You know, it’s wild to think that just 24 hours changed the entire conversation around India’s role in the global tech landscape.

Microsoft and Amazon have announced more than $50 billion in combined investments toward India’s cloud and AI infrastructure—and that’s just the headline number.

When you add Google’s $15 billion commitment, along with free AI tool access from OpenAI and Perplexity, India suddenly looks like the place where the world’s AI future is being built.

But here’s the thing—this isn’t happening by accident.

There’s a real story here about why Big Tech is betting such massive amounts on India specifically.

The $50+ Billion That Arrived in 24 Hours

Let’s start with the numbers, because they tell a powerful story:

Microsoft announced a $17.5 billion investment spread over four years, aimed at expanding hyperscale infrastructure, embedding AI into national digital platforms, and getting 20 million Indians trained in AI by 2030.

Microsoft CEO Satya Nadella called this the company’s largest investment in Asia to date. The money will fund sovereign public and private cloud services that align with India’s push for technological independence and AI-first infrastructure.

Amazon followed up with plans to invest over $35 billion by 2030—on top of the $40 billion it’s already poured into India over the years. The company says this will create around one million jobs and strengthen AI capabilities across their logistics, cloud, and small business support initiatives.

Google committed $15 billion over five years to build a state-of-the-art AI hub in Visakhapatnam, Andhra Pradesh. This isn’t just any data center—it’s designed to become one of Google’s largest AI centers globally, complete with new international subsea cable networks and renewable energy infrastructure.

Meanwhile, companies like Anthropic (behind Claude AI) opened a Bengaluru office in early 2025, marking their second Asian hub. Usage of Claude has grown fivefold in India since June 2025. Google, OpenAI, and Perplexity have all made their AI tools free for millions of Indian users, essentially betting that capturing India’s AI mindshare early will pay dividends for years to come.

When you add it all up, we’re talking about $67+ billion in commitments. That’s not spare change—that’s a serious commitment to India’s AI infrastructure.

Why Is India So Attractive? Four Big Reasons

Here’s what Big Tech executives keep saying when they talk about India, and honestly, it makes sense:

1. Abundant Physical Resources for Data Centers

Let me be real—data centers need a lot of space and power. Most Asian data center hubs have hit their limits. Singapore, one of the region’s oldest centers, struggles with land constraints. Japan, Australia, China—these markets have matured and have limited room for large-scale expansion.

India? India has the opposite problem. The country has abundant land available for massive data center developments. And here’s the kicker—India’s electricity costs are significantly lower than European data center hubs while the country is rapidly expanding its renewable energy capacity. That’s crucial because AI compute infrastructure is incredibly power-hungry.

Add in growing demand from India’s booming e-commerce sector and potential new rules around storing social media data locally, and you’ve got what industry analysts call a “sweet spot” in the global data center market. India is entering a moment where global cloud providers, AI companies, and domestic digitalization are all converging at once.

2. A Massive, High-Skilled Talent Pool

Here’s something that amazes me: India ranks first globally in AI skill penetration according to Stanford’s AI Index 2024. While India hasn’t built many foundational AI models (that’s still dominated by the US and China), it has something almost as valuable—millions of talented software engineers, data scientists, and developers who can build AI applications.

The numbers back this up. GitHub reports that India accounts for 24% of all global open-source projects. The Indian AI workforce has grown 14-fold since 2016. India’s AI industry alone is projected to reach $28.8 billion by 2025, growing at 45% annually. By 2026, demand for AI professionals in India is expected to hit one million jobs.

As S. Krishnan, secretary at India’s Ministry of Electronics and Information Technology, explained: “Having a model or computing is not enough... it requires companies making application layer and a large talent pool to deploy them.”

That’s India’s real competitive advantage. While building foundational AI models is expensive and takes years, deploying AI applications at enterprise scale needs talent. India has that in abundance.

3. A Massive Digital User Base Ready to Adopt AI

Let’s talk about scale. India has 1.4 billion people. That’s roughly 18% of the world’s population. The digital adoption rate is accelerating rapidly—from smartphones to online payments (India leads the world in digital transactions), to emerging AI tools.

Companies like Microsoft are betting that what works for Indians today will set global standards tomorrow. India’s linguistic and cultural diversity also makes it an ideal training ground for AI systems that need to work across multiple languages and contexts. Building AI that understands 22 official Indian languages and hundreds of regional variations? That creates models that are naturally more robust and adaptable globally.

Plus, 80% of Indian companies now consider AI a core strategic priority according to BCG research. In fact, 69% of Indian tech companies plan to increase their tech investments in 2026, with one-third allocating over $25 million to AI initiatives. This isn’t just external investment—India’s own companies are hungry for AI solutions.

4. Government Support and AI Sovereignty Push

India isn’t sitting passively. The government launched the IndiaAI Mission in 2024 with an allocation of ₹10,300 crore (about $1.25 billion) over five years. Here’s what that includes:

• 18,693 GPUs in high-end computing facilities (initially 10,000 available, with more coming). For context, that’s nearly 9 times the computing capacity of DeepSeek’s infrastructure and about two-thirds of ChatGPT’s operating capacity

• BharatGen—a government-funded multilingual AI model designed for public service delivery

• Sarvam-1—a large language model optimized for Indian languages with 2 billion parameters

• Chitralekha—an open-source video transcreation platform in Indic languages

• Data and AI Labs in Tier 2 and Tier 3 cities, democratizing AI access beyond major metros

Prime Minister Narendra Modi has emphasized “AI sovereignty”—building India’s own AI capabilities rather than remaining dependent on Western tech. This creates alignment between Big Tech’s interests and Indian government goals. Foreign companies investing in India’s AI infrastructure are simultaneously supporting India’s strategic tech vision.

As Union Minister Ashwini Vaishnaw put it, these massive investments signal that India is becoming “a reliable global technology partner, accelerating the shift from digital to AI public infrastructure.”

Where Are These Data Centers Actually Going?

Here’s something interesting that shows India’s thinking ahead. Traditionally, India’s data centers clustered in Mumbai and Chennai—cities close to undersea cable landing stations. That made sense logistically.

But global cloud providers are expanding capacity in Bangalore, Hyderabad, and Pune—India’s major IT hub cities. Why? Because that’s where the engineers are. Building data centers near your talent pool makes sense for operations, maintenance, and innovation. Plus, cities like Visakhapatnam (where Google’s new hub will be) have geographic advantages for international connectivity without the limitations of older coastal cities.

The Bigger Picture: India’s Data Center Market Is About to Explode

Here’s a stat that shouldn’t get lost: India’s data center market is projected to grow from $4.5 billion in 2023 to $11.6 billion by 2032—that’s a compound annual growth rate of nearly 11%. We’re still in the early innings of this growth curve.

And it’s not just foreign companies. Indian conglomerates like Reliance, Adani, and Tata are also making massive data center investments. You’ve got healthy competition between global hyperscalers (Microsoft, Amazon, Google) and Indian players, which should drive innovation and efficiency.

What About AI Talent and Job Creation?

Let’s address something important: Microsoft’s commitment to train 20 million Indians in AI by 2030 isn’t charity—it’s investment in India’s AI ecosystem. Microsoft is also integrating Azure AI into India’s Ministry of Labour and Employment platforms and the National Career Service, embedding AI deeper into government operations.

Amazon’s promise of one million jobs by 2030 comes from a company that’s already invested $40 billion. These aren’t random numbers. They reflect real infrastructure build-out that requires workers—everything from data center operations to customer support to application development.

The skills gap is real, though. While India has abundant talent, training 20 million people in AI in five years is ambitious. But it’s exactly what India needs to move from being a service provider to being a center of AI innovation.

The Indian AI Startup Angle

Here’s something that might surprise you: while India has amazing AI application startups (capturing 65% of the startup ecosystem), it only produces about 3% of global AI infrastructure and foundational model startups. That gap is by design, honestly.

As Rahul Agarwalla, founding partner at Sense AI Ventures, points out: “India has a large pool of software developers with AI knowledge. Instead of building infrastructure, they rely on global providers like Google, Microsoft, and AWS.” This is actually smart capital allocation. Indian startups focus on applications while leveraging global infrastructure.

In 2024, Indian AI startups received $1.35 billion in funding, with Bengaluru capturing over 40% of that. The startup ecosystem is ready to consume the infrastructure that Big Tech is building.

Looking Ahead: India’s AI Economic Opportunity

India’s digital economy is being driven by e-commerce, fintech, digital public services, and increasingly, AI-powered applications. These investments aren’t just about infrastructure—they’re about positioning companies to capture value from India’s digital transformation.

The convergence of three factors is creating an unprecedented opportunity: (1) massive government support and AI sovereignty initiatives, (2) abundant talent and a digital user base of 1.4 billion people, and (3) physical resources for data center expansion. This combination simply doesn’t exist anywhere else in the world right now.

When Tarun Pathak from Counterpoint Research says Microsoft’s investment gives it “first-mover advantage in GPU-rich data centers” in India, he’s highlighting something important: whoever builds the best infrastructure now will have enormous leverage in India’s AI ecosystem for the next decade.

The Bottom Line

What we’re witnessing is the early stages of India becoming a core engineering and deployment hub for global AI, not just a market or service provider. The $50+ billion committed in 24 hours isn’t an anomaly—it reflects Big Tech’s serious belief that India’s AI future is a central piece of their global AI strategy.

For Indians—whether you’re an entrepreneur, developer, or someone interested in AI—this is good news. Infrastructure investment creates jobs, drives skill development, and builds the foundation for Indian companies to build world-class AI applications.

The real race isn’t over who can build the fanciest AI model. It’s over who can build the infrastructure and ecosystem that makes AI useful at scale. And right now, India looks like the place where that’s happening.

About the author: Rupesh Bhambwani is a technology enthusiast specializing in the broad technology industry dynamics and international technology policy.

When not obsessing over nanometer-scale transistors, energy requirements of AI models, real-world impacts of the AI revolution and staring at the stars, he can be found trying to explain to his relatives why their smartphones are actually miracles of modern engineering, usually to limited success.

Leave a comment

When Nick Turley joined OpenAI in 2022 as head of ChatGPT, his job description was refreshingly vague: “help commercialize OpenAI technology.”

Three years later, he’s figured it out alright—ChatGPT now boasts 800 million weekly active users, and Turley is plotting its transformation into something that sounds suspiciously like the next Windows or macOS.

Except, you know, with more artificial intelligence and fewer printer driver updates.

In a recent interview on the outskirts of San Francisco’s Fort Mason—a former military post that seems like an appropriately strategic location to plan the conquest of the operating system market—Turley laid out his vision for ChatGPT’s future.

And it’s ambitious enough to make even Steve Jobs do a double-take from the great beyond.

Turley has a somewhat unflattering take on the current state of ChatGPT: it’s basically in its “command line era.” For those who didn’t grow up typing arcane commands into MS-DOS, that’s not a compliment. Command lines are powerful, sure, but they’re about as user-friendly as a porcupine in a balloon factory.

“It’s really powerful, but it’s lacking something very important, which is affordances,” Turley explains. The fact that ChatGPT has scaled to 800 million weekly users without proper affordances is, in his words, “kind of bonkers.” He’s not wrong. Most successful tech products don’t make you work quite this hard to use them.

The solution?

Transform ChatGPT into something more like macOS or Windows—a proper operating system where you can open applications instead of remembering obscure commands. “If you want to write, there’s an app for that. If you want to code, there’s an app for that,” Turley says, channeling his inner Apple marketing executive.

The irony of using Apple’s famous “there’s an app for that” slogan to describe what’s essentially competing with Apple’s ecosystem is not lost on us.

Learning from History

When asked where he draws inspiration for building ChatGPT, Turley gives the kind of answer that job candidates everywhere should bookmark: you need first principles thinking because there’s literally no playbook for this.

“When it comes to ChatGPT or Sora, there’s just zero precedent,” he notes.

So where does he look for guidance when you’re building something that’s never been built before?

Turley mentions browsers—specifically how they’ve quietly become the de facto operating system over the past decade. When was the last time you used a desktop app that wasn’t Excel, PowerPoint, or Spotify? Most of our digital lives now happen in Chrome, Safari, or Edge, powered by web applications that feel native but technically aren’t.

He also studied early Apple PowerBook advertisements, which had the delightful problem of explaining what a portable computer could do when nobody quite understood it yet. The ads literally listed basic functions: “It’s a calculator, it’s an alarm clock.” Sound familiar? ChatGPT faces a similar challenge—it can do so much that explaining it becomes comically difficult.

The Apps Are Coming

OpenAI isn’t exactly new to the app game. In 2023, the company launched ChatGPT plugins and the GPT Store with considerable fanfare. Those initiatives didn’t exactly set the world on fire. But at DevDay 2025, OpenAI seems to have learned from its previous stumbles.

The company introduced apps inside ChatGPT, allowing users to discover and use third-party applications directly within the chat interface. Unlike the previous attempts, these apps are integrated into ChatGPT’s core experience rather than tucked away in a separate store collecting digital dust.

The vision borrows from both WeChat and Salesforce AppExchange: centralize user attention, make third-party services interoperable, and turn ChatGPT into an operating system for reasoning.

Launch partners include heavy hitters like Expedia, DoorDash, Uber, Instacart, and even Coursera—companies that bring real utility and, more importantly, potential transaction revenue.

This is OpenAI’s pitch to developers: access to 800 million weekly users during their everyday conversations. That’s not a niche audience. That’s roughly the population of Europe, and they’re all actively engaged in ChatGPT conversations where your app could potentially be useful.

For context, that’s up from 400 million users just months ago, making it one of the fastest-growing platforms in tech history.

The Messy Business of Being a Platform

Of course, running an operating system isn’t all sunshine and app installations. It comes with thorny questions that make product managers wake up in cold sweats.

Take app discovery: If I tell ChatGPT I want to order snacks, should it suggest DoorDash or Instacart? Turley’s answer is refreshingly straightforward—show both options initially, then learn user preferences over time. “We have multiple partners in the same product categories. I think the most graceful and respectful way to handle that is to serve both apps,” he explains.

But then comes the inevitable question every platform faces: will companies be able to pay for preferential placement? In other words, can DoorDash buy its way to the top of ChatGPT’s recommendation list?

Turley won’t commit to a hard “no,” which should tell you everything you need to know. OpenAI is taking a “let’s figure it out as we go” approach, engaging with developers to understand what distribution mechanisms feel fair. “We want a great user experience,” Turley insists. “So if that would lead to apps surfacing that are irrelevant to the user, I don’t think we’d like it.”

Translation: paid placement might happen, but OpenAI will try really hard not to make it feel like you’re browsing a digital billboard. Whether they succeed is anyone’s guess.

Privacy: The Elephant in the ChatRoom

Then there’s the small matter of user data. Third-party developers naturally want access to ChatGPT user information—how else can they personalize experiences? But OpenAI can’t exactly hand over everyone’s chat history willy-nilly without causing a privacy nightmare.

OpenAI’s official line is that app developers must “gather only the minimum data required to perform the tool’s function.” What does “minimum” mean in practice? Good question. Turley admits OpenAI is still figuring this out.

The company plans to implement fine-grained access controls, taking inspiration from Apple’s approach where users can share data “just this time” or “all the time.”

Turley even floats the idea of “partitioned memory” in ChatGPT—separate compartments for different types of conversations. You might want to share your music preferences with Spotify but keep your health discussions private. Makes sense, though implementing this without making the user experience feel like filling out government forms is the challenge.

“The thing that’s uncompromisable for us is transparency,” Turley emphasizes. Users should always understand what data might be shared with third parties, even if the controls to manage that sharing are still being built out.

It’s the kind of statement that sounds great until you remember that “transparency” and “user-friendly” don’t always go hand in hand.

Beyond ChatGPT

Turley makes it clear that OpenAI’s ambitions extend far beyond a single chatbot, even one with 800 million users. “You should really think about what we’re building as a family of products and applications that are tied together by your account, personalization, and identity layer,” he explains.

OpenAI is reportedly developing a web browser (which Turley neither confirms nor denies, but describes as “really interesting”). The company is also working with Jony Ive and former Apple designers on hardware devices.

Suddenly, the vision becomes clearer: ChatGPT as the software platform, custom hardware as the access point, a browser as the web gateway, and apps as the connective tissue bringing it all together.

It’s an ecosystem play worthy of the tech giants. Apple has done it with iOS, the App Store, Safari, and its family of devices. Google has done it with Android, Chrome, and its web services. Now OpenAI wants to do something similar, except with AI at the core instead of search or mobile computing.

Perhaps the most fascinating part is how Turley reframes ChatGPT’s consumer business in relation to OpenAI’s nonprofit mission. The company’s stated goal is to develop AGI and ensure it benefits all of humanity. Some researchers worry that the consumer business—with its revenue targets and growth metrics—might overshadow that mission.

Turley’s response is a masterclass in corporate reframing.

ChatGPT isn’t just funding the mission or demonstrating the research, he argues. It is the mission. “Our mission is to ensure that AGI benefits all of humanity, and reaches people,” he explains. “If you combine that with the insight that AGI is probably not this single moment in time, but rather a gradual thing, you have to think of product as the delivery vehicle of the mission.”

He’s not wrong, exactly. Those 800 million weekly users include people teaching themselves to code at age 69, parents using ChatGPT to model social interactions for autistic children, and countless others learning languages or solving daily problems. That’s impact at scale. That’s technology benefiting humanity.

But it’s also a convenient framing that turns every ChatGPT Plus subscription and enterprise license into a noble act of mission fulfillment. Cynical? Maybe. Effective messaging? Absolutely. As Turley himself might say, it’s all about the affordances—including rhetorical ones.

The Road Ahead

OpenAI’s transformation of ChatGPT into an operating system faces significant challenges. The company has tried and failed at similar platform plays before. Privacy concerns are real and growing. Competition from Google, Microsoft, and others isn’t going away. And there’s always the risk that trying to be everything to everyone means being nothing special to anyone.

But if there’s one thing Turley’s interview makes clear, it’s that OpenAI is thinking big. Really big. Transform-how-humans-interact-with-computers big.

The company isn’t satisfied with having built the fastest-growing consumer application in history. It wants to build the platform that defines how we’ll interact with AI for the next decade.

Will it work?

That’s the billion-dollar question—or in OpenAI’s case, given recent valuations, the hundred-billion-dollar question. But one thing’s certain: the command line era of ChatGPT is coming to an end.

Whether what replaces it will be revolutionary or just another app platform will depend on execution, timing, and a bit of luck.

As for Turley, he seems energized by the challenge. When asked if OpenAI is boxing itself in, he responds with refreshing candor: “Even if we were just the ChatGPT company, there would be infinite things to build, but our ambition on what we can do for people just goes way beyond that.”

In other words, buckle up. The operating system war just got a new contender, and this one comes with conversational AI built in. Windows and macOS, meet your chatty new competitor.

Let’s just hope it handles printer drivers better than its predecessors.

About the author: Rupesh Bhambwani is a technology enthusiast specializing in the broad technology industry dynamics and international technology policy.

When not obsessing over nanometer-scale transistors, energy requirements of AI models, real-world impacts of the AI revolution and staring at the stars, he can be found trying to explain to his relatives why their smartphones are actually miracles of modern engineering, usually to limited success.

Leave a comment

If you thought your company’s quarterly planning meetings were intense, imagine mapping out an entire nation’s technological future for five years straight.

Welcome to China’s 15th Five-Year Plan (2026-2030), a document that makes most corporate strategies look like hastily scribbled grocery lists.

This isn’t just another bureaucratic blueprint gathering dust in Beijing – it’s a declaration of technological independence that could reshape global tech dynamics.

The Grand Vision

China has set itself an audacious goal: becoming a global leader in science and technology by 2035.

The 15th Five-Year Plan focuses on economic self-sufficiency and technological innovation, prioritizing strategic investments in core technologies like semiconductors, AI, quantum tech, and rare earth elements.

At a State Council press briefing in September, Science and Technology Minister Yin Hejun delivered what can only be described as a technological highlight reel. During the current 14th five-year plan period (2021-2025), China’s national R&D investment surged by 48 percent compared to 2020.

The country maintains the world’s largest pool of researchers, has climbed to 10th place in the Global Innovation Index, and has led the world in high-impact international journal publications and patent applications for five consecutive years.

Not bad for a country that Western observers often dismissed as merely capable of copying technology rather than innovating.

The Semiconductor Saga

Let’s talk chips – not the kind you eat, but the microscopic marvels that power everything from your smartphone to that smart fridge you never asked for.

China’s semiconductor journey has been rocky, to put it mildly. While the nation has achieved front-runner status in areas like space exploration, 5G, and new energy, semiconductors remain the proverbial thorn in its side, particularly as US-led export curbs have hobbled Beijing’s access to advanced chip manufacturing technology.

In other words, China needs to stop playing someone else’s game and create its own playbook.

Currently, China stands about five years behind global leaders in high-volume manufacturing of leading-edge logic semiconductor chips, and continues to trail in memory chips and semiconductor manufacturing equipment, though Chinese firms have made significant inroads in semiconductor design and production of legacy chips.

China has achieved a degree of self-sufficiency in mature process nodes (28 nanometers and above), but the high-end market remains dominated by American, Japanese, and European companies.

The real bottleneck?

Advanced process equipment, particularly those finicky lithography machines and their core components – the kind that require export licenses thicker than their instruction manuals.

The FDSOI Alternative

Here’s where things get interesting. As transistor sizes approach their physical limits (we’re talking about features smaller than a coronavirus), the global semiconductor industry is actively exploring new frontiers. China sees an opportunity in FDSOI (fully depleted silicon-on-insulator) technology.

Compared to the mainstream FinFET (fin field-effect transistor) technology, FDSOI offers some compelling advantages: simpler manufacturing processes, lower costs, high speed, low power consumption, and – here’s the kicker – less demanding requirements for advanced lithography tools.

Professor Ye also highlights emerging technologies like photonic chips and quantum computing as potential pathways for performance leaps. Beijing is introducing significant new guidelines to boost the adoption of RISC-V, the open-source chip architecture, in a determined move to lessen China’s reliance on Western semiconductor technology.

The RISC-V Revolution

If you’re not familiar with RISC-V (pronounced “risk-five,” not “risk-vee”), think of it as the Linux of chip architectures – open, free, and unburdened by licensing fees or geopolitical restrictions.

China’s RISC-V adoption is no longer grassroots — it’s now national policy, with all domestic IoT chips mandated to adopt the ISA by 2027.

In November 2018, China established the China RISC-V Alliance under the Chinese Academy of Sciences, with the aim of rapidly advancing the development of a domestic RISC-V ecosystem.

The beauty of RISC-V is that it sidesteps traditional architecture monopolies. While Arm and x86 architectures are controlled by specific companies (and subject to US export restrictions), RISC-V is genuinely open. It’s like building with LEGO blocks that anyone can manufacture, modify, and use without asking permission.

The Internal Competition

Not everything in China’s semiconductor garden is rosy. Professor Ye points out a thorny problem: “In almost every segment where localization has been achieved, there are more than five domestic companies competing.”

This low-level, redundant internal competition is partly a result of US restrictions pushing firms into horizontal expansion, and partly due to policy incentives attracting capital and new entrants who replicate mature products, thereby wasting innovative resources.

It’s the classic “too many cooks in the kitchen” problem, but with billions of dollars in funding and national pride at stake. When everyone’s rushing to fill the same gap, innovation suffers. You end up with ten companies making similar 28nm chips instead of one company pushing toward 7nm.

The solution?

Ye advocates for what he colorfully describes as “building our own house with our own blueprints” – establishing China’s own development model rather than following foreign technological roadmaps. It’s a call for strategic coordination and genuine innovation rather than parallel development and product duplication.

The AI Catalyst

If semiconductors are the foundation, artificial intelligence is the catalyst accelerating everything. Chen Xiaohong, an academician of the Chinese Academy of Engineering, emphasizes that the 15th five-year plan period will be crucial for the accelerated innovation and widespread application of AI.

China’s AI strategy involves three key pillars:

First, concentrating efforts to make breakthroughs in foundational algorithms, development frameworks, and high-end chips. This is the unsexy but essential groundwork – the mathematical foundations and specialized processors that make AI actually work.

Second, promoting the development of brain-inspired computing to reduce the energy consumption of training and running large models. (Current AI models consume enough electricity to power small cities, which is neither economically nor environmentally sustainable.)

China is also advancing explainable AI to tackle issues of model “hallucination” – those embarrassing moments when AI confidently tells you that the Eiffel Tower is in London – and enhancing cross-modal representation technologies spanning text, images, sound, and video.

Third, fostering the integration of data, algorithms, computing power, and computing networks. This involves improving data markets, promoting authorized use of public data, building open-source algorithm ecosystems, and establishing national computing power scheduling platforms. Think of it as creating a nationwide AI nervous system.

Real-World Applications

All this high-level strategy means nothing without practical applications. Priority fields include humanoid robots, intelligent connected vehicles, smart green transport, and the low-altitude economy (think delivery drones and air taxis).

Minister of Industry and Information Technology Li Lecheng announced that the 15th five-year plan period would also actively pioneer future industries such as brain-computer interfaces, the metaverse, and quantum information.

Brain-computer interfaces might sound like science fiction, but they’re rapidly becoming science fact. Quantum information could revolutionize everything from encryption to drug discovery. And the metaverse – well, at least someone’s still optimistic about it.

Growth in computing infrastructure, intelligent driving, and smart manufacturing is expected to be a key market driver during 2026-2030. The rapid development of AI creates explosive demand for specialized AI chips while simultaneously transforming chip design and manufacturing processes through automation, advanced materials screening, and improved equipment efficiency.

Forward-Looking Investments

Professor Li Xianjun of the Beijing Academy of Social Sciences argues that China must not only overcome existing bottlenecks (like developing EUV lithography) but also make forward-looking investments in emerging and disruptive technologies.

This includes advancing chiplet packaging to achieve near-advanced performance using mature nodes and exploring entirely new manufacturing techniques, such as nanoimprint lithography.

Chiplet packaging is particularly clever – instead of trying to manufacture one massive, complex chip (which requires the most advanced processes), you create multiple smaller chips and package them together. It’s like building with modular components rather than trying to carve everything from a single block of marble.

You get comparable performance without needing the most cutting-edge manufacturing equipment.

The Geopolitical Elephant in the Room

Let’s address what everyone’s thinking: this entire strategy exists because of US-China tensions.

The semiconductor industry is facing what Ye calls “the dual challenges of a technology blockade and supply chain decoupling.” US export controls have effectively locked China out of the most advanced chip-making equipment and technologies.

But here’s where it gets interesting: restrictions often spur innovation.

When you can’t buy what you need, you either invent it yourself or find alternative approaches. China is doing both simultaneously. The strategy of catch-up and substitution, while building capabilities within the global ecosystem, had created a dependency on foreign technological roadmaps. Now, facing decoupling efforts, China is forced to establish its own development model.

Whether this will succeed remains to be seen. History is full of technological catch-up stories (South Korea, Taiwan) and cautionary tales (the Soviet Union’s computer industry).

China has advantages – massive funding, a huge domestic market, a large pool of engineers – but also faces significant challenges, particularly in areas requiring decades of accumulated manufacturing expertise.

The Talent Question

Behind every technological breakthrough are actual humans with brains (at least until AI advances further). China’s success depends heavily on cultivating and attracting scientific talent. The country maintains the world’s largest pool of researchers, but quantity doesn’t automatically translate to quality or innovation.

Brain drain remains a concern, with many of China’s brightest minds still attracted to opportunities in the United States and Europe. However, improving research conditions, better funding, and growing opportunities in China’s tech sector are making domestic careers increasingly attractive. Plus, when you’re building a technological ecosystem from scratch, there’s no shortage of interesting problems to solve.

The Bottom Line

China’s 15th Five-Year Plan represents more than just another policy document. It’s a comprehensive strategy to achieve technological sovereignty in an era of increasing geopolitical competition.

The plan acknowledges both China’s achievements and its vulnerabilities, charting a course that emphasizes indigenous innovation, alternative technological pathways, and strategic patience.

Will China achieve its 2035 goal of becoming a global leader in science and technology? That remains uncertain. What is clear is that the next five years will be decisive.

China is betting heavily on emerging technologies like RISC-V, FDSOI, chiplet packaging, and AI-driven innovation to leapfrog traditional development paths.

For the rest of the world, this means increased competition, diverging technological standards, and potentially a bifurcated global tech ecosystem. For China, it means a period of intense innovation, massive investment, and the challenge of transforming from a fast follower into a genuine technology leader.

The next five years might just determine whether we’re heading toward a truly multipolar technological world or whether existing hegemonies will maintain their grip.

Either way, China’s 15th Five-Year Plan ensures that the competition will be fierce, innovative, and consequential for everyone with a stake in the future of technology.

Grab your popcorn – the show’s just getting started.

About the author: Rupesh Bhambwani is a technology enthusiast specializing in the broad technology industry dynamics and international technology policy.

When not obsessing over nanometer-scale transistors, energy requirements of AI models, real-world impacts of the AI revolution and staring at the stars, he can be found trying to explain to his relatives why their smartphones are actually miracles of modern engineering, usually to limited success.

Leave a comment

When DeepSeek dropped its updated AI model in late August, the sound you heard wasn’t applause—it was Nvidia shareholders collectively gasping.

The Chinese AI startup’s announcement that it was pivoting toward domestically produced chips sent the American chip giant’s stock into a nosedive faster than Maverick's F-14 Tomcat hitting the hard deck.

But here’s the twist: this wasn’t just another tech company making headlines. It was a shot heard around the semiconductor world, signaling that China’s quest for tech independence isn’t just talk—it’s happening, and it’s happening fast.

Let’s rewind for a moment.

The US has spent years building what it thought was an impenetrable fortress around its chip technology, particularly when it comes to AI accelerators.

Export controls, blacklists, sanctions—the works. The logic seemed sound: cut off China’s access to advanced chips, and you effectively kneecap their AI ambitions.

Except, well, someone forgot to tell the Chinese tech sector about this plan.

Enter Huawei Technologies, the company that’s become Beijing’s poster child for technological resilience. After being blacklisted by the US in 2019 over national security concerns (a polite way of saying “we don’t trust you”), Huawei did what any self-respecting tech giant would do when told they can’t play with the cool kids’ toys: they decided to build their own playground.

Last week, Huawei pulled back the curtain on its Ascend chip series roadmap—the first time they’ve done so since the blacklisting—and the tech world collectively sat up and paid attention.

This wasn’t just incremental improvement; this was Huawei laying out an ambitious blueprint for the future of Chinese AI computing.

Huawei’s Master Plan

Deputy chairman Eric Xu Zhijun didn’t mince words when describing the Ascend chips as “the foundation of Huawei’s AI computing strategy.” The company’s vision is audacious: double the compute power every year, support more data formats, enhance usability, and increase bandwidth.

You know, just casual Moore’s Law on steroids.

The Ascend 950PR, designed for pre-fill and recommendation tasks, is slated for Q1 2026. The 950DT, optimized for the heavy lifting of decoding and training, arrives in Q4 2026.

Then come the big guns: the Ascend 960 and 970 processors, scheduled for the fourth quarters of 2027 and 2028 respectively.

But here’s where it gets really interesting. Huawei didn’t just announce new chips—they unveiled what they’re calling “the world’s most powerful” supernode computing clusters.

The Atlas 950 and 960 SuperPoDs, along with their SuperCluster siblings, can aggregate up to one million Ascend neural processing units (NPUs). To put that in perspective, that’s like taking a million specialized AI brains and getting them to work together as one. If that doesn’t sound impressive, you might want to check your pulse.

Kevin Xu, founder of Interconnected Capital, noted that Huawei’s approach leverages its networking heritage. “Huawei’s announcement shows that its AI accelerators are inching closer to Nvidia’s capabilities by leaning into its networking roots,” he observed. In other words, Huawei is playing to its strengths, compensating for what it lacks in cutting-edge chip manufacturing with superior interconnection technology.

The Performance Gap

Now, before we start planning the victory parade, let’s inject some reality. Charlie Zheng, chief economist at Samoyed Cloud Technology Group, points out that while Huawei leads in publicly available products when it comes to card capacity, they’re still “one or two generations behind” Nvidia in single-chip performance.

But here’s where Chinese engineering ingenuity comes into play. Huawei founder and CEO Ren Zhengfei hinted at a workaround: chip stacking and clustering.

Think of it like this—if you can’t build the fastest single processor, why not stack multiple processors together and coordinate them brilliantly? It’s the technological equivalent of “work smarter, not harder.”

Sure, chip stacking comes with its own headaches—power consumption that would make your electricity bill weep, cabling challenges, and reliability concerns. But Zheng’s optimism isn’t unfounded. The company’s CloudMatrix 384 system, based on the Supernode384 architecture, is already making waves in industrial applications.

The crown jewel is the Atlas 950 SuperPoD, a beast of a system featuring 8,192 cards, sprawling across about 1,000 square meters (roughly two basketball courts), housed in 160 cabinets. It’s set to launch in late 2026, and if it delivers on its promises, it could be a genuine game-changer.

DeepSeek’s Cryptic Announcement

Remember that DeepSeek announcement we mentioned? Here’s where the plot thickens. In August, the Hangzhou-based startup released its V3.1 model, which was trained using a proprietary method called UE8M0 FP8 scaling (yeah too technical and I could have avoided the jargon if I could).

The technical details matter less than what DeepSeek said next: this format was “specifically tailored for next-generation home-grown chips to be released soon.”

Cue the speculation. Su Lian Jye, chief analyst at Omdia, explained that “the architecture is specifically designed to accommodate the hardware logic of Chinese chips, which enables a model to run smoothly on this hardware.”

DeepSeek has built its AI model knowing it would need to run on Chinese chips, not Nvidia’s.

While neither Huawei nor DeepSeek has officially confirmed their partnership (because keeping people guessing is apparently part of the fun), industry insiders are betting the farm on Ascend chips powering future DeepSeek models.

Liu Jie, an engineer at a Shanghai-based GPU developer, noted that Huawei already has experience running DeepSeek’s existing models on Ascend chips for inference—the process of actually using the AI model. That expertise would naturally extend to training new models.

Huawei’s Xu essentially confirmed the collaboration without quite confirming it, stating that between DeepSeek-R1’s January launch and April 30, their teams worked closely to ensure the Ascend 910B and 910C chips met customer needs.

The Broader Ecosystem

Here’s something that should concern Nvidia even more: Huawei isn’t the only game in town.

DeepSeek is likely to adopt a combination of domestic chips to optimize training efficiency and minimize power consumption. Cambricon Technologies, MetaX, and Moore Threads are all potential suppliers whose products support DeepSeek’s FP8 format.

Speaking of Cambricon, let’s talk about this Beijing-based upstart that’s become one of the priciest stocks in China’s onshore market.

Founded in 2016, the AI chipmaker has surged nearly 500% over the past year—and no, that’s not a typo. In August, Cambricon reported a jaw-dropping 4,348% year-on-year increase in first-half revenue to $404 million. That’s not growth; that’s an explosion.

The enthusiasm is fueled by genuine momentum. Major Chinese tech companies are going all-in on domestic chips. Tencent Holdings recently announced that its cloud computing unit had “fully adapted to mainstream domestic chips.”

Alibaba’s semiconductor design unit T-Head developed a new AI chip reportedly on par with Nvidia’s H20 processors. Even the state broadcaster China Central Television is hyping domestic chip achievements—when state media gets involved, you know it’s serious.

In July, Huawei joined the Model-Chips Ecosystem Innovation Alliance with other Chinese semiconductor and AI companies, creating a united front for tech self-sufficiency. Beijing has also reportedly urged tech giants to stop buying Nvidia chips specifically designed for the Chinese market—chips that were already watered down to comply with US export restrictions.

The Reality Check

Before we get carried away with the narrative of China’s unstoppable tech rise, let’s pump the brakes. As with any good underdog story, there are significant obstacles in the way.

First, there’s the software ecosystem. Having powerful hardware is one thing; having the software tools, libraries, and developer communities to fully exploit that hardware is quite another.

Nvidia’s CUDA platform didn’t become the industry standard overnight—it took years of investment and ecosystem building. Chinese chipmakers are still playing catch-up in this arena.

Then there’s the manufacturing bottleneck. China’s foundries, particularly Semiconductor Manufacturing International Corporation (SMIC), are also under US sanctions. They’re producing chips using 7-nanometer processes, which is impressive given the restrictions, but they’re doing it with far lower yields than Taiwan Semiconductor Manufacturing Company (TSMC), the world’s premier chip manufacturer.

Charlie Zheng estimates that China’s 7-nanometer production lines will supply about 600,000 to 700,000 AI chips in 2026. If Huawei’s million-card cluster becomes reality, it would consume one-sixth of that annual production capacity. That’s an enormous strain on fabrication plants already struggling with yield issues.

Kevin Xu of Interconnected Capital raises another critical question: “It is also an open question whether Huawei’s ‘supernode’ architecture can work seamlessly so that many chips can function smoothly as one computer, which is important for large-scale AI training.” Getting a million NPUs to act as a single coherent system isn’t just difficult—it’s the kind of engineering challenge that keeps smart people awake at night.

The Geopolitical Chess Game

US trade restrictions, intended to slow China’s AI development, have instead catalyzed a wave of innovation driven by necessity. Gary Ng, director and senior economist for Asia Pacific at Natixis Corporate and Investment Bank, put it succinctly: with US restrictions as the push factor and China’s self-sufficiency as the pull factor, “DeepSeek will look for alternatives for chips. As the national champion, Huawei will play a major part.”

The irony is palpable. American policymakers sought to maintain technological superiority by restricting access to advanced chips. Instead, they’ve created a scenario where China is developing an entirely parallel ecosystem that could eventually compete with—or even surpass—Western technology. It’s the law of unintended consequences writ large across the semiconductor industry.

This dynamic mirrors historical patterns. When the Soviet Union was cut off from Western technology during the Cold War, it developed its own computing systems. While those never quite matched Western capabilities, they were functional enough for Soviet purposes.

The difference today is that China has far more resources, a much larger talent pool, and a more integrated global economy (sanctions notwithstanding) than the USSR ever did.

The Innovation Pressure Cooker

There’s something to be said for necessity being the mother of invention. Without access to the latest Nvidia chips, Chinese companies have been forced to innovate in ways they might not have otherwise.

DeepSeek’s development of efficient training methods that can squeeze maximum performance from available hardware is a prime example. When you can’t throw unlimited cutting-edge chips at a problem, you have to get creative with algorithms and architecture.

This efficiency-first approach could have interesting implications. While Western AI labs have sometimes been criticized for brute-forcing problems with massive computational resources, Chinese developers are being forced to find more elegant solutions. If they succeed, they might develop AI systems that are not only competitive but also more cost-effective and energy-efficient than their Western counterparts.

Looking Ahead

That statement encapsulates both the predicament and the determination of China’s tech sector. They’re behind, they know they’re behind, but they’re not giving up. Instead, they’re finding alternative paths—chip stacking, superior networking, efficient algorithms, and sheer scale.

The next few years will be fascinating to watch. Will Huawei’s Ascend chips close the gap with Nvidia? Can Chinese foundries improve their yields enough to supply the massive demand? Will DeepSeek and other AI companies prove that you don’t need the absolute latest chips to build world-class AI systems?

The answers to these questions will shape not just the tech industry but the broader balance of power in the 21st century. Because make no mistake—whoever leads in AI will have significant economic, military, and geopolitical advantages.

The Bottom Line

The partnership between DeepSeek and Huawei represents more than just two companies working together. It symbolizes China’s broader strategy of achieving technological self-reliance in the face of Western restrictions.

Whether this strategy ultimately succeeds remains to be seen, but one thing is certain: dismissing Chinese tech innovation as mere imitation or state-sponsored copying would be a serious mistake.

As Nvidia shareholders nervously watch their stock prices and policymakers in Washington reassess their strategies, Chinese engineers are busy stacking chips, optimizing algorithms, and building supercomputers. They’re behind, but they’re catching up. And they’re doing it without asking permission.

In the end, this story isn’t really about chips or AI models. It’s about what happens when you tell a nation of 1.4 billion people—including some of the world’s smartest engineers—that they can’t have access to cutting-edge technology.

Sometimes, they just build their own. And sometimes, that homegrown technology turns out to be pretty damn good.

About the author: Rupesh Bhambwani is a technology enthusiast specializing in the broad technology industry dynamics and international technology policy.

When not obsessing over nanometer-scale transistors, energy requirements of AI models, real-world impacts of the AI revolution and staring at the stars, he can be found trying to explain to his relatives why their smartphones are actually miracles of modern engineering, usually to limited success.

Leave a comment

The relationship between Silicon Valley and the Pentagon has fundamentally transformed. After years of carefully avoiding public connections to the military-industrial complex, technology giants like Google, Microsoft, Meta, and Amazon have embraced lucrative defense contracts worth billions of dollars.

This shift represents more than a business pivot—it signals a profound realignment of corporate priorities that raises urgent questions about ethics, transparency, and the future of warfare itself.

The Trump Effect and the Trillion-Dollar Push

Donald Trump’s return to the White House has accelerated Big Tech’s military ambitions. The administration’s plan to invest a trillion dollars by 2026 to “modernize” the armed forces—with artificial intelligence at its core—has opened floodgates for Silicon Valley.

The message is clear: AI-powered defense is the future, and tech companies are racing to claim their stake.

The scale of engagement is staggering.

OpenAI, Google, Anthropic, and Elon Musk’s xAI have each secured contracts worth up to $200 million to develop advanced AI capabilities for the Department of Defense.

These aren’t peripheral projects but central initiatives aimed at integrating cutting-edge technology into national security infrastructure. OpenAI’s recent $200 million Pentagon contract specifically targets both “warfighting and enterprise domains,” marking a dramatic shift from the company’s earlier prohibition on military applications.

Meta’s transformation has been particularly striking. The company announced plans to invest $14.3 billion in Scale AI, whose founder Alexandr Wang now works in Meta’s general AI research division.

Scale AI has been selected by the Pentagon to test and evaluate large language models for military use. In May, Meta partnered with defense startup Anduril to develop virtual and mixed reality headsets specifically designed for soldiers—technology that could fundamentally change how combat operations are conducted.

When Executives Become Officers

Perhaps nothing symbolizes the blurring lines between Silicon Valley and the military more than the creation of Detachment 201, the Army’s Executive Innovation Corps.

In June, four technology executives were commissioned as reserve lieutenant colonels: Adam Bosworth, Meta’s CTO and Mark Zuckerberg confidant; Kevin Weil, OpenAI’s Product Manager; Shyam Sankar, Palantir’s CTO; and Bob McGrew, a former Executive at both Palantir and OpenAI.

This reverse hiring process—where tech executives don military fatigues—represents an unprecedented fusion of corporate and military interests.

These individuals now serve dual roles: advancing their companies’ commercial interests while shaping military technology strategy from within the armed forces. The arrangement raises obvious questions about conflicts of interest and where loyalties ultimately lie when corporate profits and national security objectives intersect.

The Policy Pivot

The corporate embrace of military work required fundamental changes to company policies that had previously restricted such engagements. In February, Google quietly removed prohibitions on developing weapons and mass surveillance tools from its code of conduct.

This reversal came years after the 2018 Project Maven controversy, when thousands of Google employees protested the company’s partnership with the Pentagon to analyze drone surveillance footage.

OpenAI followed suit in January 2024, eliminating its ban on using ChatGPT technology for “military and war” tasks. The revised policy now permits “national security use cases that align with our mission”—a carefully worded change that opens vast possibilities while maintaining rhetorical commitment to ethical principles.

Microsoft has been equally forthcoming about its military engagements. The company acknowledged that since Israel’s invasion of Gaza began, it has sold advanced AI technology and cloud computing services to the Israeli army. These aren’t abstract software licenses but tools actively used in ongoing military operations, raising profound ethical questions about corporate complicity in warfare.

The Economy of Genocide

The most controversial dimension of Big Tech’s militarization involves Israel. Project Nimbus, a $1.2 billion joint contract between Google, Amazon, and the Israeli government, has become a flashpoint for employee activism and ethical concerns. The agreement provides cloud infrastructure, artificial intelligence services, and data centers to Israeli government agencies and military forces.

UN Special Rapporteur Francesca Albanese has described the intersection of technology companies, cloud services, and arms manufacturers in occupied Palestinian territories as an “economy of genocide.” Her report implicates Microsoft, HP, IBM, Google, and Amazon in surveillance technologies deployed there.

IBM has contributed to biometric databases tracking Palestinians, while Microsoft and Palantir provide cloud services supporting Israeli government and military systems.

These aren’t minor technical contributions. The technologies enable sophisticated surveillance networks, facial recognition systems, and data analysis capabilities that fundamentally shape how occupation and military operations function.

The companies involved maintain that they’re simply providing neutral technology platforms, but this distinction becomes increasingly meaningless when those platforms are integral to military and security operations.

Employee Resistance and Corporate Retaliation

The rapid militarization of Big Tech has sparked significant internal resistance. In April 2024, Google employees staged coordinated sit-ins at company headquarters in New York, Sunnyvale, San Francisco, and Seattle to protest Project Nimbus. The demonstrations, organized by the group No Tech for Apartheid, demanded that Google and Amazon terminate their Israeli government contracts.

Google’s response was swift and severe: 28 employees were fired for participating in the protests. The company accused them of disrupting operations and violating workplace policies, but activists saw it as retaliation for speaking out against military contracts.

Additional firings followed, with reports indicating approximately 50 employees have been terminated for protesting Project Nimbus.

Microsoft has similarly cracked down on dissent. In April 2025, the company fired two employees who publicly criticized AI supplies to Israel. In February, five workers were ejected from a meeting with CEO Satya Nadella at Microsoft’s Redmond headquarters for protesting military contracts.

These corporate actions send a clear message: employee activism against military work will not be tolerated. The firings have created a chilling effect within tech companies, where workers increasingly fear speaking out about ethical concerns regarding their employers’ military engagements.

Justifying the Shift

Tech executives have framed their military pivot as a matter of democratic values and national security imperatives. Google DeepMind founder Demis Hassabis told media that “Western democratic values are under threat” and that his company has “a duty to be able to help with what we are uniquely qualified and positioned to do.” He cited defending against AI-powered cyberattacks and biological threats as examples of necessary military applications.

This framing presents the militarization as defensive and value-driven rather than profit-motivated. However, critics argue this narrative conveniently aligns corporate interests with patriotic duty.

As Heidy Khlaaf, Chief AI scientist at the AI Now Institute, explains, positioning themselves as protagonists in a “quasi-civilizational crusade” protects tech companies from regulatory scrutiny. Any call for accountability can be branded as “a detriment to national interests.”

The national security argument also allows companies to position themselves as “not only too big, but also too strategically important to fail.” This status grants them extraordinary power to shape policy, attract investment, and resist regulation—all while pursuing profit through military contracts.

The Cybersecurity Paradox

Ironically, Big Tech’s military involvement may create the very vulnerabilities it claims to address. Building national security systems on widely available foundation models like Meta’s Llama or OpenAI’s GPT-4 introduces significant cybersecurity risks. These systems can be manipulated through data poisoning, creating vectors for hostile actors to compromise military infrastructure.

AI companies have largely circumvented the rigorous military standards that traditional defense systems must meet. The “move fast and break things” ethos of Silicon Valley stands in direct tension with the reliability and security requirements of national defense. Yet these companies have successfully promoted what Khlaaf calls “an unfounded narrative of an AI arms race” to justify bypassing established protocols.

Data, Consent, and Surveillance

A disturbing dimension of Big Tech’s military work involves the use of personal data. As Khlaaf emphasizes, the personally identifiable information used to train AI models enables military applications like ISTAR (intelligence, surveillance, target acquisition, and reconnaissance) capabilities. This data allows systems to monitor and target specific populations.

Crucially, this happens without user consent. Whether or not individuals use AI tools, their data makes military and surveillance applications possible. The billions of users generating content, images, and information that feed these models have become unwitting participants in building military infrastructure.

Historical Continuity and Corporate Power

The militarization of technology is not new. As ethics expert Lorena Jaume-Palasí notes, “Our Western concept of modern technology has its genesis in the military or security sphere.” The internet originated as a secure communications system for armed forces. GPS guided missiles and submarines were used before directing civilian navigation.

The pattern of military origins and civilian adaptation is well-established.

What has changed is scale and concentration.

Eight of the world’s ten largest companies by market capitalization are American technology firms: Nvidia, Microsoft, Apple, Amazon, Alphabet, Meta, Broadcom, and Tesla. Only Saudi Aramco and Berkshire Hathaway represent other sectors.

This unprecedented consolidation means that a handful of corporations wield extraordinary influence over military technology development.

These companies have successfully framed AI advancement as a matter of national security, even when development is primarily profit-driven and environmentally harmful.

Trump himself has repeatedly emphasized that American companies must beat China in the AI arms race—rhetoric that serves corporate interests by justifying increased investment and reduced regulation.

The Changing Defense Context

The shift toward military work reflects broader geopolitical changes. As Raquel Jorge of Spain’s Elcano Royal Institute explains, recent conflicts have directly affected U.S. national security interests in ways previous wars did not. The war in Ukraine beginning in 2022 and the Gaza conflict since 2023 have created a defense context that makes technology companies “feel more comfortable talking about” military engagements.

Trump’s demands for increased defense spending and greater NATO ally contributions further normalize military work as patriotic duty. In this environment, companies that were previously “very careful with their narrative” around defense contracts now openly embrace military partnerships.

Power, Accountability, and Democracy

Big Tech’s militarization raises fundamental questions about power, accountability, and democratic governance. These companies operate with minimal oversight, make decisions affecting global security behind closed doors, and silence internal dissent through aggressive retaliation. Their products and services touch billions of lives, yet democratic institutions have struggled to effectively regulate or even monitor their activities.

The fusion of commercial technology corporations with military objectives creates a new power structure that transcends traditional categories. These aren’t purely private companies operating in a free market, nor are they government agencies subject to democratic accountability.

Instead, they occupy a hybrid space where profit motives, strategic interests, and technological capabilities intersect with minimal transparency or public input.

The technologies being developed—AI-powered surveillance systems, autonomous weapons capabilities, advanced analytics for targeting—will shape warfare and security operations for decades. Yet the public has little insight into how these systems work, what safeguards exist, or what ethical considerations guide their deployment.

Employee activism within tech companies represents one form of resistance, though corporate retaliation has made such advocacy increasingly risky. Regulatory efforts face the challenge of governing entities that have positioned themselves as too important to constrain.

International frameworks remain underdeveloped for addressing the role of multinational corporations in warfare.

As Silicon Valley continues its transformation into a central pillar of the military-industrial complex, the fundamental question remains: who benefits from this arrangement, and at what cost?

The trillion-dollar investments, the lucrative contracts, and the concentration of power suggest that Big Tech’s war business serves corporate interests extraordinarily well.

Whether it serves broader human interests, democratic values, or long-term security remains very much in question.

About the author: Rupesh Bhambwani is a technology enthusiast specializing in the broad technology industry dynamics and international technology policy.

When not obsessing over nanometer-scale transistors, energy requirements of AI models, real-world impacts of the AI revolution and staring at the stars, he can be found trying to explain to his relatives why their smartphones are actually miracles of modern engineering, usually to limited success.

Leave a comment

Remember when your biggest privacy concern was whether your mom would read your diary? Well, congratulations—you’ve upgraded.

Now AI is reading everything you do online, and unlike mom, AI doesn’t even pretend to feel guilty about it.

Artificial intelligence is the digital equivalent of that friend who “just wants a bite” of your fries and ends up eating half your plate. Except in this case, the fries are your personal data, and AI isn’t asking—it’s already deep in the basket.

For AI to be effective, particularly AI agents (those conversational robots managing your emails, schedules, and shopping), it needs to know you intimately. What you search for, what you buy, what embarrassing health conditions you Google at 2 AM—all of it becomes fuel for the machine.

And while sometimes you’ve technically given permission for this data feast, let’s be honest: clicking “I agree” on a 47-page terms of service document while trying to access a service hardly counts as informed consent.

According to Hervé Lambert, Global Consumer Operations Manager at Panda Security, this AI access to data creates risks of “commercial manipulation, exclusion, or even extortion.” So basically, it’s not paranoia if they’re actually watching you.

Your Browser Assistant Is a Snitch

Recent research from University College London presented at the USENIX security symposium, revealed just how chatty AI browser assistants can be. Spoiler alert: they’re basically the gossips of the digital world.

During tests using an invented user profile, researchers discovered that AI web browser assistants were sharing search information with their servers, along with banking data, health information, and IP addresses. They demonstrated the ability to guess user attributes like age, sex, salary, and interests, using this information to personalize responses across different browsing sessions.

Only one assistant, Perplexity, didn’t show evidence of profiling or personalization—making it the digital equivalent of the one friend who can actually keep a secret.

“Although many people are aware that search engines and social media platforms compile information about them for targeted advertising, AI web browser assistants operate with unprecedented access to user online behavior,” explains Anna Maria Mandalari, the study’s primary author. These assistants peer into areas of your online life that should remain private, often without transparency, consent, or regard for privacy legislation—and sometimes even violating their own companies’ terms of service.

Because apparently, rules are more like suggestions when there’s money to be made.

The “Black Box” Problem

AI models, especially deep learning models with opaque algorithms, are often “black boxes,” making it difficult to explain how they make decisions.

The sheer volume of information in play makes AI arguably pose a greater data privacy risk than earlier technological advancements. We’re not talking about a filing cabinet of information anymore—we’re talking about the digital equivalent of the Library of Congress, except it’s all about you, and anyone with the right access can read it.

Tech Giants Playing Privacy Limbo

Google’s “Improvement” Initiative

Google recently updated its privacy terms under the friendly guise of wanting to “improve our services.” (Pro tip: When a tech company says they’re “improving services,” grab your personal data and run.) The company now admits to using interactions with Gemini, its AI application, and has launched a “temporary chat” feature for users who want to opt out.

But here’s the catch—you have to be proactive. You need to deactivate the “keep activity” function and manually manage and delete Gemini app activity. If you don’t, congratulations! Your entire digital life becomes training data.

From September 2nd onwards, files, videos, screens, photos, and audio recordings submitted to Gemini have become fair game for “improving Google services for everyone.”

Google assures users that it disconnects chats from accounts before sending them to “service providers”—which is corporate speak for “we’re sharing your data, but we promise we took your name tag off first.” That should make everyone feel better, right?

WhatsApp and Meta: Trust Us (But Also Don’t Tell Us Anything Important)

WhatsApp insists that “your personal messages with friends and family are off limits.”

The AI is only trained through direct interaction with their artificial intelligence application. But they do offer this chilling warning: “What you send to Meta may be used to provide you with accurate responses or to improve Meta’s AI models, so don’t send messages to Meta with information you don’t want it to know.”

The use of anthropomorphic interfaces, such as human-sounding voices used in assistants like Alexa and Siri, creates a unique problem: people are inclined to interact with technology as if it were human, developing trusting relationships and consequently sharing increasingly personal information.

We’re basically being catfished by our devices.

The WeTransfer Panic

File-sharing service WeTransfer recently sparked panic by modifying its terms of service in a way that seemed to grant unlimited access to user data for AI training. The backlash was swift and severe.

The company had to issue clarifications faster than a politician caught in a scandal, essentially saying: “NO, we’re not stealing your stuff to train our AI. YES, your content is yours. PLEASE don’t leave us.”

It was a valuable lesson in what happens when privacy policy updates aren’t carefully worded—or perhaps, what happens when companies get too honest about their data ambitions.

The Risks Are Real (and Terrifying)

The potential dangers of AI data collection extend far beyond annoying targeted ads. Lambert cites risks including:

• Commercial manipulation: Your data being used to influence your purchasing decisions in ways you don’t realize

• Geopolitical manipulation: Because why stop at selling you products when you can influence your political views?

• Exclusion: Being locked out of opportunities based on AI profiling

• Extortion: In a world of frequent data breaches, there’s no telling who ends up with your information

• Identity theft: The classic crime, now with an AI twist

It’s a digital Wild West out there, except the outlaws have machine learning and you’re riding a horse made of your own personal data.

“But We Need It for Progress!”

Tech companies are acutely aware of the privacy problem—not just because of ethical and legal concerns, but because data limitations are slowing their AI development.

Their response? Find ways around those pesky limitations.

Meta founder Mark Zuckerberg has directed work toward “self-improving AI”—systems capable of increasing performance through advances in equipment, programming (including self-programming), and having the AI train itself. I

Startup Sakana AI created the Darwin Gödel Machine, where an AI agent adapts its own code to improve task performance. Progress? Absolutely. Slightly terrifying? Also yes.

Chris Painter, policy director at METR (a non-profit AI research organization), warns that if AI accelerates its own development capabilities, it could also be weaponized for hacking, weapons design, and human manipulation. So that’s comforting.

What Can Actually Be Done?

Eusebio Nieva, technical director of Check Point Software for Spain and Portugal, advocates for comprehensive regulations including:

• Transparency and explicit consent requirements

• Security regulations for devices

• Prohibition and restrictions on high-risk AI providers (as seen in European regulation)

In 2025, the initial EU enforcement wave bans unacceptable-risk AI uses, including manipulative techniques, social scoring, and real-time biometric surveillance. At least someone’s trying to pump the brakes on this data train.

Lambert calls for shared responsibility between users and companies, rejecting the notion that protective regulation slows progress: “Protecting our users does not mean that we are going to slow down; it means that, from the outset of a project, we include privacy and digital footprint protection, thereby becoming more effective and efficient in protecting our most important assets, which are our users.”

It’s almost refreshing to hear someone suggest that users are assets to be protected rather than resources to be mined.

The Bottom Line

We’re living in an era where AI knows you better than your best friend, your therapist, and possibly yourself. It knows what you search for at 3 AM, what you’re thinking of buying before you buy it, and probably what you’re going to eat for dinner next Tuesday.

The uncomfortable truth is that we’ve traded privacy for convenience, often without realizing the full cost of that transaction.

Every “I agree” click, every voice assistant query, every file upload is another piece of data feeding the AI beast.

So what can you do?

• Actually read privacy policies (or at least the key sections)

• Use temporary chat features when available

• Disable data collection options where possible

• Be mindful of what you share with AI assistants

• Support regulations that protect consumer privacy

• Remember that if something is free, you’re probably the product

The AI revolution isn’t coming—it’s here, and it’s hungry. The question isn’t whether AI will collect your data, but whether we’ll demand that it does so responsibly, transparently, and with genuine consent.

Because right now, AI is devouring your digital life like an all-you-can-eat buffet. And unless we set some ground rules, it’s never going to stop going back for seconds.

Or thirds.

Or your entire online existence.

The next time you’re about to click “I agree” on a terms of service update, maybe pause for just a moment. That clicking sound? It might be the sound of your privacy being uploaded to the cloud.

About the author: Rupesh Bhambwani is a technology enthusiast specializing in the broad technology industry dynamics and international technology policy.

When not obsessing over nanometer-scale transistors, energy requirements of AI models, real-world impacts of the AI revolution and staring at the stars, he can be found trying to explain to his relatives why their smartphones are actually miracles of modern engineering, usually to limited success.

Leave a comment

Who would have thought that a desert city once famous for black gold would become the Silicon Valley of the Middle East? Dubai's transformation from oil hub to tech paradise is nothing short of remarkable, and Chinese tech giants are playing starring roles in this modern-day makeover.

From 50% Oil to Less Than 1%

Let's start with some mind-blowing numbers. Oil production, which once accounted for 50% of Dubai's gross domestic product, contributes less than 1% today.

That's not a typo – we're talking about a complete economic 180-degree turn that would make even the most seasoned business strategist dizzy.

Most of Dubai's GDP (over 95%) is non-oil-based. This transformation didn't happen overnight or by accident. Dubai's leadership saw the writing on the wall decades ago and decided to diversify before the oil taps ran dry.

Smart move, considering how many oil-dependent economies are still struggling to find their footing in a post-petroleum world.

The emirate's tech-friendly policies have created what can only be described as a digital oasis in the desert. At the heart of this transformation sits Dubai Internet City (DIC), a tech park that's become the Middle Eastern equivalent of Silicon Valley – complete with palm trees and stunning coastline views and California sunshine.

Dubai Internet City: Where East Meets West

Located near the upscale Palm Jumeirah (yes, that palm tree-shaped archipelago that makes every aerial shot of Dubai look like something from a sci-fi movie), Dubai Internet City has become the go-to destination for global tech giants.

Customer numbers grew more than 17.5% in the first nine months of 2024 to reach over 4,000 businesses compared to the same period in 2023, while the community of tech talent at Dubai Internet City expanded to exceed 31,000 professionals.

The park hosts everyone from Microsoft and Google to Amazon.com, alongside their Chinese competitors who are increasingly making themselves at home in the desert heat. The number of companies operating in Dubai Internet City increased by 10 per cent annually in the first half of 2024, boosting the number of professionals in the district. That's the kind of growth that would make any tech hub jealous.

TikTok's Desert Adventure

ByteDance, the parent company of TikTok, has secured multiple floors in a DIC tower with what can only be described as Instagram-worthy coastline views. When TikTok CEO Chew Shou Zi visited in 2024, it wasn't just a courtesy call – it was a statement of intent about the company's commitment to expanding in Dubai.

Angela Ji, a Chinese national who's been working at TikTok's Dubai office for two years, offers some insider perspective: "Dubai makes me feel like I'm in an era of economic growth, where tech companies still spend big to expand and hire. That makes it so different from China and elsewhere." Coming from someone who previously worked at ByteDance's Beijing headquarters, that's quite the endorsement.

The company offers generous relocation packages for Chinese employees willing to trade Beijing's smog for Dubai's desert sunshine. With hundreds of staff housed in their Dubai office, TikTok is clearly betting big on the Middle East becoming a crucial growth market.

Huawei: Building the Infrastructure of Tomorrow

Huawei established its regional office in DIC back in 2016 and has since made Dubai its Middle East hub for telecommunications, cloud services, and consumer electronics. The company operates six consumer-facing stores across the UAE, with four located in Dubai – because apparently, even in the age of e-commerce, sometimes you need to touch the phone before you buy it.

A sales associate at one of Dubai's Huawei stores reports that Chinese smartphones have gained significant popularity among both local and Chinese consumers. The Mate X6 launched in December and the innovative trifold Mate XT released in February have been particular hits.

The new flagship Pura 80 series, unveiled just last week, had several models selling out on the first day – proof that Dubai's tech-savvy consumers have expensive tastes and quick reflexes.

In the cloud services arena, Huawei is going head-to-head with tech titans like Microsoft, Google, Amazon, and fellow Chinese giant Alibaba. At last year's Gitex Global – one of the region's largest tech fairs – Huawei introduced its hybrid Cloud Stack 8.5 specifically for the Middle East and Central Asia.

With 6,000 partners across the Middle East and Africa, they're clearly not just window shopping in the region.

Desert Innovation

Alibaba's cloud unit opened a data centre in Dubai back in 2016, showing early confidence in the region's potential. Last year, they launched a Middle East & Africa Training Centre in DIC, offering ecosystem partners and customers training, workshops, and certifications.

Because in the tech world, knowledge truly is power – and Alibaba wants to make sure everyone speaks their cloud language.

Meanwhile, AliExpress is bringing Chinese goods into the region faster than you can say "free shipping." During the midyear 618 shopping festival, something fascinating happened: 65% of Chinese smart-glass maker Rokid's global sales originated in the Middle East.

That's right – more than half the world's demand for Chinese smart glasses was coming from this region. Talk about finding your niche market!

Chinese Cars on Arabian Roads

Perhaps nowhere is the Chinese influence more visible than along Sheikh Zayed Road, one of Dubai's main arteries. Here, showrooms for Chinese automotive brands BYD, Nio, and Zeekr stand proudly alongside luxury stalwarts like Rolls-Royce and Bentley.

It's like an automotive United Nations meeting, but with better air conditioning.

According to consulting firm Roland Berger, the Middle East market holds "extremely high growth potential" for Chinese carmakers, with its current development stage reminiscent of China's automotive boom in the early 2000s. Translation: if you thought China's car industry transformation was impressive, buckle up for the Middle Eastern sequel.

The autonomous driving revolution is also gaining momentum. Guangzhou-based Pony.ai has partnered with Dubai's Roads and Transport Authority to conduct pilot tests of robotaxis later this year, aiming to launch fully driverless commercial services. Soon, you might be able to hail a taxi that doesn't need coffee breaks or ask for directions.

Chinese Talent in the Desert

The influx of Chinese tech companies has brought with it a growing community of Chinese professionals. Wang Yu, a property agent who has lived in Dubai for a decade, notes that Chinese nationals make up an increasing proportion of expatriates in the city.

Many can afford high rents in sought-after areas like Dubai Marina and Palm Jumeirah – areas where the views cost almost as much as the actual apartments.

This isn't just about companies setting up shop; it's about creating entire ecosystems of talent, innovation, and investment. When hundreds of Chinese tech workers relocate to Dubai, they bring their expertise, their networks, and their late-night coding habits (though the desert heat might discourage too much midnight oil burning).

Smart City, Smarter Strategy

Dubai's transformation is being actively promoted by the Digital Dubai Authority and other government initiatives focused on creating a "smart city" agenda. The government has established free zones, including Dubai Internet City and Dubai Silicon Oasis industrial estate, which allow fully foreign-owned companies. It's like offering a VIP pass to the global economy – no local partner required.

These tech-friendly policies are proving to be a lifeline for Chinese companies facing domestic challenges and looking to accelerate their overseas expansion. For many Chinese tech giants, Dubai offers something increasingly rare: a place where they can grow without constantly looking over their shoulders at regulatory uncertainty.

Economic Impact by the Digits

Dubai Internet City has added AED 100 billion to Dubai's GDP over the past 15 years. That's the kind of economic contribution that turns heads in government circles worldwide. When you consider that Dubai's economy expanded by 3.1% in real terms in the first nine months of 2024, with total real GDP reaching AED 339bn, it's clear that the tech sector is pulling serious weight.

The success is attracting attention from all corners of the tech world. With over 1,600 multinational companies, Fortune 500 firms, and high-growth startups, DIC plays a crucial role in Dubai's transition from oil dependency to tech innovation.

What we're witnessing in Dubai isn't just another tech hub success story – it's a masterclass in economic transformation. The emirate has managed to reinvent itself from an oil-dependent economy to a diversified powerhouse where Chinese and Western tech giants compete and collaborate in equal measure.

The presence of companies like ByteDance, Huawei, and Alibaba alongside traditional tech titans creates a unique ecosystem where East meets West in the most practical way possible: through innovation, investment, and a shared belief that the future is digital.

As Angela Ji from TikTok's Dubai office puts it, Dubai feels like "an era of economic growth, where tech companies still spend big to expand and hire." In a world where many tech hubs are tightening their belts, Dubai's willingness to roll out the red carpet for international tech investment is paying dividends.

The transformation from 50% oil dependency to less than 1% is more than just impressive statistics – it's proof that with the right vision, policies, and timing, any economy can reinvent itself. And if you can do it while maintaining some of the world's best shopping malls and tallest buildings, well, that's just showing off.

About the author: Rupesh Bhambwani is a technology enthusiast specializing in the broad technology industry dynamics and international technology policy.

When not obsessing over nanometer-scale transistors, energy requirements of AI models, real-world impacts of the AI revolution and staring at the stars, he can be found trying to explain to his relatives why their smartphones are actually miracles of modern engineering, usually to limited success.

Leave a comment

Remember when the biggest debate in entertainment was whether to watch movies in theaters or at home?

Well, hold onto your popcorn because Netflix just casually announced they've been slipping AI-generated scenes into your binge-watching sessions. And honestly, you probably didn't even notice.

The Building That Changed Everything

During one of Netflix's earnings call – you know, one of those corporate events that usually puts people to sleep faster than counting sheep – co-CEO Ted Sarandos dropped a bombshell.

Netflix had used generative AI for the first time for a scene in an Argentinean show called "El Eternauta", marking what he called "the very first GenAI final footage to appear on screen" in Netflix content.

The scene in question? A dramatic building collapse in the postapocalyptic series that adapts an Argentine comic of the same name, following survivors of a toxic snowfall that blankets Buenos Aires.

Now, I've seen my fair share of buildings dramatically collapsing on screen (thanks, Michael Bay), but this one was special – it was created entirely using AI.

And here's where it gets interesting: The generative AI used in The Eternauts helped its production team to complete a sequence 10 times faster than if they had used traditional special effects tools. Ten times faster! That's like going from a horse-drawn carriage to a Tesla in terms of production speed.

The Economics of Digital Destruction

This is actually a game-changer for smaller productions. The six-part series was made entirely by an Argentine cast and crew with a budget lower than is typical for big-budget Hollywood productions. Yet they managed to pull off visual effects that would have required a Hollywood-sized budget just a few years ago.

Sarandos made it clear that Netflix isn't just dabbling in AI for the fun of it. During the call, he emphasized that AI represents "an incredible opportunity to help creators make films and series better, not just cheaper."

The co-CEO explained that AI-powered creator tools are already helping with "pre-visualization and shot-planning work, and certainly visual effects." He pointed out something particularly interesting: "It used to be that only big-budget projects would have access to advanced visual effects like de-aging."

Remember how impressed we all were when they made Robert De Niro look young in "The Irishman"? That technology is now becoming accessible to productions with much smaller budgets.

But Netflix isn't stopping at making buildings fall down prettier. Co-CEO Greg Peters revealed that the company is using generative AI in other areas too, including personalization and search and ads. They're even planning to roll out interactive ads in the second half of this year.

Earlier in 2025, Netflix introduced AI-powered search, though let's be honest – we're all still going to spend 20 minutes scrolling through our "Continue Watching" list before settling on The Office for the millionth time.

The breadth of AI implementation shows Netflix isn't just experimenting; they're systematically integrating AI across their entire operation. It's like they're building an AI-powered entertainment ecosystem, one algorithm at a time.

The Numbers Don't Lie

Speaking of impressive numbers, Netflix's Q2 results were pretty spectacular. The company announced a robust profit of $3.1 billion in Q2 2025, a dramatic increase from the previous year's $2.1 billion. They also reported revenue of $11.08 billion, up 16% from a year earlier.

But here's the stat that really caught my attention: users watched over 95 billion hours of content in the first half of 2025. To put that in perspective, if you watched Netflix content continuously for 95 billion hours, you'd be watching for over 10.8 million years. I mean, we all knew people were binge-watching, but this is next level.

What's particularly interesting is that non-English titles accounted for one-third of all views. Shows like "El Eternauta" are proving that great storytelling transcends language barriers – and apparently, AI-enhanced visual effects don't hurt either.

Now, let's address what everyone's thinking but nobody wants to say out loud: what does this mean for the armies of VFX artists who currently make their living creating digital mayhem?

The broader anxiety around AI's impact on jobs was evident in 2024, when filmmaker Tyler Perry paused a planned $800 million expansion of his Atlanta studio, citing uncertainty about how AI-generated video content would affect employment across the industry.

It's a legitimate concern, but Sarandos framed it differently during the call. He emphasized that these are "AI-powered creator tools" and stressed that "this is real people doing real work with better tools." It's a bit like when calculators didn't eliminate mathematicians – they just changed how math gets done.

The Future of Storytelling

What's fascinating about Netflix's approach is how matter-of-fact they are about it. "The creators were thrilled with the result," said Sarandos, and honestly, that's what matters most. At the end of the day, if the audience can't tell the difference and the creators are happy with the result, does it really matter whether that spectacular explosion was crafted by a team of VFX artists working for months or an AI system working for days?

The good news for fans of the show — and for Argentina – is Season 2 is confirmed, which means we'll likely see even more AI-enhanced sequences in the future. Netflix's 'The Eternaut' has already generated $34M for Argentina, proving that AI-enhanced productions can still have very real economic benefits for the countries where they're made.

The Takeaway

Netflix's quiet introduction of AI into its content production feels like one of those watershed moments that we'll look back on and say, "That's when everything changed." It's not the loud, dramatic transformation that Hollywood loves to portray in movies about technology.

Instead, it's the kind of practical, results-driven implementation that actually changes industries.

The streaming giant has essentially proven that AI can enhance creativity without replacing it, reduce costs without sacrificing quality, and speed up production without cutting corners. Whether you're a filmmaker worried about your job security or a viewer who just wants to be entertained, Netflix's AI experiment suggests that the future of entertainment might be brighter – and more efficient – than we thought.

And the best part? You probably won't even notice the difference. You'll just keep binge-watching, blissfully unaware that some of those spectacular scenes were crafted by algorithms instead of artists. In a world where we're constantly debating whether AI will take over everything, Netflix has quietly shown us that maybe, just maybe, AI can simply make things better (or maybe not)

Now, if they could just use AI to figure out why my "Continue Watching" list never actually helps me pick something to watch, that would be most helpful.

About the author: Rupesh Bhambwani is a technology enthusiast specializing in the broad technology industry dynamics and international technology policy.

When not obsessing over nanometer-scale transistors, energy requirements of AI models, real-world impacts of the AI revolution and staring at the stars, he can be found trying to explain to his relatives why their smartphones are actually miracles of modern engineering, usually to limited success.

Leave a comment

When Alibaba CEO Eddie Wu took the stage at the company’s annual Apsara conference in Hangzhou last September, nobody expected fireworks.

The man’s known for reading from prepared statements like a high school student giving a book report on a novel they didn’t quite finish.

But this time? Wu came out swinging with four words beamed on a giant screen that would send Alibaba’s stock soaring: “Roadmap to Artificial Superintelligence.”

Wu’s 23-minute keynote explicitly invoked both AGI (artificial general intelligence) and ASI (artificial superintelligence)—making Alibaba the first established Chinese tech giant to publicly embrace these terms.

While Silicon Valley has been tossing around these concepts for years, this was different. This was China’s largest cloud company declaring that AGI isn’t the finish line—it’s just the starting point.

The ASI Declaration

Wu laid out an ambitious vision: “Achieving AGI—a system with human-level general cognitive abilities—now appears to be a certainty. However, AGI is not the endpoint of AI development; it is a brand-new starting point.

AI will not stop at AGI; it will progress toward ASI, surpassing human intelligence and capable of self-iterative evolution.”

Think about that for a moment.

While American tech companies obsess over reaching human-level AI, China’s biggest players are already planning for what comes after.

Wu described ASI as a system that could produce a generation of “super scientists” and “full-stack super engineers” capable of conquering medical challenges, inventing new materials, solving sustainable energy and climate issues, and even enabling interstellar travel. No pressure, right?

Wu outlined a three-phase roadmap toward ASI:

First, “emergent intelligence,” where AI develops reasoning by learning from humanity’s collective knowledge.

Second, “autonomous action,” where AI acquires tool use and programming abilities to assist humans—roughly where the industry stands today.

Finally, “self-iteration,” where AI connects to raw data from the physical world, learns autonomously, and ultimately surpasses humans.

And she’s right to be surprised. This isn’t just marketing fluff—it represents a fundamental shift in how China approaches the AI race.

The global AI arena has morphed into something far more complex than a simple sprint to AGI.

The big five AI hyperscalers are pouring billions into data centers and AI infrastructure, with spending expected to exceed $400 billion this year—roughly the GDP of Romania, for those keeping track at home. That’s a lot of server farms.

But here’s where it gets interesting: the competition isn’t just about throwing money at the problem anymore.

It’s about building complete ecosystems—what industry insiders call “full-stack capabilities” spanning hardware, software, and applications. Think of it as the difference between owning a fancy sports car engine versus owning the entire car factory, the supply chain, and the dealership network.

Capital spending by the top five US hyperscalers—Amazon Web Services (AWS), Microsoft, Alphabet, Meta and Oracle—rose 66% to US$211bn in 2024, with Microsoft alone committing over $105 billion in future data center deals.

Meanwhile, Jensen Huang, Nvidia’s leather-jacket-wearing CEO, predicted that ChatGPT maker OpenAI could become a “multi-trillion-dollar hyperscale company.”

That’s quite the endorsement, though Huang might be slightly biased given Nvidia’s $100 billion investment in OpenAI’s success.

Wu believes that to unlock this superintelligent future, large AI models will replace existing operating systems as the link between users, software and computational power, running on cloud computing networks like Alibaba Cloud.

It’s a vision where ASI and business strategy are completely intertwined—a full-stack approach that makes Silicon Valley’s piecemeal investments look almost quaint.

AGI vs. ASI: Why China Chose the Harder Path

Here’s where China’s strategy diverges sharply from America’s.

While Silicon Valley executives lose sleep over achieving AGI before China does (spoiler alert: it’s not happening next Tuesday), Beijing has taken a refreshingly different—and arguably more ambitious—approach.

Zhang Peng, CEO of Chinese AI startup Zhipu AI, predicts artificial superintelligence may arrive by 2030 but will likely only surpass humans in specific areas initially, rather than achieving full superintelligence across all domains.

That’s a more measured take than some of the breathless predictions coming from Silicon Valley, where OpenAI CEO Sam Altman suggested ASI could arrive before 2030, and SoftBank’s Masayoshi Son pegged it for 2035.

But China’s focus isn’t just theoretical. DeepSeek’s achievements highlight China’s broader goal to lead AI by 2030 under the “New Generation Artificial Intelligence Development Plan,” which combines state funding and private innovation in a centralized approach unlike the US decentralized model.

Instead of getting “AGI-pilled”—as Princeton researcher Kyle Chan puts it—Chinese policymakers see AI as something more pragmatic: a turbocharger for existing industries.

China’s strategy, dubbed “AI Plus,” focuses on integrating artificial intelligence with the country’s massive manufacturing and industrial sectors.

And the results? They’re pretty spectacular.

China showcased more than 150 humanoid robots at the World Artificial Intelligence Conference (WAIC) 2025 in Shanghai, demonstrating that the country isn’t just playing catch-up—it’s running a different race entirely.

China now leads the world in industrial robot installations, with a record 2.027 million active robots, according to the International Federation of Robotics.

In March, Beijing made a bold move by designating “embodied intelligence”—AI integrated into physical machines—as a key future industry.

In January 2025, China launched an $8.2 billion National AI Industry Investment Fund, while its broader $138 billion National Venture Capital Guidance Fund targets AI-related fields, including robotics and embodied intelligence.

While America debates how many billions to spend on the next language model, China’s robots are already assembling cars, sorting packages, and conducting power inspections.

This is where China’s ASI vision gets tangible—it’s not just about creating smarter software, but about building physical systems that can self-iterate and improve.

China’s embodied intelligence market is expected to reach 5.295 billion yuan by 2025, and the momentum is accelerating.

Companies like Shanghai-based AgiBot and Hangzhou’s Unitree Robotics are landing orders from state-owned firms, while nearly half of China’s AI fundraising this year has flowed toward embodied intelligence startups.

In the first half of 2025, total financing in the humanoid robotics sector reached a record high, exceeding 10 billion RMB ($1.4 billion). And it’s not just startups getting in on the action—tech giants like Huawei, ByteDance, and even food delivery platform Meituan are developing their own robotics solutions.

Because apparently, when you’ve mastered getting dumplings delivered in 30 minutes, building a humanoid robot is the logical next step.

They predict a future where “intelligence embedded in the physical world” culminates in generalist robots performing tasks across manufacturing, services, and defense.

And the country betting on that future? Not the US.

If you needed more evidence that China’s AI strategy is working—and that the path to ASI doesn’t require unlimited budgets—look no further than DeepSeek.

This Hangzhou-based company, founded in 2023 and backed by Chinese hedge fund High Flyer, pulled off something remarkable in January 2025: it released an open-source reasoning model that matched OpenAI’s performance at a fraction of the cost.

By 27 January, DeepSeek surpassed ChatGPT as the most downloaded freeware app on the iOS App Store in the United States, triggering an 18% drop in Nvidia’s share price.

The company claims it trained its V3 model for just $6 million—compared to over $100 million for OpenAI’s GPT-4—and using approximately one-tenth the computing power of Meta’s comparable model.

The impact has been staggering. A US government report noted that downloads of DeepSeek models on the developer platform Hugging Face surged nearly 1,000% since January.

Over 700 community models based on DeepSeek have appeared on Hugging Face, amassing more than 5 million downloads by early 2025.

Here’s the kicker: scarcity pushed Chinese engineers to reinvent training methods and hardware stacks, while US labs—flush with top-tier GPUs—lacked similar constraints.

It turns out that having unlimited resources might not be the advantage everyone assumed. Sometimes constraints breed innovation—who knew?

Even OpenAI swiftly responded with its first open model in six years, but according to open-source AI expert Nathan Lambert, the gap may already be too wide to bridge.

“Qwen alone is roughly matching the entire American open model ecosystem today,” Lambert said at a recent industry conference.

That’s Alibaba’s open-source model, by the way—you know, the one Eddie Wu was talking about when he outlined his ASI roadmap.

America’s Spending Spree vs. China’s Ecosystem

Now, let’s talk numbers, because they’re both impressive and slightly terrifying.

US hyperscalers are expected to spend over $300 billion in 2025, up 25% from 2024 and nearly double 2023, with Amazon and Microsoft leading the charge at $96.4 billion and $89.9 billion respectively.

OpenAI alone has accumulated computing deals worth at least $1 trillion this year, including the ambitious $500 billion Stargate Project.

Alibaba, by comparison, announced a RMB 380 billion (US$53 billion) three-year investment plan in AI infrastructure.

Wu added that by 2032, compared with 2022, Alibaba Cloud’s global data center energy consumption will grow tenfold to meet the arrival of the ASI era. That’s a serious commitment, but still dwarfed by annual US spending.

But here’s the catch: China’s hyperscalers don’t need to match these astronomical figures to compete effectively.

The American “big three”—AWS, Microsoft Azure, and Google Cloud—command about 63% of the $900 billion global cloud computing market.

But in China, Alibaba Cloud leads with 36% of the domestic market, and that’s enough to build a thriving ecosystem aimed at ASI development.

The Chinese cloud market is vast enough to support multiple players—Alibaba, Huawei, Baidu, and ByteDance—each finding different opportunities while collectively pushing toward superintelligence.

The Open-Source Advantage

Here’s where China’s ASI strategy gets really clever.

While American companies guard their AI models like Colonel Sanders protects his secret recipe, Chinese tech giants have embraced open-source with the enthusiasm of a kid in a candy store.

Alibaba’s Qwen models, DeepSeek’s various offerings, and Huawei’s ecosystem are all freely available for developers to download, modify, and build upon.

By 2024, models like DeepSeek-V3 were outperforming Meta’s Llama 3.1 and Anthropic’s Claude 3.5 Sonnet on common language and reasoning benchmarks, thanks to architectural innovations such as Mixture-of-Experts and Multi-Head Latent Attention. Yeah, I know - nerd talk!

And they’re doing it in the open, where anyone can examine, improve, and build upon the work.

This open approach is creating what Zhao calls “a self-sufficient AI stack, free from Nvidia’s influence.”

When DeepSeek introduced its new programming language TileLang, Chinese chipmakers Hygon and Cambricon quickly announced “day zero” chip support, while Huawei developed core operators for the new model.

The depth of China’s open-source ecosystem spans from Big Tech giants such as Huawei Technologies and ByteDance to unexpected developers like food delivery giant Meituan and Alibaba’s fintech affiliate Ant Group, which open-sourced a 1 trillion-parameter model this year.

This collaborative approach suggests China believes the path to ASI runs through collective advancement rather than proprietary breakthroughs.

Different Races, Different Finish Lines

So, who’s ahead in this race?

Well, that depends on how you define “winning”—and whether you’re racing toward AGI or ASI.

If the goal is achieving AGI—that mythical point where AI matches human intelligence across all tasks—then the jury’s still out. Some US executives believe they’re in a sprint to reach AGI before China, leading to what Princeton’s Chan describes as

But if the goal is ASI—systems that can self-iterate and continuously improve beyond human capabilities—China’s full-stack, applications-focused approach might have the edge.

Morgan Stanley estimates that China’s total addressable market for robots will double to US$108 billion by 2028 from US$47 billion in 2024, with hyper-growth concentrated in collaborative robots (46% CAGR), mobile robots (35% CAGR), and service robots (25% CAGR).

And in terms of actual application and people benefiting from AI—the kind of widespread deployment that could lead to self-iterating systems—China has made significant strides.

In the first half of 2025, industrial robot production in China reached 370,000 units while service robot output hit 8.82 million units, representing year-on-year growth of 35.6 percent and 25.5 percent, respectively.

Wu’s vision goes even further: “With ASI surpassing human intelligence, humans and AI will collaborate in entirely new ways. In the future, every household, factory, and company may have numerous agents and robots working 24/7, potentially requiring hundreds of GPU chips per person.”

That’s not AGI—that’s a superintelligent infrastructure woven into the fabric of society.

The Cost Advantage - China

Here’s a fun fact that keeps Silicon Valley executives up at night: It costs 2.2 times more in the US than in China to manufacture a robotic arm of similar specifications.

China’s Unitree Go2 quadruped robot starts at 1/54 the price of Boston Dynamics’ similar Spot robot. That’s not a typo. It is one fifty-fourth. There is no way, US can beat that.

This cost advantage stems from China’s well-established manufacturing system, strong industrial support capabilities, and supply chain integration.

While American companies can certainly match the technical sophistication, they can’t match the economics—at least not yet. And in a race toward ASI where you need millions of physical robots collecting real-world data for self-iteration, economics matter. A lot.

Fueling the ASI Future

Wu introduced a provocative concept: “Tokens are the electricity of the AI world,” suggesting that in the ASI era, AI agents will consume tokens like households consume electricity.

This isn’t just metaphorical—it’s a complete reconceptualization of computing infrastructure.

Wu predicted that large AI models will replace existing operating systems as the link between users, software and computational power, with natural language becoming the new programming language, agents becoming the new software, and context becoming the new memory. In this vision, reaching ASI requires not just smarter models, but an entirely new computing paradigm.

China is developing a National Integrated Computing Network that will integrate private and public cloud computing resources into a single nationwide platform optimized for AI workloads, with the “Eastern Data, Western Computing” initiative building eight national computing hubs.

By June 2024, China had 246 EFLOP/s of total compute capacity and aims to reach 300 EFLOP/s by 2025. That’s infrastructure built explicitly for the ASI era.

The Future: Five or Six Hyperscalers

At the Apsara conference, Eddie Wu made a bold prediction: there would only be “five or six hyperscalers globally” in the future, implying that Alibaba would be one of them.

Given current trends and China’s explicit focus on ASI rather than just AGI, that prediction doesn’t seem far-fetched.

What’s clear is that the AI race has evolved beyond a simple competition of computing power or model performance. It’s become a contest of ecosystems, adoption strategies, and full-stack capabilities—with China explicitly aiming for the ultimate prize of artificial superintelligence.

The most valuable tech companies have emerged as early winners in the AI era, with the five biggest companies accounting for more than 70% of the total market value of the top 20, up from 65% last year. But the Chinese challengers are proving that massive spending alone doesn’t guarantee victory in the race to ASI.

The lesson from China’s AI strategy is surprisingly simple: it takes more than chips—and more than AGI—to win the AI race.

While American companies pour hundreds of billions into developing ever-more-powerful language models and competing for scarce GPU supplies, China is quietly building a comprehensive AI ecosystem that spans from hardware to applications, from research labs to factory floors—all explicitly aimed at achieving artificial superintelligence that can self-iterate and evolve.

The US still holds advantages in foundational AI research, semiconductor design, and venture capital. But China’s focus on practical applications, open-source collaboration, manufacturing integration, and explicit ASI development represents a fundamentally different—and potentially more sustainable—approach to AI development.

As one researcher noted, the concept of superintelligence has long guided prominent American AI companies, with OpenAI releasing an article on safe superintelligent AI development in May 2023 stating: “Now is a good time to start thinking about the governance of superintelligence—future AI systems dramatically more capable than even AGI.”

The difference? China isn’t just thinking about it—they’re building it.

Alibaba’s commitment to full-stack development—from cloud infrastructure to open-source models to physical robotics—embodies China’s broader strategy: winning not by outspending Silicon Valley in the race to AGI, but by building a more complete, integrated, and practical path to artificial superintelligence.

In the end, the AI race might not be won by whoever reaches AGI first or spends the most.

It might be won by whoever builds the most comprehensive infrastructure for ASI—systems that can self-iterate, continuously improve, and seamlessly integrate with the physical world.

And on that metric, China’s lesson is clear: forget the hype about AGI, build the full stack for ASI, and let the robots (and the self-improving AI systems) do the heavy lifting.

Now, if you’ll excuse me, I need to go check if my food delivery app has upgraded to humanoid robot couriers powered by self-iterating superintelligent systems yet.

Given current trends, it’s probably only a matter of time.

And who knows? Maybe those robots will be smarter than us by 2030. At least they’ll probably be better at remembering to include the extra soy sauce.

About the author: Rupesh Bhambwani is a technology enthusiast specializing in the broad technology industry dynamics and international technology policy.

When not obsessing over nanometer-scale transistors, energy requirements of AI models, real-world impacts of the AI revolution and staring at the stars, he can be found trying to explain to his relatives why their smartphones are actually miracles of modern engineering, usually to limited success.

Leave a comment

Remember when your biggest tech investment worry was whether to buy the latest iPhone? Those were simpler times.

Today, we're witnessing what might be the most spectacular financial circus since someone convinced people that tulips were worth their weight in gold.

Welcome to the AI investment boom—a $3 trillion bet that's making the dot-com bubble look like a modest poker game between friends.

The Numbers That Make Your Calculator Weep

Let's start with some numbers that would make even a lottery winner blush. American tech giants are throwing around cash like confetti at a billionaire's birthday party.

Meta, Amazon, Alphabet and Microsoft intend to invest as much as $320 billion this year into AI advancement technologies. Meanwhile, Microsoft plans to spend $80 billion in fiscal 2025 on the construction of data centers that can handle artificial intelligence workloads—that's roughly the GDP of Ecuador, spent on what essentially amounts to very expensive digital Lego blocks.

The scale is genuinely mind-boggling.

With over $1 trillion in investment at stake, energizers are finding ways to deliver reliable power while driving ROI. They are making substantial investments in emerging power-generation technologies—including nuclear, geothermal, carbon capture and storage, and long-duration energy storage.

We've reached the point where tech companies are considering nuclear reactors to power their data centers. Who would have thought that 10 years back?

The AI data center market is experiencing unprecedented growth. In 2024, the global data center market focused on AI was valued at $13.62 billion and is projected to grow at a remarkable 28.3% compound annual growth rate through 2030. That's the kind of growth rate that would make a bamboo forest jealous.

More Servers, Less Pickaxes

This isn't just about tech companies anymore—it's become a full-ecosystem play. Property developers are racing to build data centers, electricity companies are scrambling to provide power, and even Larry Ellison briefly became the world's richest man when Oracle announced its AI ambitions.

It's like the California Gold Rush, except instead of forty-niners with pickaxes, we have CEOs with PowerPoint presentations and very expensive GPUs

The investment frenzy has reached such heights that 50% of venture dollars were spent on AI start-ups during the first half of 2025, according to data from CB Insights.

And in those six months "AI funding exceeded spending for all of last year".

That's not growth—that's a feeding frenzy with venture capitalists as the sharks and AI startups as the... well, also sharks, but smaller ones.

But Wait, There's a Plot Twist

Here's where things get interesting (and slightly terrifying).

Goldman Sachs Research estimates that about $720 billion of grid spending through 2030 may be needed just to keep these AI data centers running. We're not just building digital castles in the sky—we're rewiring the entire electrical grid to power them.

The investment boom has been so intense that it's actually driving significant portions of economic growth. The AI boom has allegedly contributed 40% of America's GDP growth over the past year—a figure so staggering it sounds like someone added an extra zero by mistake.

To Pop or Not to Pop?

Now for the trillion-dollar question (literally): Is this a sustainable revolution or the mother of all bubbles waiting to burst?

Investors' long-running enthusiasm for artificial intelligence showed signs of faltering few weeks back as tech stocks tumbled, and even Sam Altman, OpenAI's CEO, has warned about an AI bubble forming.

The warning signs are certainly there. Many smaller, experimental AI companies, particularly those without substantial revenue or established business models, are at high risk.

Companies with "gargantuan" P/E ratios like Palantir (NYSE: PLTR) and CrowdStrike (NASDAQ: CRWD) are considered highly overvalued. When companies are valued more on promise than performance, that's usually when the financial gravity starts kicking in.

Some analysts predict the AI bubble can burst in 2025 is due to the expected resolution of GPU scarcity, that's sent Nvidia's stock into the stratosphere. It's ironic—the very success in solving supply chain issues could deflate the bubble that supply scarcity helped create.

What Happens When the Music Stops?

If this bubble does burst, the fallout could be swift and dramatic. Unlike the railway boom that left Britain with useful infrastructure, much of today's AI spending is on servers and specialized chips that become obsolete faster than your smartphone.

More than half the capital expenditure splurge has been on equipment that'll be worthless in a few years.

The good news?

The financial system might actually survive this particular bubble. Much of the investment is coming from tech giants with deep pockets rather than over-leveraged banks.

If the AI bubble bursts, tech-heavy indices may experience sharp corrections, particularly those overexposed to speculative AI stocks. While unlikely to cause global instability, the shakeout could temporarily spook innovation-heavy markets.

However, there's still cause for concern. The concentration of AI investments means that if you remove those companies from the stock index, so-called American exceptionalism disappears.

Outside the top 10 companies, the indices have been stagnating since the start of 2025. The economy has become surprisingly dependent on a handful of tech companies making very large bets on artificial intelligence.

The Reality Check

Here's the thing about bubbles—they're only obvious in retrospect.

The internet boom of the late '90s was also called a bubble, and while many companies did crash and burn (remember this line from the movie, Top Gun?), the underlying technology revolutionized everything.

The question isn't whether AI will be transformative (it almost certainly will be), but whether the current valuations and investment levels are sustainable.

Early adopters are already showing a preference for smaller, more efficient AI models, which could suggest that the massive computing infrastructure currently being built might be overkill.

The Bottom Line

We're living through a fascinating economic experiment: What happens when the world's richest companies make trillion-dollar bets on technology that might or might not live up to its hype?

The scale of investment is unprecedented, the potential returns are astronomical, and the risks are... well, also astronomical.

Whether this ends with artificial general intelligence ushering in an era of 20% annual economic growth or with a bunch of very expensive data centers being repurposed as the world's most advanced cryptocurrency mining facilities remains to be seen.

One thing's for certain: Future business school case studies are going to be absolutely fascinating.

And if you're wondering whether you should invest in this AI boom, remember the old Wall Street saying: "When taxi drivers start giving stock tips, it's time to sell."

Today's equivalent might be: "When your smart toaster starts recommending AI stocks, maybe it's time to reassess."

The $3 trillion question isn't just about money—it's about whether we're witnessing the birth of a new technological age or the most expensive cautionary tale in human history. Place your bets, but maybe keep the receipt.

About the author: Rupesh Bhambwani is a technology enthusiast specializing in the broad technology industry dynamics and international technology policy.

When not obsessing over nanometer-scale transistors, energy requirements of AI models, real-world impacts of the AI revolution and staring at the stars, he can be found trying to explain to his relatives why their smartphones are actually miracles of modern engineering, usually to limited success.

Leave a comment

Nvidia, worth more than the entire economies of most countries at over $4.5 trillion, is lending money to OpenAI so OpenAI can buy Nvidia’s chips.

Meanwhile, OpenAI is borrowing hundreds of billions to build data centers, and also buying stakes in Nvidia’s competitors. If this sounds like a game of financial hot potato played with amounts that would make a Bond villain blush, you’re not alone in feeling queasy.

Lisa Shalett, Chief Investment Officer at Morgan Stanley Wealth Management, has become the party’s designated driver—the one person willing to say “maybe we’ve had enough” while everyone else is ordering another round of AI-fueled euphoria. And her warnings should make even the most bullish tech investor pause mid-sip.

Since ChatGPT burst onto the scene in late 2022 and the market bottomed out that October, the S&P 500 has rocketed up 90%. Champagne corks popping, right?

Not so fast. Here’s the uncomfortable truth: most of those gains came from a shockingly small group of companies.

The so-called “Magnificent Seven” (think Nvidia, Microsoft, and their friends) plus about 34 AI data-center ecosystem companies have been responsible for roughly 75% of the market’s returns, 80% of earnings growth, and a jaw-dropping 90% of capital spending growth.

Meanwhile, the other 493 companies in the S&P 500? They’re up a modest 25%.

Tech companies are projected to spend roughly $400 billion this year alone on data-center infrastructure. To put that in perspective, the entire Apollo space program—you know, the one that put humans on the moon—cost about $300 billion in today’s dollars spread across more than a decade.

We’re essentially launching a new moon mission every 10 months, except instead of exploring space, we’re teaching computers to write poetry and generate images of cats wearing top hats.

Shalett points out that while “AI adoption” might be in its first inning, the “capex infrastructure buildout has been going full-out since 2022.” In other words, we’re not in the early stages of anything—we’re somewhere around the seventh inning, and the bullpen is looking nervous.

This spending spree has become so significant that AI capital expenditure contributed an estimated 100 basis points—one full percentage point—to second-quarter GDP growth.

When the Snake Eats Its Own Tail

Now we get to the really spicy part. Nvidia, currently the most valuable company in the history of the world with a market cap exceeding $4.5 trillion, sits at the center of an increasingly tangled web of investments that’s starting to look less like strategic partnerships and more like a game of financial Twister.

In September alone, Nvidia invested $100 billion in OpenAI—just days after pledging $5 billion to Intel.

But here’s where it gets weird: OpenAI is partially owned by Microsoft. Nvidia has also invested in OpenAI. Oracle has a data-center deal with OpenAI. AMD has purchasing agreements with OpenAI. And then OpenAI struck a deal allowing it to buy up to 10% of AMD.

Let that sink in: Nvidia’s main competitor could end up partially owned by OpenAI, which is partially owned by Nvidia.

It’s financial “inception”, except instead of dreams within dreams, we have investments within investments. As Shalett puts it, “It is totally circular and increases systemic risk.”

Shalett sees these interconnected deals as a red flag waving so vigorously it might achieve liftoff. “The guy at the epicenter, Nvidia, is basically starting to do what all ultimate bad actors do in the final inning, which is extending financing, they’re buying their investors,” she warns.

Oracle’s $300 Billion Bet

Speaking of eyebrow-raising deals, OpenAI struck a $300 billion data-center agreement with Oracle—roughly a week before the Nvidia deal. Analysts at KeyBanc Capital Markets estimate that Oracle will need to borrow $100 billion of that amount, at $25 billion per year for four years.

This is where Shalett gets genuinely nervous. “Every morning the opening screen on my Bloomberg is what’s going on with CDS spreads on Oracle debt,” she revealed.

For the uninitiated, credit default swaps (CDSs) are financial instruments that became infamous during the 2008 financial crisis—they’re essentially insurance policies investors buy in case a company can’t pay its debts.

“If people start getting worried about Oracle’s ability to pay,” Shalett warned, “that’s gonna be an early indication to us that people are getting nervous.”

The Cisco Moment

Shalett invoked the specter of the “Cisco moment”—that glorious period in 2000 when Cisco briefly became the world’s most valuable company before experiencing an 80% stock plunge as the dot-com bubble burst.

When asked how close we are to such a moment, Shalett estimated probably not in the next nine months, but very possibly within 24 months. Bank of America’s Vivek Arya echoed similar timing: “I’m extremely comfortable with what will happen in the next 12 months... But can there be periods of digestion in between? Yeah.”

Even John Chambers, Cisco’s former CEO and one of the faces of the original dot-com bubble, told the Associated Press in early October that he sees “irrational exuberance on a really large scale” reminiscent of the internet age. When the guy who lived through the last bubble says it looks like a bubble, you might want to pay attention.

The Bulls Make Their Case

To be fair, not everyone is running for the hills. Bank of America’s Vivek Arya argues that Nvidia’s actual ecosystem investments amount to less than $8 billion over the past 12 months—not the astronomical figures headlines suggest. He maintains that Nvidia and OpenAI will be among the four or five ecosystems that ultimately succeed, making current investments strategic rather than speculative.

“It’s not like Nvidia is just handing a $100 billion check to OpenAI [and saying] you know, go have fun,” Arya explained. The deals are structured in tranches over multiple years, with performance milestones. It’s vendor financing, not a blank check—though critics might argue that distinction matters less if the whole house of cards collapses.

Goldman Sachs CEO David Solomon, speaking at a tech conference in Turin, acknowledged the massive spending but characterized it pragmatically: “There will be a lot of capital that was deployed that didn’t deliver returns. We just don’t know how that will play out.”

Jeff Bezos called it “kind of an industrial bubble” but argued the infrastructure would pay off for years to come. Even OpenAI CEO Sam Altman, whose company sits at the center of much of this spending, acknowledged reality: “Between the 10 years we’ve already been operating and the many decades ahead of us, there will be booms and busts. People will overinvest and lose money, and underinvest and lose a lot of revenue.”

The Free Cash Flow Problem

Here’s a sobering statistic that should concern anyone paying attention: hyperscaler companies (the big tech firms building all this infrastructure) have already seen free-cash-flow growth turn negative. That’s finance-speak for “they’re spending money faster than they’re making it back.”

Strategas, an independent research firm, estimates that hyperscaler free cash flow will shrink by more than 16% over the next 12 months. This puts enormous pressure on the sky-high valuations these companies command and forces investors to demand more discipline in capital deployment.

Or as normal people might put it: at some point, someone’s going to ask “so when does this actually make money?”

The Groupthink Problem

Shalett raised an even more unsettling concern beyond just market mechanics. She pointed to media consolidation, noting that Oracle founder Larry Ellison plays major roles in a Trump-friendly consortium attempting to buy TikTok, while his son David Ellison now owns Paramount (which includes CBS News).

“That is not something that most of us have experienced in our lifetimes,” she warned. When the same small group of people controls both massive chunks of AI infrastructure and major media outlets, “groupthink” can filter into market functioning. “You stop factoring in risk premiums into markets, there is no bear case to anything.”

It’s a reminder that financial bubbles aren’t just about numbers on a spreadsheet—they’re about collective psychology, about entire industries convincing themselves that this time really is different.

So What Happens Next?

The uncomfortable truth is that nobody really knows. Shalett waved away macro theories about the labor market or Federal Reserve policy, insisting “we think that’s missing the forest for the trees because the forest is entirely rooted in this one story” about AI infrastructure.

Morgan Stanley’s bull-case mid-2026 price target for the S&P 500 is 7,200—significantly higher than current levels. But even that optimistic outlook acknowledges that risk premiums, credit spreads, and market volatility don’t seem to fully account for the vulnerabilities lurking beneath the surface.

The question Shalett poses is whether we’ll see a mild 1991-92-style recession or something worse when the music stops. “At some point,” she notes, “we’re not gonna be building any data centers for a while.”

The Bottom Line

We’re witnessing one of the largest capital expenditure booms in modern history, funded by an increasingly circular web of investments and debt, concentrated in a remarkably small number of companies that have come to dominate market returns. The parallels to previous bubbles are hard to ignore, even as bulls argue the infrastructure being built will prove valuable regardless of short-term volatility.

Perhaps the most honest assessment came from Altman’s acknowledgment of inevitable “booms and busts.” The only questions are: how big will the boom get, how hard will the bust hit, and are you personally holding the bag when it happens?

In the meantime, if you’ll excuse me, I need to go check those Oracle CDS spreads. You know, just in case

About the author: Rupesh Bhambwani is a technology enthusiast specializing in the broad technology industry dynamics and international technology policy.

When not obsessing over nanometer-scale transistors, energy requirements of AI models, real-world impacts of the AI revolution and staring at the stars, he can be found trying to explain to his relatives why their smartphones are actually miracles of modern engineering, usually to limited success.

Leave a comment

You're Jensen Huang, CEO of Nvidia, sitting on top of a $4.5 trillion empire (yes, trillion with a T), and you're wondering how to keep the money train rolling.

Then it hits you—why not convince every country on Earth that they need their own AI system?

Boom! "Sovereign AI" is born, and suddenly governments are lining up with their wallets open.

Late in 2023, Huang started pitching what might be the most successful sales campaign since someone convinced us we needed bottled water.

His pitch was simple: every country should have its own artificial intelligence system, trained on local data, aligned with national values, and built using domestic infrastructure.

He called these systems "AI factories"—because nothing says "cutting-edge technology" like good old-fashioned manufacturing metaphors that make politicians feel warm and fuzzy inside.

And boy, did it work.

Politicians are eating this up like free donuts at a town hall meeting. Earlier this year, the European Commission unveiled plans for a €20 billion ($23 billion) fund to build up to five "AI gigafactories."

That's billion with a B, folks—enough money to buy Twitter twice over (though admittedly, that bar keeps getting lower).

In just the past few months, the governments of France, Germany, Saudi Arabia, South Korea, and the United Arab Emirates have all jumped on this bandwagon faster than passengers boarding the last train home.

According to Nvidia—who definitely has no bias in this matter whatsoever (yeah, right!)—at least 20 countries are now pursuing sovereign AI initiatives.

Follow the Money

Let's be honest about why Jensen Huang is so enthusiastic about this idea.

Jefferies, an investment bank, estimates that sovereign initiatives could generate around $200 billion in cumulative revenue for Nvidia "over the coming years." Nvidia itself is even more optimistic, suggesting spending could reach $1 trillion over what they diplomatically call "an equally fuzzy period."

That's the kind of fuzzy math that would make a politician blush.

But there's more to Huang's strategy than pure greed (though there's definitely plenty of that).

His biggest customers—Amazon, Alphabet, Meta, and Microsoft—are all developing their own chips, which could eventually reduce their dependence on Nvidia.

Sovereign AI represents a brilliant hedge against this threat.

And the numbers are staggering. Saudi Arabia expects to purchase "several hundred thousand" of Nvidia's top-end processors over the next five years. Meanwhile, the UAE—apparently not to be outdone in the "who can spend more taxpayer money on AI" competition—intends to import half a million processors annually. That's a lot of silicon heading to the desert.

Here's where things get interesting. Despite all the enthusiasm and eye-watering price tags, the concept of sovereign AI remains about as clear as mud in a thunderstorm. Sure, it might help countries develop national models, protect sensitive information, and widen access to AI technology.

But will it actually reduce dependence on America?

Spoiler alert: not really.

The approaches vary wildly. France has decided to go the partnership route, with domestic model-maker Mistral teaming up with Bpifrance (a government bank), MGX (an Emirati state-owned investor), and Nvidia to build what's being marketed as Europe's largest AI data campus.

In the Gulf, governments have taken a more direct approach. In May, Saudi Arabia launched Humain, a firm specifically tasked with building AI infrastructure in the kingdom.

The UAE has handed this responsibility to G42, an AI firm part-owned by Mubadala, a sovereign wealth fund. When you have enough oil money to buy small countries, why not cut out the middleman?

The Justifications Are... Creative

Governments are justifying these massive expenditures with a variety of reasons that range from reasonable to "did you really just say that with a straight face?"

Some big spenders want to catch up to America. The European Commission hopes to propel Europe to the "forefront of AI development"—because nothing says technological leadership like being five years behind and spending twice as much to get there.

Others, like India, have more legitimate concerns about AI models trained on foreign data not incorporating local languages and values. This is actually a fair point—imagine trying to get AI assistance in Hindi only to have it respond in English with a distinctly Silicon Valley bias about the best places to get artisanal coffee.

Control over domestic data is another major concern, particularly in healthcare. Officials worry about patients' information flowing into foreign models, which is about as appealing as finding out your medical records are being used as training data for a chatbot that keeps trying to sell you supplements.

Finally, there's the access argument. Nadia Carlsten, CEO of the DCAI (which runs Gefion, Denmark's national AI supercomputer), points out that smaller companies and research institutes are "always at the back of the line" when it comes to accessing AI resources.

Gefion, launched in 2024, is being used for applications like drug discovery and weather forecasting—you know, the boring but actually important stuff that doesn't involve generating memes creating fake videos

The Inconvenient Truth About Self-Sufficiency

Here's where the sovereign AI dream hits the wall of reality harder than a bird flying into a freshly cleaned window. These projects won't offer anything close to actual self-sufficiency, and there's one simple reason: chips.

Cutting-edge processors are the most important requirement for AI systems, and Nvidia dominates that market like a monopoly player who owns Boardwalk, Park Place, and somehow convinced everyone else to pay double rent. The company accounts for about 90% of all commercially available AI chips, which explains why it has played a central role in nearly all sovereign AI initiatives.

Nvidia's only serious rival is AMD—which is also American. The servers that house these chips are mostly built by Dell and Supermicro—surprise, surprise—also American companies.

Even China, which has built something close to a self-sufficient AI stack, has yet to develop an alternative to America's "whizziest chips" (and yes, that's a technical term).

The Emperor's New AI Factory

So here we are, watching governments around the world spend hundreds of billions of dollars on what Kevin Xu of Interconnected Capital warns might create "more like a palace than a factory."

And you know what? That's probably exactly what's going to happen.

Amazon and Microsoft are already pitching sovereign clouds with enhanced data controls and dedicated local infrastructure. These cloud giants could probably achieve the same objectives more efficiently—they have the scale and bargaining power that individual countries lack. National AI models could simply be built on top of these existing platforms, saving everyone a lot of money and headaches.

But efficiency isn't always the point in government spending, is it?

Sometimes you need to build something big and expensive to show your citizens that you're serious about staying competitive in the global AI race, even if that something is essentially a very expensive way to rent processing power from Nvidia.

The Bottom Line

Sovereign AI initiatives may help governments achieve some legitimate objectives—keeping data secure, widening access to AI technology, and developing local expertise.

These aren't trivial benefits, and for some countries, they might justify the enormous costs involved.

But let's be clear about what's really happening here.

This is Jensen Huang's masterpiece—a way to turn nationalist anxieties about AI dependence into a $1 trillion revenue opportunity, all while countries remain fundamentally dependent on American hardware.

It's brilliant, really. Sell countries the dream of AI independence while making them more dependent on your products than ever before.

The sovereign AI gold rush is just getting started, and Nvidia is sitting pretty with the only shovels in town. Whether taxpayers are getting good value for their money remains to be seen, but one thing's for certain: Jensen Huang is having a very good decade.

As governments continue to chase the mirage of AI independence, perhaps the real question isn't whether sovereign AI is worth the cost, but whether we're all just participants in the most expensive marketing campaign in history.

Either way, Nvidia shareholders probably aren't complaining.

About the author: Rupesh Bhambwani is a technology enthusiast specializing in the broad technology industry dynamics and international technology policy.

When not obsessing over nanometer-scale transistors, energy requirements of AI models, real-world impacts of the AI revolution and staring at the stars, he can be found trying to explain to his relatives why their smartphones are actually miracles of modern engineering, usually to limited success.

Leave a comment

Have you ever wondered how you can predict what happens next when watching a movie or playing a game? Your brain creates a mental simulation—a “world model”—that lets you anticipate outcomes without experiencing them first.

It’s why you know not to touch a hot stove (well, after that first time), and why you can catch a ball without calculating its parabolic trajectory. Imagine if you had to do that every time you had to catch a ball. Phew

Now, researchers are giving AI systems this same remarkable ability. And before you worry about Skynet becoming self-aware, let me assure you: we’re still at the stage where AI can barely understand why cats knock things off tables. Baby steps.

An Idea Born Decades Ago

The concept of world models traces back to 1943, when Scottish psychologist Kenneth Craik proposed that organisms carry “small-scale models” of their environment. Craik was basically saying, “Hey, maybe our brains aren’t just reacting to stuff—maybe they’re running little simulations!” This was pretty revolutionary thinking, especially considering World War II was happening and people had slightly bigger concerns at the time.

But it wasn’t until much later that AI researchers began implementing this idea computationally.

In 1990, Richard S. Sutton introduced the Dyna algorithm, a foundational approach to model-based reinforcement learning that integrated learning, planning, and direct reinforcement learning. Sorry, a lot of geek talk. But stay with me, am going to simplify it.

Think of Dyna as AI’s attempt to learn from its mistakes without actually making all of them—kind of like watching your friend touch the hot stove instead of doing it yourself.

The term “world models” itself gained serious momentum in 2018 when researchers David Ha and Jürgen Schmidhuber published their influential “World Models” paper. Suddenly, everyone in AI research was talking about it.

Think of world models as AI’s imagination engine. World foundation models are neural networks that simulate real-world environments and predict accurate outcomes based on text, image, or video input. Just as you can close your eyes and imagine throwing a ball, predicting its arc and landing spot (with varying degrees of accuracy, depending on your athletic prowess), a world model allows AI to simulate scenarios internally before taking action.

World models take inspiration from the mental models that humans develop naturally, transforming abstract sensory representations into a concrete understanding of the world.

This is fundamentally different from traditional AI approaches that simply react to inputs without truly “understanding” their environment.

It’s the difference between a chess player who memorizes openings versus one who understands strategy. One is playing chess; the other is just moving pieces and hoping for the best.

Meta’s chief AI scientist, Yann LeCun, has thrust this issue into the spotlight, arguing that world models—AI systems that form internal representations of structure, dynamics and causal relationships—could be key to advancing artificial intelligence.

And when someone with LeCun’s credentials says something matters, the AI community listens.

How Do World Models Work?

Firstly, world models enable sample efficiency, which means AI systems can learn from fewer real-world interactions by practicing in their simulated environments. Researchers have even trained agents entirely inside their own hallucinated dreams (whatever that means) generated by world models, then successfully transferred these policies back into actual environments.

It’s like learning to drive in a video game and then somehow being decent at actual driving. Except, you know, it actually works for AI. (Please don’t try this at home.)

Second, world models provide predictive capabilities. They help AI predict outcomes, reason about environments, and guide decision-making in ways that mirror how even young children intuitively develop structured, adaptive representations.

Your three-year-old nephew understands that if he drops his ice cream, it will fall down (and he will be sad). That’s a world model at work. We’re trying to give AI that same intuitive physics, minus the tantrum.

Third, they offer interpretability. When an AI system has an internal model of how the world works, we can better understand its decision-making process—a crucial advantage over black-box neural networks.

Traditional neural networks are like that friend who gives you advice but can’t explain why. World models are more like that friend who not only gives advice but shows you their entire reasoning process, complete with diagrams. Both friends are helpful, but one makes you feel less like you’re blindly trusting a crystal ball.

The Neurosymbolic Connection

Here’s where things get interesting, and by interesting, I mean “the kind of fascinating that makes you forget to eat lunch.”

World models connect deeply with neurosymbolic AI, an approach that combines neural networks’ learning power with symbolic reasoning’s logical structure.

Researchers have introduced neuro-symbolic world models like WorldCloner, which learns efficient symbolic representations of transitions and uses these models to improve how agents adapt to novelties.

Think of it this way: neural networks provide the pattern recognition (like recognizing a chair), while symbolic reasoning provides the logic (understanding that chairs are for sitting, can be moved, and exist in specific locations). Together, they’re like the ultimate odd couple, except instead of arguing about cleaning, they’re solving complex AI problems.

Neurosymbolic AI, which combines neural networks with symbolic reasoning, gained wider adoption in 2025 to address hallucination issues in large language models. And yes, “hallucination” is the actual technical term. It’s what happens when AI confidently tells you that Napoleon invented the dishwasher or that penguins can fly. (They cannot. I checked. Twice.)

By grounding world models in both learned patterns and logical rules, we create AI systems that are both flexible and reliable.

Meta’s I-JEPA (Image Joint Embedding Predictive Architecture) learns by creating an internal model of the outside world, which compares abstract representations of images rather than comparing the pixels themselves.

This is a huge deal because it means AI can understand concepts rather than just memorizing pixel patterns. It’s the difference between understanding what a dog is versus just memorizing what 50,000 dog pictures look like. One approach scales; the other requires the internet’s entire collection of dog photos.

World models aren’t just theoretical—they’re already changing industries. And no, I don’t mean “changing industries” in the vague, buzzwordy way that crypto bros talked about blockchain. I mean actual, tangible applications that you could see in the real world if you looked outside right now. (Go ahead, I’ll wait.)

Autonomous Driving: Teaching Cars to Predict Your Terrible Lane Changes

World models have emerged as a transformative approach, enabling autonomous driving systems to synthesize sensor data and predict potential future scenarios while compensating for information gaps.

Companies like Wayve have developed GAIA-1, a 9-billion-parameter generative world model that creates high-resolution video predictions with fine-grained control over vehicle behavior and scene features.

Let me put that in perspective: 9 billion parameters. That’s roughly the number of neurons in a human brain, except this brain only thinks about driving and doesn’t get distracted by wondering what’s for dinner or why your high school crush never texted back.

Researchers from Meta, NYU, and Berkeley AI Research have proposed a Navigation World Model (NWM), a controllable video generation model that enables agents to simulate potential navigation plans and assess their feasibility before taking action. It’s like having a GPS that can actually visualize your entire route before you start, including that weird merge on the highway that always confuses you.

Robotics: When Machines Learn to Think Before They Act

Meta announced an AI world model in June 2025 that better understands 3D environments and movements of physical objects, advancing both robotics and self-driving car development. This allows robots to plan complex tasks by mentally simulating their actions before executing them.

This is crucial because robots, historically, have been about as graceful as a giraffe on ice skates. By letting them practice in simulation first, we’re finally getting robots that don’t immediately knock everything over like a mechanical bull in a china shop. Progress!

Warehouse Automation: Amazon’s Robots Get Smarter

Amazon applied neurosymbolic AI in its Vulcan warehouse robots and Rufus shopping assistant to enhance accuracy and decision-making, demonstrating how world models improve practical industrial applications. These robots now understand not just “move package from Point A to Point B” but “this package is fragile, those packages are heavy, and that package is blocking the path, so I should reroute.”

It’s the difference between a robot worker who follows orders blindly and one who actually understands the job. One is a machine; the other is beginning to look suspiciously competent.

The Yann LeCun Factor

LeCun predicts a “new paradigm of AI architectures” within 5 years and a “decade of robotics,” with world models at the heart of systems that promise to help machines understand the dynamics of the real world, including memory, common sense, intuition, and reasoning capabilities—traits far beyond current systems.

When someone who won the Turing Award (basically the Nobel Prize of computing) tells you something is going to revolutionize AI, you pay attention.

LeCun described how a world model could help achieve desired goals through reasoning, enabling sophisticated forecasting and planning in both digital and physical realms.

We’re not just talking about AI that can play chess or generate pretty pictures. We’re talking about AI that can reason about the world the way you and I do when we plan our day or figure out how to rearrange furniture without measuring anything. (Though let’s be honest, we should probably measure.)

The Challenges

Now, before you start imagining a utopian future where AI solves all our problems, let’s pump the brakes and talk about reality. World models face significant challenges, and I’m not just saying that to seem balanced and thoughtful (if you ask my kids, they will say that I am neither of those)

World models, like all AI models, also hallucinate and internalize biases in their training data. A world model trained largely on videos of sunny weather in European cities might struggle to comprehend or depict Korean cities in snowy conditions, or simply do so incorrectly.

This is a real problem. If your AI’s understanding of “world” is really just “places that look like the dataset,” you’re in trouble. It’s like someone who’s only seen movies filmed in Los Angeles trying to explain what winter is. Sure, they’ve heard about it, but their mental model is fundamentally flawed.

There’s a gap between AI’s command of human knowledge and its inability to match an animal’s understanding of the physical world—a crucial challenge in the quest for more intelligent machines.

Your cat understands object permanence better than most AI systems. That toy didn’t disappear when you put it under the couch—it’s still there, just hidden. Many AI systems would be genuinely confused by this magic trick that entertains six-month-old babies.

There are also computational challenges. Building and running world models requires enormous computing power. To build general world models, researchers must address several open challenges, including generating consistent maps of environments and capturing not just dynamics but complex interactions. It’s not just about making the model work; it’s about making it work efficiently enough to be practical.

And let’s not forget the technical innovations in systems like Google DeepMind’s Genie 3 introduce new challenges for safety and responsibility. When you give AI the ability to imagine and simulate worlds, you need to think carefully about what happens when those simulations affect real-world decisions. It’s the AI equivalent of “with great power comes great responsibility,” except Spider-Man’s uncle was probably thinking about something less computationally expensive.

The Future Is Imaginative

Looking ahead, world models promise to revolutionize how AI interacts with our physical world. And I use “revolutionize” carefully here—not in the “this app will revolutionize how you order pizza” way (many people do think on these lines), but in the “this actually changes fundamental capabilities” way.

Imagine:

Medical robots that can simulate surgical procedures with perfect precision before making a single cut. Your surgeon could have an AI assistant that’s practiced your exact operation thousands of times in simulation, accounting for your specific anatomy. It’s like having a surgeon who’s done your procedure a million times, except they’re doing it for the first time on you but with all that simulated experience. Wild.

Climate models where AI can predict environmental changes across decades by understanding complex ecological relationships. Instead of just extrapolating trends, these models would understand cause and effect: “If deforestation increases here, rainfall patterns change there, which affects agriculture over there, which impacts migration patterns everywhere.” It’s like having a crystal ball, except this one uses physics instead of mysticism.

Personal assistants that truly understand your daily routines, anticipating your needs by maintaining rich models of your preferences and environment. Not in a creepy surveillance way (hopefully), but in a “oh, you usually get coffee around now, and traffic is bad on your normal route, so I’ve found an alternative that passes by your favorite coffee shop” way. It’s the difference between Siri and an assistant who actually knows you.

Game AI that creates infinitely varied, realistic scenarios by understanding underlying game mechanics rather than just reacting to inputs. Imagine video games where NPCs don’t just follow scripts but actually understand their world and make decisions based on their goals and understanding. No more “I used to be an adventurer like you, then I took an arrow in the knee” repeated ad nauseam. NPCs could have actual personalities and memories and react to you in meaningful ways.

Educational systems where AI tutors understand not just the subject matter but how learning works, adapting in real-time to your understanding and struggles. Instead of one-size-fits-all education, you get teaching that’s tailored to exactly how your brain processes information. It’s like having a personal tutor who never gets tired, never gets frustrated, and has infinite patience for your questions about why calculus exists.

World models have the potential to solve general-purpose simulation and evaluation, enabling robots that are safe, reliable, and intelligent in a wide variety of scenarios. We’re talking about AI that doesn’t need specific programming for every possible situation but can reason its way through novel problems using its understanding of how the world works.

The Philosophy Bit

Here’s something that keeps me up at night: what does it mean for AI to “understand” the world? When a world model predicts that a dropped ball will fall, is it understanding gravity, or is it just pattern matching on steroids?

Humans have this amazing ability to build internal models of the world that are simultaneously wrong and useful. We don’t understand quantum mechanics, but we understand that solid objects are solid (even though they’re mostly empty space at the atomic level). Our world models are pragmatic approximations, good enough to get us through the day without requiring a PhD in physics to walk down stairs.

AI world models might be similar—not perfect representations of reality, but useful approximations that enable intelligent behavior. And maybe that’s okay. Maybe understanding doesn’t require perfect fidelity; maybe it just requires sufficient coherence to enable prediction and planning.

Or maybe I’m overthinking it. Probably that second one.

The Bottom Line (Finally)

World models represent a fundamental shift in how we approach AI. Instead of building systems that merely respond to inputs like the world’s most sophisticated reflex arc, we’re creating AI that genuinely understands its environment—systems that can imagine, predict, and plan like you and I do naturally (though hopefully with less procrastination).

By combining the pattern-learning power of neural networks with the logical structure of symbolic reasoning, and grounding everything in predictive models of how the world works, we’re moving closer to an AI that doesn’t just process information but truly comprehends it.

The future of AI isn’t just about faster computers or bigger datasets—though those certainly help, and the electricity bills are eye-watering.

It’s about building systems that can dream, imagine, and understand the world they inhabit. Systems that don’t just know that a ball falls when dropped, but understand why, can predict how it will bounce, and can reason about what happens if you change variables like the ball’s material or the surface it hits.

And that future is already beginning to unfold. With LeCun predicting major architectural changes within five years and a decade of robotics breakthroughs ahead, we’re standing at the threshold of something genuinely transformative.

Will world models solve all of AI’s problems? No.

Will they occasionally hallucinate that penguins can fly? Probably.

But will they fundamentally change how we build intelligent systems? Absolutely.

The age of AI that can imagine is here. And unlike that hot stove you touched as a kid, this is one future we don’t need to learn about the hard way—we can simulate it first.

About the author: Rupesh Bhambwani is a technology enthusiast specializing in the broad technology industry dynamics and international technology policy.

When not obsessing over nanometer-scale transistors, energy requirements of AI models, real-world impacts of the AI revolution and staring at the stars, he can be found trying to explain to his relatives why their smartphones are actually miracles of modern engineering, usually to limited success.

Leave a comment

Somewhere off the coast of Shanghai, in the murky depths where fish outnumber humans (thankfully), a giant yellow capsule full of servers is quietly humming away, living its best submarine life.

Welcome to the future of data storage, where the solution to overheating computers is apparently to dunk them in the ocean like an oversized tea bag.

Because if there’s one thing the internet needed, it’s to literally go fishing for data.

Before we dive deeper than these aquatic servers, let’s address the whale in the room: data centers are energy gluttons.

Data centers accounted for about 1.5 percent of global electricity consumption in 2024, which might not sound like much until you realize that’s roughly equivalent to powering an entire mid-sized country.

And thanks to AI’s insatiable appetite for computational power, the sector’s global electricity consumption would more than double between 2024 and 2030, reaching 945 terawatt-hours by the end of the decade, equivalent to the current electricity demand of Japan.

Think about that for a moment. ChatGPT pondering your request for a haiku about cats requires the same industrial infrastructure that keeps Tokyo’s subway running. Suddenly, those AI-generated images of dogs wearing top hats seem a bit more ethically complicated.

The problem isn’t just the computing itself – it’s the cooling. Traditional data centers run hot enough to fry an egg (not that anyone’s tried, but the fire marshals would definitely have opinions).

Most facilities currently rely on energy-intensive air conditioning systems or water evaporation to prevent their servers from melting into expensive puddles of silicon.

Enter the Underwater Solution

This is where China’s maritime equipment firm Highlander decided to think outside the box – or rather, inside a waterproof yellow capsule.

Working with state-owned construction companies, they’ve developed an underwater data center pod that’s scheduled to be submerged off Shanghai’s coast in mid-October.

The concept is elegantly simple: let the ocean do the heavy lifting when it comes to cooling.

That’s not a typo. Ninety percent. It’s the kind of efficiency that makes traditional data center operators weep into their electricity bills.

The technology isn’t entirely unprecedented. Microsoft tested a similar concept with Project Natick off the coast of Scotland back in 2018. The project proved that submerged servers failed eight times less frequently than their land-based cousins, which is impressively reliable considering they’re essentially running a computer in a giant fish tank.

However, Microsoft retrieved its pod in 2020, declared mission accomplished, and promptly wandered off to pursue other ventures, leaving everyone wondering if they’d discovered something brilliant or just really expensive.

China’s approach appears more commercially minded. The Shanghai pod will serve clients, including China Telecom and a state-owned AI computing company, making it one of the world’s first commercial underwater data center services.

It’s part of a broader government initiative to reduce data centers’ carbon footprint, backed by the kind of subsidies that make experimental technology actually viable. Highlander received 40 million yuan (approximately $5.62 million) for a similar 2022 project in Hainan province, which is still operational and presumably keeping the local fish population thoroughly confused.

The Engineering Challenge

But. Building an underwater data center isn’t quite as simple as wrapping your laptop in cling film and tossing it into the harbor (which I absolutely do not recommend, regardless of how many times Windows has crashed today).

Zhou Jun, an engineer for Highlander’s Shanghai project, admitted that “the actual completion of the underwater data centre involved greater construction challenges than initially expected,” – which is engineer-speak for “this was way harder than we thought, but we’re too proud to elaborate.”

The most obvious challenge is keeping water away from electronics, which have historically maintained a famously antagonistic relationship.

Seawater, with its enthusiastic salt content, is particularly eager to corrode metal components into expensive Swiss cheese. The Chinese project tackles this by coating the steel capsule with glass flakes, creating a protective barrier that hopefully tells the ocean to kindly back off.

The facility is constructed onshore in separate components before being assembled underwater, like the world’s most expensive IKEA furniture. To enable maintenance access (because someone eventually needs to turn it off and on again), an elevator connects the main pod structure to a segment that remains above water. It’s essentially a submarine data center with a surface lobby – the world’s most stressful commute for IT technicians.

Power delivery presents another engineering puzzle. The Shanghai installation draws nearly all its electricity from nearby offshore wind farms, with Highlander claiming that more than 95 percent of the energy used comes from renewable sources.

It’s green energy powering servers cooled by natural ocean currents – the kind of sustainable synergy that makes environmentalists cautiously optimistic and investors nervously excited.

Shaolei Ren, an expert from the University of California, Riverside, notes that projects like this are currently focused on demonstrating “technological feasibility” rather than immediate commercial viability.

Translation: they’re figuring out whether this is genius or madness, and the jury’s still out.

The Internet Connection Conundrum

Connecting an underwater data center to the broader internet involves more complexity than running an Ethernet cable really, really far.

As Ren points out, laying the internet connection between an offshore facility and the mainland is significantly more complex than with traditional land-based servers. You can’t just call your local cable company and ask them to run fiber optic to “somewhere in the ocean, you’ll know it when you see the big yellow thing.”

There’s also the fascinating security vulnerability that researchers at the University of Florida and the University of Electro-Communications in Japan discovered: submarine data centers can potentially be attacked using sound waves conducted through water.

Because apparently, the internet wasn’t hackable enough already – now we need to worry about acoustic cyber-attacks that sound like something from a James Bond film rejected for being too implausible.

Here’s where things get murky (pun absolutely intended). While underwater data centers dramatically reduce energy consumption for cooling, they don’t eliminate heat generation – they just dump it directly into the ocean. Think of it as the world’s most expensive radiator.

The warming effect on surrounding water has raised legitimate questions about impacts on marine ecosystems. Andrew Want, a marine ecologist at the University of Hull, explains that the heat emitted could attract certain species while driving others away. It’s like installing an underwater heat lamp that might turn into an unplanned fish discotheque or an aquatic exclusion zone, depending on who likes the temperature.

“These are unknowns at this point – there’s not sufficient research being conducted yet,” Want notes, which is scientist-speak for “we honestly don’t know if we’re creating underwater spa resorts for lobsters or accidentally cooking them.”

Highlander conducted a 2020 independent assessment of their test project near Zhuhai in southern China, which indicated that surrounding water stayed well below acceptable temperature thresholds. However, Ren warns that scaling up centers would proportionally scale up heat output, making thermal pollution a more serious concern.

It’s a classic technological trade-off: solve one environmental problem (energy consumption) while potentially creating another (localized ocean warming). The keyword here is “potentially” – we genuinely don’t know the full ecological impact yet because this technology is so new. It’s like asking whether smartphones would be popular in 1995; the answer seems obvious in retrospect, but you couldn’t be certain at the time.

The Commercial Reality Check

Despite the impressive energy savings and Microsoft’s successful trial, underwater data centers face significant hurdles before becoming mainstream.

The substantial construction challenges, complex maintenance requirements, vulnerability to marine conditions, and uncertain environmental impacts all contribute to a technology that’s fascinating but not yet proven at a commercial scale.

Ren suggests that offshore facilities will likely complement rather than replace traditional data centers, providing services to “niche segments” rather than becoming the industry standard. It’s not a silver bullet for data center energy consumption – more like a specialized tool that might work brilliantly in specific circumstances.

There’s also the practical matter that most companies prefer their critical infrastructure to be easily accessible. When your server crashes at 3 AM, “it’s underwater off the coast of Shanghai” is not the ideal response from your IT department.

The elevator access helps, but it’s still more complicated than walking into a climate-controlled building in a suburban office park.

The Bigger Picture

The emergence of underwater data centers reflects a broader reality: the tech industry is desperately searching for solutions to its growing energy problem.

From 2024 to 2030, data centre electricity consumption grows by around 15% per year, more than four times faster than the growth of total electricity consumption from all other sectors. That’s unsustainable by any reasonable metric.

AI development has turbocharged this crisis. Every time someone asks an AI to write a sonnet or generate an image of a cat riding a dragon, they’re contributing to an exponentially growing demand for computational power.

Please, for heaven’s sake, stop generating cat images.

The technology we’re building requires infrastructure that our current energy systems struggle to support.

Some solutions are pragmatic (better cooling systems, more efficient chips), while others are wonderfully absurd – Jeff Bezos recently announced plans to build AI data centers in space, because when you’re a billionaire, “underwater” apparently isn’t exotic enough.

The solution might involve underwater pods, orbital servers, innovative cooling technologies, or more likely, a combination of everything we can think of.

Brilliantly Experimental or Experimentally Brilliant?

China’s underwater data centers represent genuinely innovative thinking about a real problem. The 90 percent reduction in cooling energy consumption is remarkable, the renewable energy integration is commendable, and the sheer audacity of dunking servers in the ocean deserves respect.

However, significant questions remain unanswered.

The long-term environmental impact needs a thorough study. The commercial viability at scale remains unproven. The maintenance complexity and security vulnerabilities present real challenges.

And there’s something fundamentally amusing about solving digital infrastructure problems by literally throwing computers into the sea – it feels like the technological equivalent of “have you tried turning it off and on again, but underwater?”

What’s certain is that the data center energy crisis demands creative solutions. Whether those solutions involve underwater pods, space stations, or something we haven’t imagined yet, the industry must innovate. Traditional approaches won’t scale to meet AI’s growing appetite for computational power.

In the meantime, somewhere off Shanghai’s coast, a giant yellow capsule will soon be humming away beneath the waves, simultaneously representing humanity’s technical ingenuity and our somewhat desperate attempts to keep our digital infrastructure from overheating. The fish, presumably, have no opinion on the matter.

But here’s the thought that keeps some engineers awake at night: if we’re already putting data centers underwater and planning them for space, where exactly do we go from there?

Underground volcano lairs? Antarctic ice shelves? The possibilities are both endless and slightly concerning.

For now, we’ll have to wait and see whether China’s deep dive into ocean cooling proves to be the future of data infrastructure or just a really expensive science experiment. Either way, it’s certainly making waves in the industry.

In the end, perhaps the real question isn’t whether we should put data centers underwater, but whether the internet itself has become so absurdly large that we need to consider every conceivable location – terrestrial, aquatic, or otherwise – just to keep the memes flowing and the AI chatbots chatting.

About the author: Rupesh Bhambwani is a technology enthusiast specializing in the broad technology industry dynamics and international technology policy.

When not obsessing over nanometer-scale transistors, energy requirements of AI models, real-world impacts of the AI revolution and staring at the stars, he can be found trying to explain to his relatives why their smartphones are actually miracles of modern engineering, usually to limited success.

Leave a comment

Welcome to the Dead Internet Theory, where the web has become less “worldwide” and more “bot-wide.”

There’s a delicious irony in Sam Altman, CEO of OpenAI—the company that gave us ChatGPT—experiencing what can only be described as a Frankenstein moment.

Altman recently posted on X (because apparently we’re still not calling it Twitter): “I never took the dead internet theory that seriously, but it seems like there are really a lot of LLM-run Twitter accounts now.”

The Dead Internet Theory, for the uninitiated, proposes that the majority of online content and interactions are now generated by algorithms and bots rather than actual humans.

The Numbers Don’t Lie (But the Bots Might)

Let’s talk statistics, shall we?

According to security firm Imperva’s 2024 Bad Bot Report, approximately half of all internet traffic came from bots last year.

That’s right—for every legitimate cat video you watch, there’s a bot somewhere generating content trying to sell you knockoff designer handbags.

But wait, it gets better.

Predictions suggest that by 2030, more than 70% of online content might be AI-generated. Given that we’re already in 2025, we’re essentially living in the prologue to this apocalyptic scenario. It’s like watching a horror movie where the protagonist just walked into the creepy abandoned house—we know what’s coming, but we can’t look away.

A May 2025 review from NewsGuard found over one thousand news sites run almost entirely by bots, with 167 masquerading as Russian local news websites. That’s not just a bot problem; that’s a bot industry.

These sites have names like “iBusiness Day” and “Ireland Top News”—legitimate-sounding enough that your uncle might share them on Facebook without a second thought.

Why Your Friends Are More Gullible Than You (Or So You Think)

Here’s where things get properly dystopian. Researchers at the University of Southern California have identified what they call the “Susceptibility Paradox”—a phenomenon where, on average, your social media friends are more susceptible to influence than you are. It’s like the Dunning-Kruger effect’s evil twin.

The study reveals that knowing how someone’s friends behave online is often enough to predict that person’s behavior. For AI systems hoarding our personal data, this is essentially a treasure map to viral manipulation.

Imagine an army of bots that not only know what makes you tick but also understand your entire social network’s behavioral patterns. It’s targeted advertising on steroids, except instead of selling you shoes, it might be selling you political ideologies or conspiracy theories.

The Viral Mathematics of Misinformation

Think information spreads like a simple game of telephone? Think again.

What goes viral online follows a pattern similar to sixth-generation fires—the kind that can’t be extinguished with conventional methods.

The information mutates as it spreads, becoming stronger and more infectious with each share. It’s evolution in real-time.

AI doesn’t just participate in this process—it optimizes it. These systems can create content specifically engineered to exploit these viral pathways, like a biological weapon designed to spread through social networks with mathematical precision.

Amid all this doom and gloom, there are voices of reason. Aaron Harris, Global CTO at Sage, offers a refreshingly pragmatic take: “I don’t know if I would call it ‘the dead internet,’ but it’s certainly changing rapidly.”

Harris argues for an “ethical internet,” but warns it won’t happen by accident. Companies need to make their AI systems auditable and explainable, with the same rigor we apply to financial systems. After all, if your bank can track every penny, surely we can figure out why an algorithm is recommending you join a conspiracy theory group.

The key principle?

AI should enhance human capabilities, not replace them. It’s a nice idea, though one wonders if we’re already past the point of no return.

Researchers studying this phenomenon have proposed a series of principles reminiscent of Isaac Asimov’s laws of robotics, but for AI content generation:

1. AI must not manipulate users to serve its own interests or create social harm

2. AI must not facilitate strategies that negatively affect society, like behavioral conditioning or institutional discrediting

3. AI must not unduly restrict user freedom

These sound wonderful on paper—like a Bill of Rights for the digital age. The challenge, of course, is enforcement.

It’s one thing to write principles; it’s quite another to get every AI developer worldwide to follow them, especially when breaking them is so profitable.

Harris suggests that accountability should be non-negotiable. Transparency, auditability, and explainability need to become the foundation of AI development.

Interestingly, some research suggests that trusted news organizations might actually benefit from the proliferation of AI-generated content, as readers become more discerning about their sources.

Perhaps this is evolution, too.

As the internet becomes increasingly polluted with bot-generated content, humans might develop better filters, both mental and technological.

We might actually start reading past headlines, checking sources, and questioning whether that perfectly formatted thread about a controversial topic might just be automated engagement bait.

Or maybe that’s wishful thinking. After all, humans have been falling for scams since someone first traded magic beans for a cow.

Are We Already Living in the Dead Internet?

Here’s the uncomfortable question: how much of your online interaction today was with actual humans?

That supportive comment on your post? Could be a bot.

That heated political argument? Possibly an AI trained to maximize engagement through controversy.

That recipe blog with seventeen paragraphs of backstory before the ingredients? Almost certainly... well, that one’s probably human. No AI would be that cruel.

The Dead Internet Theory isn’t necessarily claiming the internet is completely dead—it’s more like it’s becoming a zombie.

It looks alive, it moves around, it even seems to interact with you. But underneath, something fundamental has changed. The authenticity has been replaced by simulation, and we’re all playing along, sometimes without even knowing it.

Keeping the Internet Human

So what’s the solution? Do we rage against the machine? Retreat to carrier pigeons and handwritten letters?

Probably not—though the postal service would appreciate the business.

The answer likely lies in conscious engagement. We need to:

• Question the source of content before sharing it

• Support platforms that prioritize transparency and accountability

• Demand better regulation that puts human interests first

• Develop and use tools that can identify AI-generated content

• Remember that if something seems designed to make you angry or afraid, it probably is—and that’s a red flag

Most importantly, we need to maintain our humanity online.

Be kind, be thoughtful, be authentic. If the internet is being overrun by bots, the best defense is to be undeniably, messily, gloriously human.

The Dead Internet Theory started as a fringe conspiracy theory, the kind of thing you’d find in the darker corners of obscure forums. Now, it’s being discussed by the very people who built the technology that might make it reality.

There’s something poetically tragic about Sam Altman’s realization.

The internet isn’t dead yet, but it’s certainly not well. Whether it recovers or succumbs to bot-induced sepsis depends on the choices we make right now.

We created this problem; we can fix it—assuming, of course, that “we” are still mostly human (and around).

Now, if you’ll excuse me, I need to go argue with someone on social media about whether cats are better than dogs

And yes, I’m fully aware they might be a bot. But they’re still wrong about cats.

Don’t hate me for that.

About the author: Rupesh Bhambwani is a technology enthusiast specializing in the broad technology industry dynamics and international technology policy.

When not obsessing over nanometer-scale transistors, energy requirements of AI models, real-world impacts of the AI revolution and staring at the stars, he can be found trying to explain to his relatives why their smartphones are actually miracles of modern engineering, usually to limited success.

Leave a comment

Are you still using ChatGPT to compose emails, provide structure to your blog articles, research on topics, or fact-check information?

Well, those days are officially over.

At its annual developers event on Monday, OpenAI unveiled a feature that’s less “incremental update” and more “wait, ChatGPT is becoming an operating system?”

The company announced that ChatGPT users can now connect with third-party apps directly within the chatbot, turning it from a conversational AI into something that looks suspiciously like the next evolution of how we interact with digital services.

Want to create a weekend playlist? Ask ChatGPT, and it’ll tap into Spotify without you lifting a finger to switch apps.

Need a three-bedroom house in a specific neighborhood? ChatGPT will pull up Zillow listings faster than you can say “mortgage calculator.”

From Chatbot to Super Assistant

“We never meant to build a chatbot when we built ChatGPT,” confessed Nick Turley, who runs the ChatGPT team. “We set out to build a super assistant and we got a little sidetracked.”

Talk about an understatement. It’s like saying Uber got “a little sidetracked” from being a black car service, or that Instagram was “briefly” a photo-sharing app before becoming a shopping platform, dating hub, and existential crisis generator.

The new feature, available to logged-in ChatGPT users outside the EU across Free, Go, Plus and Pro plans, includes integrations with Booking.com, Canva, Coursera, Figma, Expedia, Spotify, and Zillow.

OpenAI calls this capability “talking to apps,” which is perhaps the most straightforward name the company has ever given anything. Users need to sign in to these services the first time, after which ChatGPT can work its magic across platforms.

Brad Lightcap, OpenAI’s chief operating officer, sees enormous potential in this shift. “There’s an opportunity for builders to create entirely new applications that are even native to ChatGPT, and of course for services you love to be able to benefit there, too,” he said during Monday’s event.

By opening ChatGPT to third-party apps and payments, OpenAI is taking a major step toward transforming ChatGPT from a chatbot into a full-fledged AI operating system that combines conversational intelligence, rich interfaces, and embedded commerce.

Think of it as the App Store moment for AI, except instead of tapping icons, you’re just... talking.

The Stock Market’s New Kingmaker

Here’s where things get weird. When Sam Altman mentioned specific companies during the developer event, their stock prices jumped. Like, immediately. Figma, Expedia Group, and Booking Holdings all saw their shares pop after getting the ChatGPT seal of approval.

“This is a strange new thing that started happening,” Altman said during a press briefing, with the slightly bewildered tone of someone who just discovered they can accidentally move markets with their words. “We are trying to figure out how to adjust for this kind of world, but it’s weird.”

Weird indeed. The last person who could casually move stock prices just by mentioning companies was probably Elon Musk on Twitter, and look how well that turned out for everyone involved.

The Numbers Behind the Hype

Let’s talk scale for a moment, because the numbers here are genuinely bonkers.

Altman revealed that OpenAI now has more than 800 million weekly ChatGPT users. To put that in perspective, that’s roughly equivalent to the entire population of Europe checking in with ChatGPT every single week.

This astronomical growth has translated into serious revenue. In September, OpenAI CFO Sarah Friar said the company was on track to generate $13 billion in revenue this year, up from $4 billion in 2024. That’s a 225% increase in a single year—the kind of growth that makes venture capitalists lose sleep (the good kind of sleep-losing, apparently).

But here’s the catch: despite raking in billions, OpenAI remains unprofitable. The company burns through billions of dollars annually on computing costs, research expenses, and its recent spending spree on chips and data centers. It’s the Silicon Valley equivalent of making it rain while standing in a hurricane of expenses.

“Obviously, one day we have to be very profitable and we’re confident and patient that we will get there,” Altman said with the calm reassurance of someone spending money they don’t technically have yet. “But right now, we’re in a phase of investment and growth.”

They’re pulling a classic tech startup move—grow first, profit later, and hope the later part happens before the investors get nervous.

Beyond app integrations, OpenAI also released AgentKit, a software development kit designed to help developers build AI agents. AI agents are systems or programs capable of autonomously performing tasks on behalf of users or other systems by designing their workflow and utilizing available tools.

Think of agents as ChatGPT’s more proactive cousins—instead of waiting for you to ask questions, they go out and do things for you. In 2025, multi-agent collaboration is becoming more common, where multiple AI agents work together to execute complex, end-to-end tasks like inventory tracking and customer lifecycle management.

Companies like Albertsons, Box, and Canva have already built agents using OpenAI’s tools. “Software used to take years or months to build,” Altman noted. “It can take minutes now.”

Sure, and websites used to be built with <marquee> tags and geocities templates, but here we are. The real question isn’t how fast we can build things—it’s whether we should be building them at all. But that’s a philosophical debate for another article, preferably one written by humans while we still have jobs.

In a move that might surprise people familiar with OpenAI’s historically proprietary approach, the company announced that its foray into open-source models has generated 23 million downloads on Hugging Face.

The company also introduced smaller voice and image-generation models optimized for speed and lower costs, acknowledging that not everyone needs the computational equivalent of a Ferrari when a reliable Honda will do the job.

The Mystery Hardware Project

Adding star power to Monday’s event was Jony Ive, the legendary Apple designer responsible for making your iPhone look so pretty you forgive it for breaking every two years. Ive and Altman have teased an AI hardware collaboration, but details remain maddeningly vague.

Ive revealed that his team has generated about 15 to 20 “really compelling product ideas,” but deciding which ones to pursue has been challenging. “It would be easy if we knew there were three good ones,” he said. “It’s just not like that.”

This is designer-speak for “we have too many good ideas and executive dysfunction is real.” Given Ive’s track record, whatever eventually emerges will probably be beautiful, minimalist, and cost approximately one kidney.

ChatGPT as Your Digital Brain

According to Nick Turley, over the next six months, ChatGPT will evolve even further into something resembling an operating system where users can access more software and new services. Unlike OpenAI’s previous GPT Store, which was a separate app marketplace, this launch puts apps directly in ChatGPT’s responses and lets users call up third-party tools in their everyday conversations.

This gives developers better distribution for their apps and gives users a more seamless experience. Instead of jumping between seventeen different apps to plan a vacation, book accommodations, create an itinerary, and make a playlist for the flight, you could theoretically do it all through one conversation with ChatGPT.

The implications are massive. We’re potentially looking at a future where traditional app interfaces become less relevant, replaced by conversational interactions that feel more natural. Instead of learning how each app works, you just... talk. It’s the ultimate user interface—one that’s been around since humans developed language.

OpenAI isn’t operating in a vacuum here. Google has been pushing its Gemini AI hard, Microsoft is deeply invested in AI through its partnership with OpenAI (talk about hedging your bets), and countless startups are building AI agents and assistants. The race to become the platform through which people access digital services is heating up fast.

What makes OpenAI’s approach interesting is the speed of execution. The company completed a secondary share sale that made it the most valuable startup in the world, launched Sora (a social video app that quickly rose to the top of Apple’s App Store), and announced multiple large infrastructure deals with Nvidia and AMD—all within a few frenzied weeks.

It’s the kind of pace that makes you wonder if they’re using AI to schedule their product launches. Actually, given the company we’re talking about, they probably are.

Of course, there’s a question that should keep everyone up at night: what happens when one company has access to your music preferences, travel plans, home searches, design work, and basically every other digital service you use? The article doesn’t mention privacy considerations, which is itself telling.

Apps are available to users outside of the EU—a caveat that likely reflects Europe’s stricter data protection regulations. For everyone else, the convenience of having one AI assistant handle everything comes with the implicit understanding that OpenAI is sitting in the middle of an enormous amount of personal data.

It’s worth thinking about, even if most people will inevitably trade privacy for convenience faster than you can say “I accept the terms and conditions without reading them.”

Strip away the hype, the market movements, and the tech industry’s perpetual need to declare everything “revolutionary,” and what are we left with? A genuinely useful development that could change how we interact with technology.

The promise is simple: instead of context-switching between apps, learning new interfaces, and remembering passwords (okay, having your browser remember passwords), you talk to ChatGPT and it handles the logistics. Planning a trip becomes a conversation, not a tab-juggling exercise. Creating a presentation goes from fighting with software to describing what you want.

The reality will likely be messier. Early integrations will probably be clunky. Some tasks will work brilliantly through conversational AI, while others will still be faster the old-fashioned way. Security issues will emerge. Someone will inevitably try to jailbreak the system to do something ridiculous.

But the direction is clear: OpenAI is betting big that conversational AI is the future interface for digital services. And with 800 million weekly users already on board, they might just be right.

Given their trajectory so far, betting against them might not be the wisest move. Just ask the stock market.

About the author: Rupesh Bhambwani is a technology enthusiast specializing in the broad technology industry dynamics and international technology policy.

When not obsessing over nanometer-scale transistors, energy requirements of AI models, real-world impacts of the AI revolution and staring at the stars, he can be found trying to explain to his relatives why their smartphones are actually miracles of modern engineering, usually to limited success.

Leave a comment

In a move that perfectly encapsulates the “we’re making this up as we go” energy of the AI boom, Nvidia is investing up to $2 billion in Elon Musk’s xAI as part of a massive $20 billion funding round.

Yes, you read that correctly: the company selling the chips is now investing in the company buying the chips, who will then rent out those chips to pay back the debt... which is backed by the chips.

If this sounds like the financial equivalent of a snake eating its own tail, you’re not alone. But in the wild west of AI infrastructure financing, this might actually be genius.

The $20 billion raise is split between approximately $7.5 billion of equity and as much as $12.5 billion of debt, structured through a special purpose vehicle (SPV). Think of it as a financial Trojan horse, except instead of soldiers, it’s filled with GPUs and venture capital optimism.

Here’s how this circular economy works:

The SPV buys Nvidia processors, xAI rents them out for five years, and Wall Street financiers eventually recoup their investment. It’s like a lease-to-own scheme, except instead of a Honda Civic, we’re talking about billions of dollars worth of cutting-edge AI hardware.

The brilliance?

The debt is backed by the GPUs themselves, not the company. This could provide a playbook for tech firms looking to decrease their balance sheet exposure while still gorging on the silicon they desperately need.

All these GPUs are destined for Colossus 2, xAI’s Memphis-based data center that’s quickly becoming the stuff of legend. As of June 2025, the Colossus supercomputer consists of 150,000 H100 GPUs, 50,000 H200 GPUs, and 30,000 GB200 GPUs, with another 110,000 GB200 GPUs planned for a second data center in the Memphis area.

To put this in perspective: Nvidia shipped a total of 3.76 million data center GPUs in 2023. xAI is essentially hoarding a significant chunk of global GPU supply in a single location. The original Colossus cluster achieved fame for being built in just 122 days—a timeline that makes most construction projects look like geological processes.

xAI is moving toward a roadmap of 1 million GPUs, which is either visionary or completely bonkers, depending on whom you ask. Probably both.

All this isn’t happening in a vacuum. The AI industry has transformed into a capital-intensive arms race that would make defense contractors blush. Earlier this week, OpenAI announced a multi-year deal with AMD chips, while Meta has signed several multibillion-dollar deals in recent months, including a $29 billion financing package for data centers. Oracle raised a $38 billion debt package for its infrastructure.

In US bond markets alone, tech companies have raised about $157 billion this year—up 70% from last year. At this rate, “billions” is becoming the “millions” of the 2020s.

Here’s where things get peak Elon:

In September, Musk posted on X that xAI was “not raising any capital right now.” Fast forward to October, and the Elon magic is at work again.

To be fair, “right now” is technically accurate if you interpret it with enough creative flexibility. It’s the corporate equivalent of “I’m five minutes away” when you haven’t left the house yet.

The Burn Rate That Burns

xAI was valued at $200 billion in September, making it among the most valuable startups in the world, behind OpenAI. That’s an impressive valuation for a company that already raised about $10 billion earlier this year.

Why the constant need for capital? According to the original reporting, xAI has been burning through approximately $1 billion per month. That’s $33 million per day. Or if you prefer your existential dread in smaller units, roughly $23,000 per minute.

At that rate, even Scrooge McDuck would need a Series B.

The Nvidia Strategy: Invest in Your Best Customers

Nvidia’s CFO Colette Kress told a Goldman Sachs conference in September that the company’s priority was using cash to help other companies deploy AI more quickly. This isn’t charity—it’s strategic genius.

By investing in its customers, Nvidia creates a flywheel effect:

xAI needs GPUs to build AI models, Nvidia provides capital to buy more GPUs, xAI scales faster, Nvidia sells more chips. It’s vertical integration without the messy business of actually acquiring companies and dealing with antitrust regulators.

Plus, when your product is literally the bottleneck for the entire AI industry, helping customers afford more of it is just good business sense.

Who Else Is Writing Checks?

Apollo Global Management is participating in the debt raise, as is Diameter Capital Partners, while Valor Capital is leading the equity portion. These aren’t your typical seed-stage angels betting on a PowerPoint deck—these are serious institutional players betting that AI infrastructure is the new oil.

And honestly, given the capital intensity required, they might be right. You can’t train frontier AI models in a garage anymore (gone are the days when you can launch your startup from just a garage). You need industrial-scale data centers, cooling systems that could chill a small city, and enough electricity to power a small country.

Musk being Musk, he’s turned xAI into a family affair. SpaceX has already invested in xAI, and later this year, Tesla shareholders will vote on whether the electric carmaker should join the party.

xAI is even bolstering its Colossus 2 data center with 168 Tesla Megapacks, because when you own an electric vehicle company, you might as well use it to power your AI ambitions. It’s called synergy, people.

Musk has framed AI as foundational to many of his futuristic products, including self-driving cars and fully autonomous robots. Whether you view this as visionary empire-building or a tech billionaire’s expensive hobby depends largely on your tolerance for chaotic innovation.

A New Playbook for Infrastructure Financing?

The unique structure of this deal—with debt backed by GPUs rather than the company—could fundamentally change how tech firms finance infrastructure. It’s asset-backed lending for the AI age.

Instead of putting the entire company’s creditworthiness on the line, firms can securitize their hardware purchases. The GPUs become the collateral, not the business itself. For an industry that’s moving at breakneck speed with uncertain business models, this could be transformative.

Or it could be a spectacular way to create a new class of financial instruments that explode when the AI bubble pops. Time will tell!

However, not everyone is convinced that more compute necessarily equals better AI. Some researchers argue we’re hitting diminishing returns on simply throwing more GPUs at the problem. There are questions about whether data center capacity is truly the constraint on AI progress, or if we’re just in a very expensive cargo cult phase.

The Memphis community has also raised concerns about the power consumption of these massive facilities. When your data center requires 168 Tesla Megapacks just for backup power, you’re not exactly winning environmental awards.

The Bottom Line

Whether this is brilliant financial engineering or the setup for a “where are they now?” documentary in five years remains to be seen. What’s certain is that the AI infrastructure race has reached a scale that would have seemed absurd just a few years ago.

Nvidia investing in its own customers to buy its own products, which they’ll rent out to pay back debt backed by those same products, is either the future of tech financing or the world’s most expensive example of circular reasoning.

Probably both.

In the meantime, Memphis has become an unlikely hub of AI supercomputing, Wall Street is writing checks with more zeros than a binary joke, and Elon Musk is building what might be the world’s largest collection of GPUs in what was once a riverport town.

As of this writing, representatives for Nvidia declined to comment, a spokesperson for xAI did not respond to requests for comment, and Elon Musk is presumably posting something controversial on X to distract us from asking too many questions about the math.

About the author: Rupesh Bhambwani is a technology enthusiast specializing in the broad technology industry dynamics and international technology policy.

When not obsessing over nanometer-scale transistors, energy requirements of AI models, real-world impacts of the AI revolution and staring at the stars, he can be found trying to explain to his relatives why their smartphones are actually miracles of modern engineering, usually to limited success.

Leave a comment

You’re at a restaurant, confidently ordering the most expensive items on the menu—the wagyu beef, the vintage wine, the flambéed dessert—while quietly hoping your credit card doesn’t get declined. That’s essentially what’s happening in the AI industry right now, except instead of an awkward dinner bill, we’re talking about an $800 billion gap between ambition and reality.

According to Bain & Company’s latest Global Technology Report, the AI industry is racing headfirst into what might be the most spectacular case of “champagne taste on a beer budget” the tech world has ever seen. And folks, the numbers are absolutely wild.

Let’s break down this financial horror show—or comedy, depending on your perspective.

By 2030, AI companies will need $2 trillion in combined annual revenue to fund the computing power needed to meet projected demand. That’s trillion with a T, enough to make even the most optimistic venture capitalist’s eyes water.

But here’s where it gets spicy: even if companies shifted all of their on-premise IT budgets to cloud and reinvested the savings from applying AI in sales, marketing, customer support, and R&D (estimated at about 20% of those budgets) into capital spending on new data centers, the amount would still fall $800 billion short.

To put that in perspective, that $800 billion shortfall is roughly equivalent to the GDP of Switzerland.

We’re not talking about a minor budgeting hiccup—this is more like planning a backyard barbecue and accidentally ordering enough food to feed a small country.

The Data Center Gold Rush

The tech giants—Microsoft, Amazon, Meta, and friends—aren’t exactly pumping the brakes either. They’re expected to collectively spend more than $500 billion on AI infrastructure by the early 2030s.

In 2025, the dollar value contributed to GDP growth by AI data center expenditure surpassed the total impact from all U.S. consumer spending—the first time this has ever occurred.

Yes, you read that correctly. Americans’ collective decision to build giant computer warehouses is now contributing more to economic growth than their legendary shopping habits. Move over, retail therapy—we’ve got data center therapy now.

OpenAI: The Poster Child of Optimistic Spending

If you want to see this financial high-wire act in action, look no further than OpenAI. The company behind ChatGPT is essentially the poster child for “fake it till you make it” in the AI world—except they’re being refreshingly honest about the “not making it yet” part.

OpenAI’s revenue in the first half of 2025 reached $4.3 billion, already surpassing its full-year 2024 revenue, but due to persistently high R&D expenditures, the company posted an operating loss of $7.8 billion.

Let that sink in: they made $4.3 billion and still somehow managed to lose $7.8 billion.

OpenAI CEO Sam Altman said the company should prioritize growth and its investments in training and compute “for a long time,” even if it delays its path to profitability. In other words, “We’ll worry about actually making money later. Much later. Like, 2029 later.”

The company expects to burn $8 billion in cash in 2025 on compute and other costs. That’s roughly the budget of a small space program, except instead of reaching for the stars, they’re reaching for... better chatbots?

The Revenue Reality Check

Here’s where things get uncomfortable for AI evangelists. While everyone’s been busy building data centers the size of small towns, the actual money-making part has been, shall we say, challenging.

A June report from McKinsey found that eight out of ten companies implementing AI reported no substantial increases in profits, often due to the complexities of integrating the technology into existing workflows and measuring its return on investment.

So 80% of companies are basically saying, “Yeah, we spent all this money on AI and... crickets.”

The problem isn’t that AI doesn’t work—it’s that monetizing it is harder than teaching a dog to fetch. Services like ChatGPT are wildly popular, with millions of users, but converting all that usage into sustainable revenue that covers astronomical infrastructure costs? That’s the trillion-dollar question. Or in this case, the two-trillion-dollar question.

The Power Problem (Because Why Not Add More Problems?)

As if the money situation wasn’t entertaining enough, there’s also the small matter of, you know, physically powering all these AI ambitions. Analysis indicates a projected 156 gigawatts of AI-related data center capacity demand by 2030, with 125 incremental gigawatts added between 2025 and 2030.

Bain estimates that global incremental AI computing requirements could soar to 200 gigawatts by 2030, with the US accounting for half of that demand. For context, that’s enough power to run several million homes. We’re literally choosing between lighting up neighborhoods and lighting up neural networks.

It’s like deciding to host a massive party without checking if your house has enough electrical outlets—except the party is global, the guests are power-hungry AI models, and nobody’s quite sure who’s bringing the snacks (or paying the electricity bill).

The Silver Linings (Sort Of)

Before you completely despair for the future of AI, there are some potential bright spots—or at least interesting sideshows.

Quantum Computing: Bain predicts that quantum computing could unlock $250 billion in market value across finance, pharmaceuticals, logistics, and materials science. Of course, quantum computing has been “just around the corner” for about as long as flying cars, so take that with a grain of quantum salt.

AI Agents: Companies are expected to allocate up to 10% of tech spending to building autonomous AI agents over the next three to five years. These are AI systems that can perform multi-step tasks with limited human guidance—basically, digital assistants that might actually be useful.

Humanoid Robots: Yes, those too. Though Bain notes that deployments are at an early stage and “heavily rely on human supervision,” which is code for “they’re expensive, somewhat clumsy, and need babysitting.” But hey, at least they look cool in promotional videos.

What This All Mean

The Bain report essentially sounds a warning bell that the AI industry might be building a house of cards—or more accurately, a data center of cards. The gap between what companies are spending and what they’re earning isn’t just a crack in the foundation; it’s a canyon.

This doesn’t necessarily mean AI is doomed or that it’s all been a massive waste. AI technology is genuinely transformative and will likely reshape countless industries. The question is whether the current business model—spend astronomical sums now, figure out profitability later—is sustainable.

Think of it this way: The AI industry is like someone who bought a Ferrari before getting their driver’s license, a mortgage before getting a job, and a lifetime gym membership before ever working out. The items themselves aren’t bad; it’s the sequence and the math that’s questionable.

The Road Ahead

Bain didn’t explicitly spell out what happens if AI companies continue to fall short of profitability through 2030, but reading between the lines isn’t difficult.

Without sustainable revenue models, we could see:

• A significant market correction (tech speak for “prices coming back down to Earth”)

• Consolidation, with smaller AI companies being absorbed by tech giants

• A shift from “growth at all costs” to “maybe we should actually try to make money”

• Infrastructure projects being scaled back or delayed

Or, in the more optimistic scenario, AI companies could finally crack the code on monetization, finding ways to generate revenue that actually justifies their spending.

Maybe ChatGPT Plus subscriptions will explode. Maybe enterprise AI will finally deliver ROI. Maybe quantum computing will solve everything. (Okay, probably not that last one, but we can dream.)

The Bottom Line

The AI industry finds itself in an interesting paradox: it’s simultaneously the most hyped and most unprofitable sector in tech. Companies are spending like there’s no tomorrow while hoping that tomorrow brings enough revenue to justify today’s bills.

Everyone is watching to see who blinks first. Will the tech giants keep pouring billions into AI infrastructure based on faith and FOMO? Will AI companies finally figure out how to turn impressive technology into impressive profits? Will someone, anyone, please check if we have enough electrical outlets for all these data centers?

One thing’s certain: the next few years will be fascinating to watch. The AI industry is writing one of the most expensive business case studies in history, and unlike GPT-4, nobody knows how this story ends.

Let’s just hope the ending isn’t “and then they all ran out of money.”

That would be a rather anticlimactic conclusion to humanity’s great AI adventure. Though admittedly, it would be kind of funny in a “told you so” sort of way.

Stay tuned, folks. This party’s just getting started—someone just needs to figure out who’s paying the tab.

About the author: Rupesh Bhambwani is a technology enthusiast specializing in the broad technology industry dynamics and international technology policy.

When not obsessing over nanometer-scale transistors, energy requirements of AI models, real-world impacts of the AI revolution and staring at the stars, he can be found trying to explain to his relatives why their smartphones are actually miracles of modern engineering, usually to limited success.

Leave a comment

Remember when everyone thought AI meant massive, cloud-hungry behemoths that could answer any question about the universe?

Well, the AI world is having a bit of a rethink.

Just like the tech industry learned that not every phone needs to be a flagship model, we're discovering that not every AI task needs a "God-like" large language model (LLM).

When tech folk talk about the lacklustre progress of LLMs, they often draw an analogy with smartphones. The early days of OpenAI's ChatGPT were as revolutionary as the launch of Apple's iPhone in 2007. But advances on the frontier of AI have started to look like ho-hum phone upgrades rather than genuine breakthroughs.

This comparison hits home because it's so relatable. We've all experienced that moment of mild disappointment when the "revolutionary" new phone turns out to be incrementally better than last year's model. The same thing is happening with AI right now, and it's pushing businesses and researchers toward a more practical approach.

What Exactly Are Small Language Models?

Think of language models like engines. There is no precise definition of what constitutes small versus large language models. The distinction comes down to the number of parameters they are trained on—ie, the amount of numerical settings in a model's brain that help it make sense of data. LLMs stretch to hundreds of billions of parameters. SLMs may be trained on 40bn or fewer, down to under 1bn in the case of tiny ones.

To put this in perspective, that's like comparing a massive cargo ship engine to a motorcycle engine. Both can get you where you need to go, but one is clearly more suited for specific tasks than the other.

Smaller models are more efficient, making them quicker to train and run. That's good news for anyone wanting a more affordable on-ramp. Also smaller models work with a fraction of the computational resources.

Why Companies Are Switching

Here's where things get interesting for businesses. As David Cox, head of research on AI models at IBM, a tech company, puts it: "Your HR chatbot doesn't need to know advanced physics." This simple statement captures why so many companies are reconsidering their AI strategies.

The economics are compelling. Companies have moved on from a spend-whatever-it-takes approach, employed in the early days of generative AI, to a greater focus on return on investment. One venture capital executive perfectly summarized this shift: you may need a Boeing 777 to fly from San Francisco to Beijing, but not from San Francisco to Los Angeles.

IBM's Docling product is a perfect real-world example. IBM's Cox points to a product developed by his company, called Docling, which transforms PDF documents, such as receipts, into storable data. It runs on a "tiny" model with about 250m parameters. He calls it a useful tool, but one that would not be cost-efficient if it were run on an LLM.

Recent research supports this trend. This year, corporate demand for them is projected to grow twice as fast as it is for LLMs, though from a much lower base. Companies are discovering that specialized, smaller models often outperform their massive counterparts on specific tasks.

How Small Models Compete

The improvement in small language models has been nothing short of remarkable. Small models are increasingly "taught" by larger ones, rather than having to crawl the web to learn for themselves. This teacher-student approach has led to some surprising results.

Artificial Analysis, a benchmarking firm, says that on a variety of tests, a 9bn-parameter model called Nvidia Nemotron Nano outperforms a Llama model, which is 40 times bigger. Let that sink in – a model 40 times smaller outperforming its massive competitor.

Recent comprehensive surveys from late 2024 and early 2025 have documented significant advances in small language model capabilities, showing how these models are catching up to their larger counterparts through improved training techniques and architectural innovations.

Beyond the Cloud

One of the most significant advantages of small language models is their hardware flexibility. Small models can also run on cheaper varieties of chip. The smallest models can use central processing units (CPUs), the workhorses of general computing, rather than graphics processing units (GPUs) like those that have made Nvidia the world's most valuable company.

This hardware independence is crucial as over 15 billion edge devices are currently deployed worldwide, and the entire edge computing market is expected to grow at a CAGR of 18.3% through 2032.

By 2025, Edge AI will be responsible for half of all enterprise data created, according to an estimate by Gartner Inc. This shift toward edge computing makes small language models not just attractive, but essential.

The Future of AI Agents

Perhaps the most compelling argument for small language models comes from their potential in AI agents. A little-noticed paper published in June by Nvidia Research, the chipmaker's investigations arm, states boldly that "small, rather than large, language models are the future of agentic AI".

The reasoning is both practical and economic. It notes that currently most agents are powered by LLMs, hosted by cloud service providers. The investment pouring into AI-related cloud infrastructure suggests that the market assumes that LLMs will remain the engines of agentic AI.

The paper challenges that assumption, arguing that SLMs are sufficiently powerful to handle agentic tasks, and more economical (for instance, a 7bn-parameter model can be 10 to 30 times cheaper to run than a model up to 25 times bigger).

According to a recent report from McKinsey, these intelligent agents are set to transform complex problem-solving in sectors like manufacturing and logistics, enabling real-time adaptations that boost efficiency by up to 30%.

The vision they propose is fascinating: It says that SLMs may lead to a "Lego-like" approach to building agents, with firms using small, specialised experts, rather than one monolithic LLM intelligence.

Apple’s Patience Pays Off?

Apple's approach to AI has been widely criticized as too cautious, but it might turn out to be prescient. Apple disappointed investors when it launched "Apple Intelligence", its AI offering, last year, because it did not work well. Its shares slid on September 9th after the launch of the iPhone 17, partly because there was no news of progress on Apple Intelligence.

However, their strategy of using small models for on-device tasks while delegating complex operations to the cloud could be the winning approach. But the firm's approach of using SLMs to do some tasks on the iPhone while delegating harder jobs to the cloud could be the shape of things to come, says Counterpoint's Agrawal. Such is Apple's clout, even if it has "missed the boat" in the early stages of the AI race, "it can always get on the next boat", he says.

Apple's machine learning research has explored novel architectures like projected networks (PN), which allow high capacity networks whose parameters can be linearly projected into smaller ones, enabling specialized accuracy even when specialization data is unknown during pretraining.

Beyond performance and convenience, small language models offer significant environmental advantages. Unlike traditional AI models, which often require cloud-based processing, Edge AI handles computation locally. This reduces the need to transfer large volumes of data to centralized locations, cutting down on latency and network bandwidth usage.

This local processing capability is becoming increasingly important as a market research report anticipated a 33.9% compound annual growth rate for edge AI between 2024 and 2030.

What This Means for the Future

The shift toward small language models doesn't mean large models will disappear. Whether SLMs eventually displace LLMs or not, heterogeneity is on the rise.

All-knowing LLMs will remain important for consumer apps like OpenAI's ChatGPT. Yet even OpenAI is shifting its focus. GPT-5 has internal models of different sizes and strengths, which it uses depending on the complexity of the task.

This represents a maturation of the AI industry. Instead of the "one-size-fits-all" approach of early large language models, we're moving toward a more nuanced ecosystem where different tools are used for different jobs.

The Bottom Line

The rise of small language models represents something bigger than just a technical trend – it's a sign that the AI industry is growing up. We're moving past the "bigger is better" mentality toward a more practical, economical, and environmentally conscious approach to artificial intelligence.

Given the relative merits of SLMs, Apple's take-it-slow approach may turn out to be justified in the long run. Sometimes, the tortoise really does beat the hare.

For businesses, this shift opens up new possibilities for deploying AI without massive infrastructure investments. For consumers, it means faster, more responsive AI experiences on their devices.

The future of AI isn't just about building the biggest, most powerful models – it's about building the right model for the right job. And increasingly, that means thinking small.

About the author: Rupesh Bhambwani is a technology enthusiast specializing in the broad technology industry dynamics and international technology policy.

When not obsessing over nanometer-scale transistors, energy requirements of AI models, real-world impacts of the AI revolution and staring at the stars, he can be found trying to explain to his relatives why their smartphones are actually miracles of modern engineering, usually to limited success.

Leave a comment

We're drowning in what experts politely call "AI slop" – that endless stream of artificially generated content that's flooding our feeds like digital spam. Think of it as the fast food of information:

cheap to produce,

addictive to consume, and

probably not great for our long-term health.

This isn't just about obviously fake videos of celebrities endorsing cryptocurrency (though those are multiplying faster than rabbits). We're talking about AI-generated news articles, synthetic social media personalities, and chatbots so convincing they could probably get elected to public office. In fact, some probably already have.

The problem isn't that AI creates fake content – humans have been doing that since cave paintings. The problem is the sheer scale.

We've gone from occasional propaganda to a fire hose of synthetic reality, and our brains simply weren't designed for this information apocalypse.

The Hallucination Highway

AI systems have a charming tendency to "hallucinate" – a technical term meaning they confidently make stuff up, like that friend who insists they met a celebrity at Starbucks but can never quite remember which one.

These hallucinations aren't malicious; they're just a byproduct of how AI works. But when these confident fabrications get mixed into news feeds, research papers, and casual conversations, we end up with a culture where "I read it online" carries about as much weight as "my cousin's friend told me."

The result?

We're developing what psychologists are starting to call "epistemic fatigue" – basically, we're getting tired of trying to figure out what's true. It's like being a fact-checker in a world where everyone speaks in riddles, and the riddles are generated by machines that learned language from the entire internet (which explains a lot, actually).

Confusing, right?

Now you get the point

How We Traded Villages for Algorithms

For millennia, humans survived and thrived through one simple principle: stick together. We lived in tribes, villages, and communities where everyone knew everyone, gossip traveled at walking speed, and your reputation actually mattered because you'd see the same faces for decades.

Fast forward to today, and we've somehow convinced ourselves that 847 LinkedIn connections constitute a meaningful social network. We've traded deep relationships for broad reach, intimate conversations for public broadcasts, and genuine community for algorithmic engagement.

The Social Media Paradox

Social media promised to connect us all, and technically it succeeded. We're more connected than ever – to our phones. The average person now checks their device 96 times per day, which means we're basically in a committed relationship with a glowing rectangle that knows more about our habits than our actual family members.

We've become digital nomads wandering through virtual landscapes, collecting likes instead of memories, building follower counts instead of friendships. It's like we've replaced actual human connection with a sophisticated points-based game, and somehow convinced ourselves this is progress.

The irony is thick: we have instant access to billions of people, yet loneliness rates are skyrocketing. We can video chat with someone on the other side of the planet, but we don't know our next-door neighbor's name. We're hyper-connected and simultaneously isolated, like prisoners in adjacent cells who can only communicate through tapping codes.

The Algorithm Knows Best (Or So It Claims)

Perhaps the most insidious change is how we've outsourced our decision-making to algorithms. We let AI decide what news we see, what products we buy, whom we date, and what music we hear.

We've essentially hired a digital butler that never sleeps, always watches, and has opinions about everything – except this butler was trained by every conspiracy theorist, marketing executive, and keyboard warrior on the internet.

These recommendation systems create what researchers call "filter bubbles" – personalized reality zones where everyone agrees with you, confirms your biases, and never challenges your assumptions.

It's like living in a house of mirrors, except the mirrors are programmed to show you exactly what you want to see, making you feel smart and right about everything. Sounds nice, until you realize you're slowly losing touch with actual reality.

Rediscovering Our Analog Selves

The good news is that we're not helpless victims in this technological tsunami. We still have agency, even if we've temporarily forgotten how to use it.

The solution isn't to become digital hermits (though the idea has its appeal), but to be more intentional about how we engage with technology.

Individual Actions: Becoming a Human Again

Create Tech-Free Zones: Designate sacred spaces where phones fear to tread. Start with the bedroom – your sleep doesn't need to be interrupted by notifications about discount mattresses. Then expand to the dinner table, where actual humans can practice the ancient art of conversation.

Practice Digital Skepticism: Before sharing that shocking news article, ask yourself: "Does this seem too outrageous to be true?" If yes, it probably is. Develop a healthy paranoia about online content. Not everything needs to be fact-checked, but your brain deserves better than a steady diet of clickbait and conspiracy theories.

Cultivate Real Relationships: Remember those things called "friends"? The ones you can touch without swiping? Invest in offline relationships like they're going out of style (because they kind of are). Schedule regular phone-free meetups, join clubs based on actual interests, volunteer for causes you care about.

Learn a Skill That Requires Patience: Take up gardening, knitting, cooking, woodworking – anything that can't be optimized by an algorithm and requires you to fail repeatedly before succeeding. These activities reconnect us with the satisfying reality of gradual improvement and tangible results.

Community Solutions: Building Villages in a Digital Age

Support Local Everything: Local newspapers, bookstores, cafes, farmers markets – these are the antibodies against digital monoculture. They provide spaces for serendipitous encounters and conversations that algorithms can't predict or optimize.

Create Gathering Spaces: Communities need physical spaces where people can meet without screens mediating the interaction. This might mean converting unused buildings into community centers, creating tool libraries, or organizing regular neighborhood events that don't require an app to attend.

Promote Digital Literacy: We need to treat AI literacy like we treat driver's education – essential knowledge for navigating a potentially dangerous environment. Schools and communities should teach people how to identify AI-generated content, understand algorithmic bias, and maintain healthy skepticism about online information.

Establish Local Information Networks: Support and create local news sources that prioritize accuracy over engagement. These outlets serve as important counterbalances to the algorithmic echo chambers that dominate social media.

The Simple Life: Not Simple, But Worth It

The path back to authentic human connection isn't about rejecting all technology – it's about using it intentionally rather than being used by it.

This means choosing quality over quantity in our digital diet, prioritizing depth over breadth in our relationships, and remembering that the most important conversations happen when we put down our phones and look each other in the eye.

The simple life isn't actually simple – it requires constant vigilance against the gravitational pull of digital distraction. But it offers something that no algorithm can provide: the messy, unpredictable, genuinely human experience of being present in the world.

In the end, the greatest act of rebellion against our algorithmic ruler might just be having a long, meandering conversation with a friend over coffee, where nobody's tracking the engagement metrics and the only recommendation engine is your own curiosity about what the other person might say next.

After all, we managed to build civilization for thousands of years without push notifications. We might just be able to do it again.

Now if you'll excuse me, I need to go check my notifications. This article isn't going to promote itself.

About the author: Rupesh Bhambwani is a technology enthusiast specializing in the broad technology industry dynamics and international technology policy.

When not obsessing over nanometer-scale transistors, energy requirements of AI models, real-world impacts of the AI revolution and staring at the stars, he can be found trying to explain to his relatives why their smartphones are actually miracles of modern engineering, usually to limited success.

Leave a comment