How China Has Successfully Hacked the Semiconductor Embargo
Fresh & Hot curated AI happenings in one snack. Never miss a byte 🍔
This snack byte will take approx 4 minutes to consume.
Giddy traders in Shanghai joke that Cambricon—a Chinese chip firm—isn't just offering substitutes for Nvidia's processors, but for its stock as well. While Cambricon remains a mere fraction of Nvidia's market cap, its share price has skyrocketed by a jaw-dropping 350% over the past year—approximately 15 times Nvidia's own impressive growth.
I've witnessed countless predictions about China's technological rise. Most have proved premature. But this time feels different. Despite America's increasingly stringent tech sanctions, Chinese AI continues its relentless march forward.
How Taiwan Became the Chip Capital of the World
To understand today's semiconductor geopolitics, we need to rewind to the 1980s. While Silicon Valley pioneered chip design, manufacturing gradually shifted overseas as American firms embraced the "fabless" model—designing chips but outsourcing production.
Taiwan seized this opportunity with remarkable foresight. The government established the Industrial Technology Research Institute (ITRI) in 1973, creating a foundation for semiconductor knowledge transfer. In 1987, Morris Chang, a Texas Instruments veteran, founded Taiwan Semiconductor Manufacturing Company (TSMC) with government backing.
Chang's "pure-play foundry" model—manufacturing chips designed by others without competing with customers—revolutionized the industry. While Intel and other integrated device manufacturers (IDMs) maintained their own fabrication plants, TSMC's specialization allowed it to achieve economies of scale and accelerate down the learning curve faster than anyone else.
By the early 2000s, TSMC had established itself as the world's premier chip manufacturer. Today, it produces roughly 90% of the world's most advanced semiconductors. Intel, once the undisputed leader, now trails TSMC by at least two process generations.
As one industry veteran quipped to me last year, "Taiwan doesn't have oil, but it struck silicon gold."
The U.S. government's approach to China's semiconductor ambitions has evolved from cautious engagement to aggressive containment.
The watershed moment came in October 2022, when the Biden administration unveiled sweeping export controls targeting China's semiconductor industry. These restrictions went beyond previous measures by:
Blocking sales of advanced AI chips (including Nvidia's A100 and H100)
Restricting equipment used to manufacture logic chips at 14nm and below
Preventing U.S. personnel from supporting advanced Chinese fabs
The restrictions only intensified throughout 2023 and 2024. Last month, Nvidia was effectively barred from selling even its China-specific H20 chip, which it had developed specifically to comply with earlier regulations. The Trump administration recently announced plans to rescind a complicated export-licensing regime scheduled for May 15th, likely replacing it with even stricter bilateral measures.
Despite these unprecedented sanctions, Chinese AI development continues at a blistering pace. Far from being crippled, internet giants like Alibaba and Tencent are constructing massive AI data centers as if unconstrained by hardware limitations.
This resilience stems from several factors:
First, algorithmic innovations by companies like DeepSeek have significantly reduced the computing requirements for training large AI models. When you can't get more hardware, you make your software smarter.
Second, a shadow supply chain continues funneling advanced Nvidia chips into China through various gray-market channels. The details of these operations remain obscure, but industry insiders describe elaborate networks involving multiple countries and shell companies.
But most impressively—and concerning for U.S. policymakers—China's domestic chip industry is achieving breakthroughs that seemed fantastical just three years ago.
The Dragon's New Silicon Teeth
Huawei has emerged as the standard-bearer for China's semiconductor independence movement. Last month, it began shipping its CloudMatrix chip cluster, which interconnects 384 of its Ascend AI chips using sophisticated networking technology. According to multiple industry reports, this system outperforms Nvidia's popular NVL72 cluster in raw computational throughput, albeit with higher power consumption. Not exactly a green solution in today's climate-conscious world, but in the chip race, you take your wins where you can get them!
During an online presentation I recently attended, a Huawei representative couldn't resist a bit of nationalistic chest-thumping: "Western sanctions forced us to develop what we otherwise might have purchased. In that sense, we should thank them."
Huawei isn't alone in this renaissance. A growing ecosystem of Chinese chip designers has emerged with processors intended to replace the Nvidia A100. Cambricon is reportedly already delivering its substitute to customers, while Hygon has completed testing and is expected to begin shipments within months.
Even in high-bandwidth memory (HBM)—crucial components for AI processors traditionally dominated by South Korean giants SK Hynix and Samsung, along with America's Micron—China is making remarkable progress. CXMT, based in Hefei, is rapidly closing the technology gap despite December's U.S. restrictions on HBM sales to China.
Perhaps most impressive are the advances in semiconductor manufacturing equipment—the highly specialized tools required to produce chips. AMEC recently unveiled an etching tool for NAND memory chips, breaking Lam Research's long-standing monopoly in this niche. Meanwhile, Naura has developed techniques for layering silicon-germanium onto chip wafers, challenging American giant Applied Materials.
As Bernstein analyst Lin Qingyuan noted, "These types of developments may come faster than most imagined."
For all this progress, significant hurdles remain. Most customers purchasing domestically produced chips are state-owned enterprises rather than commercial companies with strict performance requirements. Huawei's latest Ascend 910C processor still relies heavily on foreign components according to analyses by SemiAnalysis.
China also lacks domestic alternatives to the extreme ultraviolet (EUV) lithography tools produced exclusively by ASML of the Netherlands. These machines—each costing approximately $150 million and requiring 40 shipping containers to transport—are absolutely essential for manufacturing chips at 7nm and below. Without them, China remains structurally unable to produce truly cutting-edge semiconductors at scale.
This leaves Chinese chip designers dependent on SMIC, a state-owned foundry that trails TSMC by several process generations. Though SemiAnalysis has reported that Huawei continues circumventing sanctions by purchasing TSMC-manufactured wafers through intermediaries—claims both companies vehemently deny.
The software situation presents another critical challenge. Nvidia's CUDA platform remains the gold standard for programming AI chips, creating what economists call a "moat" around their business. Nearly all AI developers learn CUDA, and it works exclusively with Nvidia hardware. The network effects are enormous—imagine trying to create a new social media platform when all your friends are on an existing one!
Huawei has developed CANN as a CUDA alternative for its Ascend chips, but by all accounts, it remains years behind Nvidia's mature ecosystem and suffers from persistent bugs.
What's changed most dramatically isn't just the technology but the mindset. Before American sanctions, Chinese companies regularly complained about government pressure to use domestic chips, citing higher costs and lower reliability. Now, these same firms increasingly view technological self-sufficiency as existential.
The consensus amongst several executives from several Chinese tech firms in clear: they would gladly pay premium prices and tolerate initial quality issues to support domestic suppliers.
This shift in attitude might prove more consequential than any individual technological breakthrough. History suggests that with sufficient determination, funding, and talent, technological gaps eventually close. The question isn't if China can catch up, but when—and at what cost.
Looking Ahead
The semiconductor industry has always operated on boom-and-bust cycles, but the current geopolitical fragmentation adds new dimensions of uncertainty. American policymakers bet that China's technological progress could be significantly delayed through export controls. That bet now looks increasingly risky.
At a recent conference, I found myself seated next to a veteran American semiconductor executive now consulting for Chinese firms. "Five years ago, I'd have said China was two decades behind. Three years ago, maybe one decade. Now? Perhaps just 5-7 years in most areas," he mused. "The pace of catch-up is accelerating, not slowing."
For companies and investors, this rapidly evolving landscape demands constant reassessment. The $500+ billion semiconductor market is fragmenting along geopolitical lines, creating parallel ecosystems with different standards, suppliers, and innovation trajectories.
In the semiconductor world, we often talk about Moore's Law doubling transistor density roughly every two years. Perhaps we need a new law to describe the pace of China's semiconductor independence—call it the Dragon Curve, where the gap halves with each passing year.
One thing is certain: the next chapter of the global semiconductor saga will be written in multiple languages.
And while Silicon Valley pioneered this industry, its future increasingly depends on decisions made in Washington, Beijing, and Taipei.
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.