The ban had forced their hand, but it had also given them permission
When governments draw lines around technology, they sometimes draw the map their rivals needed most. The U.S. decision in September 2022 to ban Nvidia and AMD's most powerful chips from Chinese markets did not simply deny China access to American silicon — it clarified the destination and accelerated the journey. Chinese startups like Shanghai Biren, already quietly building domestic alternatives, suddenly inherited a captive market and a geopolitical mandate, raising the deeper question of whether restriction, in a globally entangled economy, can ever be a clean instrument of containment.
- Washington's export ban on advanced AI chips, intended to slow China's military computing ambitions, instead handed Chinese startups a guaranteed customer base overnight.
- Biren's BR100 processor emerged with benchmark performance rivaling the very American chips now forbidden — a timing so precise it reframed the ban as a market signal rather than a setback.
- The real battleground is not transistor speed but software: Nvidia's CUDA ecosystem has years of developer loyalty baked in, and no Chinese chip can win without a compelling alternative platform.
- American venture capital — including Sequoia China and Lightspeed — has poured $2.5 billion into the Chinese AI chip sector, creating an uncomfortable loop where U.S. money funds the alternatives U.S. policy is trying to prevent.
- The trajectory points toward a bifurcated global AI infrastructure, with Chinese firms racing to prove not just performance but reliability, iteration, and the developer trust that turns hardware into an ecosystem.
When the U.S. government banned Nvidia and AMD from selling their most powerful AI chips to China in early September 2022, it produced an unintended consequence: Chinese startups that had been quietly building domestic alternatives suddenly found themselves with a captive market. Companies locked out of American technology needed computing power, and firms like Shanghai Biren Intelligent Technology were ready.
Founded by alumni of Alibaba and Nvidia, Biren had just released its BR100 processor. Independent benchmarks showed it could match the performance of the very American chips now forbidden to Chinese buyers. The timing was striking. Alongside competitors like Cambricon, Moore Threads, and Alibaba's PingTouGe, Biren had collectively raised $2.5 billion in recent years — from Chinese government-backed funds and, notably, American venture capital firms including Sequoia China and Lightspeed China Partners. The ban transformed that long-term bet into immediate opportunity.
But raw speed was never the whole contest. Nvidia's deeper advantage was CUDA, a software platform so thoroughly embedded in AI development that switching to a new chip meant rewriting code, retraining engineers, and rebuilding workflows accumulated over years. That software moat was the real barrier Chinese firms faced — and the real measure of whether their chips could become indispensable rather than merely available.
History offered a precedent. When the U.S. embargoed Intel chips from certain Chinese research institutions decades earlier, China responded by designing its own processors for supercomputers. Computer scientist Jack Dongarra, who oversees global supercomputer rankings, noted the pattern plainly: constraint had produced capability before. Analysts like Wayne Lam of CCS Insight saw Biren as a genuine potential success story, even as skeptics insisted that reliability, hardware iteration, and developer ecosystems mattered as much as benchmark numbers.
The episode also surfaced an uncomfortable irony about American capital. Defense-aligned investors warned that U.S. venture money was helping build the very AI chip alternatives that U.S. policy sought to prevent — a loop that no export control had yet addressed. What the ban had clarified, above all, was that technological competition does not move in straight lines: the restriction designed to slow China's progress had instead sharpened its focus, justified its investment, and handed it the customers it needed to grow.
When the U.S. government banned Nvidia and AMD from selling their most powerful artificial intelligence chips to China, it did something counterintuitive: it handed opportunity to the very companies Washington was trying to contain. Chinese startups that had been quietly building alternatives to American silicon suddenly found themselves with a captive market of customers desperate for computing power and locked out of the West.
The export restrictions, announced in early September 2022, were part of a broader effort to slow China's development of advanced computing capabilities that could feed military applications—nuclear weapons design, weapons systems, surveillance. But the ban created an opening that Chinese chip makers had been preparing for. Shanghai Biren Intelligent Technology, a startup founded by alumni of Alibaba and Nvidia, had just released a new processor called the BR100. When an independent benchmarking group published performance data, Biren's chip showed it could match the speed of the very American chips now forbidden to Chinese customers.
The timing was almost too perfect. Biren and competitors like Cambricon, Alibaba's PingTouGe, and Moore Threads had been raising capital for years—$2.5 billion in recent years alone—from a mix of Chinese government-backed funds and American venture capital firms. Sequoia China, Lightspeed China Partners, and other Silicon Valley investors had been betting on the idea that China would eventually need homegrown alternatives. Now that bet was paying off. Chinese data centers and cloud companies that had relied on Nvidia's dominance suddenly had a domestic option, and Biren was positioned to capture that business.
But speed alone does not win markets. Nvidia's real power lay not just in fast chips but in CUDA, its software platform that had become the industry standard for artificial intelligence work. Every researcher, every engineer, every company building AI systems had learned to write code for CUDA. Switching to a Chinese chip meant rewriting software, retraining teams, rebuilding workflows. That was the real barrier Biren and its competitors faced. A chip could be fast. A software ecosystem took years to build and required the trust of thousands of developers.
Jack Dongarra, a computer science professor who oversees the world's ranking of fastest supercomputers, had seen this pattern before. When the U.S. had embargoed Intel chips to certain Chinese research institutions decades earlier, China responded by designing its own processors for its supercomputers. History suggested it could happen again. "The result was that China designed its chips for its supercomputers," Dongarra said. This time, the stakes were higher and the market was larger. Chinese computing groups would have to retool their systems, rebuild their infrastructure, and learn to work with new hardware. But they would do it.
Analysts saw genuine potential. Wayne Lam, a chip analyst at CCS Insight, called Biren a possible "success story." Chinese computing groups would "have to re-tinker their systems and figure out how to build for something that they can get," he said. The ban had forced their hand, but it had also given them permission to invest in alternatives they might otherwise have delayed.
Still, skeptics remained. Some American chip executives and analysts argued that Chinese startups would need to prove more than raw performance. They would need to demonstrate reliability, show they could iterate and improve hardware at the cutting edge, and most critically, offer a software ecosystem compelling enough to pull developers away from Nvidia's entrenched platform. Paul Triolo, a China strategist at Albright Stonebridge Group, laid out the challenge plainly: Chinese firms had to become not just fast but trustworthy, not just competitive but essential.
The ban had also raised uncomfortable questions about American capital. Some venture investors and defense-focused entrepreneurs worried that U.S. venture capital firms were funding Chinese AI chip development that could eventually serve military purposes. Matt Ocko, managing partner of DCVC, a firm with deep ties to U.S. defense and intelligence work, said it was "not acceptable for major pools of U.S. capital to fund AI chips and other PRC military tech that threatens U.S. national security." The irony was sharp: American investors had helped build the very alternatives the American government was trying to prevent.
What emerged was a picture of technological competition that did not move in straight lines. Restrictions meant to slow China's progress had instead accelerated it by creating market demand and justifying investment. Chinese startups had the capital, the talent, and now the customers. What remained to be seen was whether they could build the software layer that would make their chips indispensable—and whether the U.S. would eventually restrict its own capital from flowing to them.
Citações Notáveis
The result was that China designed its chips for its supercomputers— Jack Dongarra, computer science professor and Top500 supercomputer ranking leader, on historical precedent
It's not acceptable for major pools of U.S. capital to fund AI chips and other PRC military tech that threatens U.S. national security— Matt Ocko, managing partner of DCVC venture capital firm
A Conversa do Hearth Outra perspectiva sobre a história
When the U.S. banned these chips, did anyone expect it would actually help Chinese companies?
The ban was designed to slow China down, but it created the opposite effect. Chinese startups had been building alternatives for years. Suddenly they had a huge customer base that couldn't buy American chips anymore.
So Biren just happened to release a competitive chip at exactly the right moment?
Not exactly luck. They'd been funded heavily for years—by Chinese government funds and even American venture capital. The timing worked out because the infrastructure was already there. The ban just flipped a switch.
But a fast chip isn't enough, right? Nvidia's real power is something else?
CUDA—their software platform. Every AI engineer in the world learned to code for it. Switching to Biren means rewriting everything. That's the real moat Nvidia built, and it's much harder to replicate than raw speed.
Has this happened before in tech history?
Yes. When the U.S. embargoed Intel chips to Chinese supercomputing centers decades ago, China just built its own processors. The embargo forced them to become self-sufficient. This could follow the same pattern.
What about the American venture capital firms funding these Chinese startups? Isn't that a problem?
That's the uncomfortable part. U.S. investors have been betting on Chinese alternatives for years, partly because they saw the market opportunity. Now some are worried they've helped create technology that could serve military purposes.
So what actually determines whether Biren wins?
They need three things: a fast chip—they have that. Reliable hardware that improves over time—they need to prove it. And a software ecosystem that makes developers want to switch from CUDA. That last part is the hardest.