Nvidia's share of China's AI chip market has collapsed from 95% to effectively zero, and by the time Washington reversed course, Chinese hyperscalers had already moved on.
Jensen Huang doesn't mince words. "In China, we have now dropped to zero," the Nvidia CEO said recently, describing the company's current share of China's AI accelerator market. Not shrinking. Not declining. Zero. For a company that held 95% of that market just two years ago, it's a remarkable statement, and the reason for it sits squarely in Washington's policy record.
The mechanics are straightforward enough. US export controls progressively restricted what Nvidia could sell into China: first the A100, then the H100, then the H20, a chip specifically downgraded to meet compliance thresholds. Each restriction pushed Chinese hyperscalers, including Alibaba, ByteDance, and Tencent, to accelerate their pivot toward Huawei's Ascend line and other domestic alternatives. By the time the Trump administration reversed the H200 export ban in January 2026, approving sales under a 25% tariff and case-by-case review, the damage was done. Customers had already retooled their infrastructure. Huawei had already locked in the orders.
Bernstein analysts had seen this coming. They projected Nvidia's China market share falling from roughly 40% in 2025 to around 8% in 2026, while Huawei's climbs toward 50%. Huawei expects its AI chip revenue to reach approximately $12 billion this year, up from $7.5 billion in 2025, a gain built almost entirely on demand from the same customers Nvidia used to serve. Huawei plans to ship around 600,000 Ascend 910C chips in 2026 regardless of what Washington decides, and Chinese domestic chipmakers collectively, including Cambricon, Moore Threads, and MetaX, are approaching 80% of the market that Nvidia once owned outright.
Huang has argued publicly that the controls have "already largely backfired." He told the Special Competitive Studies Project that "conceding an entire market the size of China probably does not make a lot of strategic sense." That framing is self-serving, obviously, but it's also accurate. The logic behind the restrictions was to deny China access to frontier compute and slow its AI development. What happened instead is that Beijing accelerated its domestic chip push, Huawei got a captive market it might never have fully captured otherwise, and Nvidia absorbed a $4.5 billion charge in Q1 fiscal 2026 tied to H20 inventory and purchase obligations it could no longer fulfill.
The January reversal on H200 exports illustrates the timing problem precisely. Chinese firms placed orders for over two million H200 chips once licenses became available. The demand is real. But the procurement cycles, software stack commitments, and supply chain decisions that Chinese hyperscalers made during the restriction period don't simply unwind because a policy changed. ByteDance and Alibaba spent the better part of two years adapting their training infrastructure to Huawei's Ascend ecosystem. You don't undo that with a tariff reversal.
Frontier model training in China still leans on Nvidia hardware where it can be obtained, so the story isn't entirely binary. The H20 chips that were sold before export restrictions tightened are still running in Chinese data centers. But the trajectory is not ambiguous. Domestic alternatives that were once genuinely inferior to Nvidia's products have narrowed the gap through forced investment and a massive captive demand base. Huawei's Ascend 910C is now competitive enough that TechRadar's reporting suggests the chip's performance trajectory was itself a factor in triggering the H200 policy reversal, Washington apparently deciding that a partial re-entry was preferable to ceding the market entirely.
For Nvidia investors, China had been generating somewhere between 20% and 25% of the company's data center revenue before the heaviest restrictions hit. That pool of revenue doesn't disappear overnight, but it's structurally smaller now than it was, and the competitive position that made Nvidia's China business so lucrative, near-total market dominance, is gone and won't return at anything close to the same scale. The $4.5 billion inventory charge gives a rough sense of how abruptly that transition happened.
The broader policy question is harder. Export controls exist for legitimate national security reasons, and the US-China technology competition is real. But the China chip case suggests that unilateral restrictions without allied coordination tend to create market vacuums that domestic competitors fill, rather than capability gaps that persist. As Tom's Hardware's reporting noted, Chinese fabs "can barely keep up" with Huawei's order volume right now. That is not a sign of a technology sector being successfully contained. It's a sign of one being accelerated.
Whether the same dynamic plays out for other chipmakers depends on how Washington handles the next round of restrictions. Intel, AMD, and Qualcomm all have China exposure. The Nvidia case is now the clearest data point available on what happens when export controls move faster than the policy's underlying theory can absorb: you hand the incumbent's customers a reason to build alternatives, you give those alternatives time to mature, and by the time you reconsider, the switching cost has already been paid.
Also read: Germany is betting €300 billion on AI to solve a labor crisis that AI cannot fully fix • Sovereign wealth funds are betting that AI's real money is in the wires and the watts, not the models • The AI buildout has turned America's power utilities into the most contested assets on Wall Street