NVIDIA H200 AI GPU Shipments to China Begin as Export Rules Ease
NVIDIA’s H200 AI GPUs are reportedly shipping to China at last, marking a major shift in the ongoing battle over advanced artificial intelligence hardware and U.S. export restrictions. According to a senior U.S. official cited in recent reports, the first shipments of NVIDIA H200 chips have now begun moving toward China and Hong Kong, although the initial volume is still described as very limited compared with demand.
The development is significant because NVIDIA’s high-end AI accelerators have been at the center of U.S.-China technology tensions for years. While the newest Blackwell GPUs remain restricted from sale in China, the older Hopper-based H200 appears to be returning to the market under a more flexible export approach.
NVIDIA’s China AI chip business has had a turbulent path. When the Hopper architecture first arrived, U.S. export controls were already limiting what could be sold to Chinese companies. To keep serving the market, NVIDIA introduced the H20, a China-focused AI GPU designed to comply with export rules.
The H20 was far less powerful than NVIDIA’s top-tier global offerings, but it still became an important option for Chinese cloud providers, AI labs, and technology firms. Demand reportedly increased as companies looked for alternatives to more advanced restricted GPUs. However, China later encouraged a stronger shift toward domestic AI chips, putting pressure on the H20’s position in the market. That move eventually led NVIDIA to pause H20 production for China.
Despite official restrictions, Hopper GPUs remained highly desired in China. Limited availability and strong demand helped create unofficial supply channels, with some chips reaching the market through gray-market routes. The new H200 shipments suggest that at least part of that demand may now be addressed through approved channels.
The H200 is no longer NVIDIA’s latest AI accelerator, but it remains a powerful chip for large-scale AI training and inference workloads. It was first introduced in 2023 as part of the Hopper generation and brought improvements over the earlier H100, including faster memory and higher bandwidth. Since then, NVIDIA has moved ahead with Blackwell and Blackwell Ultra GPUs, but those newer products are still blocked from China under current export rules.
The reopening of H200 sales follows discussions involving U.S. and Chinese leadership earlier this year. During a May meeting between Donald Trump and Xi Jinping, NVIDIA CEO Jensen Huang was reportedly among the technology executives present. After that, reports indicated that select major Chinese AI firms would be allowed to purchase H200 GPUs.
The initial plan was said to include around 10 major companies, with permission to buy up to 75,000 H200 chips. Firms reportedly included leading names such as Alibaba, Tencent, ByteDance, JD, DeepSeek, and others. More recent claims suggest that the potential purchase limit could rise as high as 200,000 H200 GPUs, though the actual number now being shipped appears to be far smaller.
For NVIDIA, even a small number of approved H200 shipments could be important. CEO Jensen Huang has previously stated that NVIDIA’s official AI GPU market share in China had fallen to 0% due to export restrictions and local policy shifts. The return of approved H200 sales could help the company regain at least a small foothold in one of the world’s largest AI markets.
Still, the scale of the opportunity remains uncertain. Chinese companies are racing to build and deploy large AI models, cloud infrastructure, and agentic AI systems, all of which require enormous computing resources. Demand for NVIDIA AI GPUs remains high, but supply approvals, export policy, and geopolitical conditions could continue to limit access.
At the same time, China is investing heavily in homegrown AI semiconductor alternatives. Domestic chipmakers are working to close the performance gap with NVIDIA, and government-backed efforts are pushing major firms to rely more on local hardware. That means NVIDIA’s return with the H200 may face both regulatory limits and rising competition from Chinese AI accelerators.
Looking ahead, NVIDIA is preparing its next major platform, Vera Rubin, featuring Rubin GPUs and Vera CPUs. Volume production for the new platform is already underway. While GPU exports to China remain tightly controlled, CPUs are not currently restricted in the same way, which could allow NVIDIA to continue supplying certain products to Chinese customers.
The broader AI hardware market is also changing. As agentic AI grows, demand is expanding beyond GPUs alone. CPUs, networking chips, memory, and full data center platforms are becoming increasingly important for companies building advanced AI systems. This shift could create new opportunities for NVIDIA, even in markets where its most powerful GPUs remain limited.
For now, the start of NVIDIA H200 AI GPU shipments to China signals a cautious reopening rather than a full return. The chips are older than NVIDIA’s latest Blackwell lineup, and the number of units appears small. But for Chinese AI companies hungry for high-performance computing, and for NVIDIA seeking to rebuild its official presence in China, even limited H200 availability could have a meaningful impact.
