Why Jensen Huang says Nvidia went from 95% to 0% in China’s AI chip market

What 0% does and does not mean

The safest reading is not that every Nvidia GPU in China has vanished or stopped working. It is that Nvidia’s visible, compliant, new high-end AI-chip business in China has been reduced to essentially zero under the current policy environment.

That distinction matters. Reports citing supply-chain information from Caijing said Nvidia shipped roughly 600,000 to 800,000 H20-series chips in China in 2024, while one major domestic AI chip shipped roughly 300,000 to 400,000 units; the same reports said Nvidia’s fiscal 2025 revenue from mainland China and Hong Kong was $17.1 billion, or 13.1% of company revenue. Those figures make it hard to read Huang’s line as a denial of historical sales or installed hardware. It is better understood as a forward-looking business and policy statement.

Huang reinforced that reading in November 2025, when he said U.S. export restrictions had stalled Nvidia’s chip sales to China and that he expected China sales to be zero for the next two quarters. He also said China’s AI-chip market was about $50 billion and could grow to $200 billion by the end of 2030.

How export controls boxed Nvidia in

The direct cause is the tightening of U.S. controls on advanced AI chips exported to China. After Washington began restricting high-end AI-chip exports in October 2022, Nvidia designed China-market versions such as the A800, H800 and H20 to comply with the rules. But the rules kept tightening. New restrictions in April 2025 forced the H20 to stop selling, according to reports, and left Nvidia with about $4.5 billion in inventory losses and roughly $8 billion in potential lost revenue.

That turned the China problem from a product problem into a rules problem. Even if Nvidia designs a chip that fits today’s limits, the next round of licensing rules or performance thresholds can change the market again. Reports said Nvidia wanted a China-specific version based on its Blackwell GPU to receive U.S. export-control approval, but whether that can happen depends on Washington’s licensing decisions.

Huang’s broader argument is not simply that Nvidia wants to sell more chips. He has argued that if the United States wants global AI to be built on American technology, that technology must be available and reliable enough for the world to depend on. Restricting Nvidia’s exports to China, he said, limits the chance for China’s large AI developer community to build on U.S. technology.

For Nvidia, the near-term issue is revenue. The long-term issue is access

China is not all of Nvidia’s business, but it is still material. Nvidia’s fiscal 2025 revenue from mainland China and Hong Kong was reported at $17.1 billion, equal to 13.1% of total revenue, even as the company told shareholders it was modeling China business at zero and treating any progress there as upside.

The deeper risk is ecosystem access. Nvidia’s advantage is not only the raw speed of a single GPU. Reports describe the company as building an integrated production system spanning chips, networking, servers, software and algorithms. If Chinese customers cannot reliably buy Nvidia AI chips, they have a strong reason to test alternative hardware, alternative software stacks and alternative supply chains.

That is the strategic warning inside the 95% to 0% claim. Huang is telling U.S. policymakers that export controls may slow China’s access to advanced chips, but they may also encourage Chinese developers and customers to move away from Nvidia and, more broadly, away from U.S. technology platforms.

For Huawei and local rivals, it is an opening, not an automatic win

Huawei is one of the clearest potential beneficiaries. Reports citing the South China Morning Post said Huang’s comments were consistent with his long-running view that if Nvidia cannot sell in China, the market will be taken over by Chinese competitors such as Huawei.

Other Chinese AI-chip companies are also part of the opening. Reports have pointed to Huawei Ascend and Cambricon as domestic firms moving to fill high-end computing gaps. A Bernstein Research forecast cited in Chinese media projected that by 2026 Nvidia’s share of China’s AI-chip market could fall to 8%, while Huawei could reach 50%, AMD 12%, and Cambricon could rank third. That is a forecast, not a confirmed market outcome.

The important caveat is that replacing Nvidia is not like swapping one card for another. If Nvidia’s strength comes from the combination of chips, networking, servers, software and algorithms, any alternative must solve the same system-level problems: supply at scale, software maturity, developer migration, cluster stability and industry-specific deployment.

For China’s AI-chip market, procurement logic changes

For Chinese buyers of AI computing capacity, the central question becomes less about whether Nvidia chips are preferred and more about whether they can be bought, deployed and upgraded with enough certainty. As long as compliant supply of advanced Nvidia GPUs remains unstable, customers have to adapt more seriously to domestic or non-U.S. options.

That shifts competition from the best single accelerator to the best usable system. Training frameworks, operator support, inference deployment, cluster interconnects, operations tooling and model optimization all become part of the replacement process. In the short term, that raises migration and adaptation costs. Over time, real orders and real workloads could give Chinese AI-chip vendors more chances to improve, but whether that becomes durable competitiveness depends on product quality, software ecosystems and deployment maturity.

The bottom line

Huang’s 95% to 0% line should not be read as a clean installed-base statistic. It is a compressed way of describing Nvidia’s current visibility into compliant high-end AI-chip sales in China under U.S. export controls. For Nvidia, the short-term effect is lower China revenue expectations; the longer-term risk is losing a major AI developer ecosystem. For Huawei and other Chinese chipmakers, it creates a demand window, but not a guaranteed victory. For China’s AI market, it likely accelerates localization while raising the cost of software migration, cluster adaptation and system rebuilding.