Nvidia resumes H200 AI chip production for China amid shifting export policy

Nvidia has quietly restarted production of its H200 AI processors aimed at the Chinese market. The move comes after months of regulatory back-and-forth between Washington and Beijing over advanced semiconductor exports. CEO Jensen Huang confirmed the development publicly, saying demand from China has returned and that orders are already flowing in.

This is not a small footnote in the chip industry. China is one of the largest buyers of AI computing hardware in the world, and Nvidia's ability to sell into that market directly affects its revenue projections, its relationships with hyperscale cloud providers in the region, and how competitors position their own products.

Why production stopped in the first place

The H200 is one of Nvidia's most capable chips for large-scale AI workloads. It succeeded the H100 and was designed to handle the memory bandwidth demands of training and running large language models. When the Biden administration tightened export controls on advanced semiconductors in late 2023 and again in 2024, chips exceeding certain performance thresholds were effectively blocked from being sold to Chinese buyers without a license.

Nvidia had previously developed a China-specific version of its A100, called the A800, to stay within those thresholds. The H200 presented a similar problem. Its raw compute performance and memory specs put it well above the export control limits, which meant the standard version could not be shipped to China without triggering federal review.

What changed in Washington's export policy

The policy environment shifted in 2025. The Trump administration, returning to office in January, took a notably different approach to chip export controls than its predecessor. While it maintained restrictions on certain military-adjacent technologies, there were indications that commercially available AI hardware, particularly chips sold to private companies rather than state entities, could be treated differently. Nvidia's lobbying efforts, combined with pressure from domestic chipmakers worried about losing market share to Chinese alternatives, helped move the conversation.

The resumption of H200 production for China does not mean the export rules are gone. It means Nvidia has found a workable path, whether through specific licensing arrangements, modified chip configurations, or a combination of both. The exact technical and legal details have not been fully disclosed, which is standard in these situations.

Jensen Huang's confidence in Chinese demand

Huang has been vocal about China's importance to Nvidia's business. At a recent industry event, he described China as a market where AI investment is accelerating, not slowing. Chinese tech companies like Alibaba, Baidu, Tencent, and ByteDance have been building out massive AI infrastructure. They need compute. Nvidia still makes the most capable chips for that purpose, even with domestic alternatives like Huawei's Ascend 910B gaining traction.

Huawei's Ascend chips are good enough for certain workloads, but they lag behind Nvidia's hardware on memory bandwidth and software ecosystem maturity. Chinese AI developers who can access Nvidia chips generally prefer them for training large models. That preference is what Huang is counting on.

What this means for the broader chip competition

Nvidia's return to producing H200s for China puts pressure on AMD, which has also been navigating export restrictions for its MI300X accelerator. Intel's Gaudi line faces a similar situation. If Nvidia secures a regulatory path that others cannot easily replicate, it gains a structural advantage in a market that could be worth tens of billions of dollars annually.

There is also a longer-term dynamic worth watching. Every quarter that Chinese AI labs use Nvidia hardware is another quarter they build their engineering workflows, tooling, and developer talent around CUDA, Nvidia's proprietary software platform. That creates switching costs that do not disappear even if domestic alternatives improve.

The risk Nvidia is taking

Not everyone in Washington is comfortable with this development. Some members of Congress who pushed for stronger export controls have argued that advanced AI chips, regardless of who buys them commercially, can eventually find their way into military or surveillance applications. That concern has not gone away. A change in political winds, a high-profile incident, or a shift in US-China relations could tighten restrictions again, potentially leaving Nvidia with production capacity and supply commitments it cannot fulfill.

Nvidia is essentially betting that the current policy environment holds long enough to justify the production restart. Given that the company reportedly lost around $5 billion in annual revenue from the initial export restrictions, the financial logic is straightforward. The geopolitical risk is harder to quantify.

Orders are already arriving, according to Huang. Production is running. The next few quarterly earnings reports will show whether this bet pays off in actual shipments and revenue, or whether another regulatory turn disrupts things again.