China is preparing to allow its top artificial intelligence firms to purchase a limited number of Nvidia H200 chips, according to a report from The Information, ending a domestic restriction that had kept the country's AI leaders from buying the Silicon Valley giant's most capable China-compliant hardware.

The move is a significant shift in the global AI chip trade and a major signal for the latest AI developments in the intensifying US-China technology rivalry.

The Numbers Behind the Decision

According to The Information, whose reporting was picked up by Reuters and Bloomberg, the Chinese government may approve purchases of roughly 200,000 H200 chips. That figure is less than half of what Chinese companies have actually requested, signaling that Beijing intends to ration supply rather than open the floodgates.

The companies reportedly on the shortlist for approval include:

  • Alibaba, whose cloud and Qwen model division is among China's largest AI operations
  • ByteDance, the parent company of TikTok and a major buyer of AI compute
  • DeepSeek, the model lab whose low-cost training breakthroughs have shaken the global AI market

The government is still deliberating the exact number of chips that will be allowed, the report notes, and the final allocation could shift.

Why China Restricted Its Own Firms in the First Place

The H200 is a China-specific chip that Nvidia developed to comply with US export control laws. After Washington imposed sweeping restrictions on advanced semiconductor sales to China, Nvidia built compliant variants that fall below the performance thresholds set by American regulators.

But in a twist, the Chinese government then limited its own domestic companies from buying those compliant chips. The restriction was part of Beijing's broader effort to counter US export controls and to keep tight oversight on which firms receive advanced compute. Recent reporting also indicated that Beijing grew concerned about the cybersecurity capabilities of Western AI systems, particularly Anthropic's Mythos models.

Now, that self-imposed limit appears to be easing. Nvidia CEO Jensen Huang confirmed to CNBC in May that his company had secured China's approval to sell its chips to local firms. At the time, Huang noted that Nvidia had not included any revenue from China sales in its financial guidance, and told reporters at GTC in March that the company was preparing to supply chips to Chinese customers.

What the H200 Means for Chinese AI Labs

For Chinese AI companies, access to even a rationed supply of H200 chips could meaningfully accelerate model training and inference. Nvidia's hardware remains the gold standard for large-scale AI compute, and Chinese labs have been forced to rely on a mix of older Nvidia inventory, domestic alternatives from Huawei, and increasingly capable in-house chips from companies like DeepSeek and Alibaba.

The 200,000-chip figure, while substantial, would need to be shared among the country's most compute-hungry players. With demand running at more than double that level, the allocation will inevitably force Beijing to pick winners — favoring firms it considers strategically important to China's AI ambitions.

A Market Thirsting for Nvidia Hardware

The potential approval is a welcome signal for Nvidia, which has navigated a complex regulatory landscape between Washington and Beijing for years. China was once one of Nvidia's largest markets, and the loss of unrestricted sales there has been a persistent drag on the company's growth in the region.

It is worth noting how far Nvidia's product line has moved. The H200 belongs to Nvidia's Hopper generation. The company's latest offerings are now two generations ahead: the Rubin platform is slated to begin shipping this fall, and the next-generation Rubin Ultra GPUs are planned for 2027, though some Rubin Ultra chips may slip to 2028 due to production constraints.

That means even if China approves H200 sales, the approved hardware will be several generations behind what Nvidia sells to customers in the United States and allied countries — a gap that US export controls were designed to preserve.

Geopolitical Implications

The decision sits at the center of a multi-layered tech standoff. The United States has used export controls to slow China's AI progress; China has responded with its own restrictions and a push for semiconductor self-sufficiency. Allowing limited H200 purchases suggests Beijing is recalibrating — prioritizing the near-term needs of its AI champions over a purely adversarial posture.

For the firms on the approval list, the chips would provide a bridge while domestic alternatives mature. For Nvidia, it reopens a sliver of the Chinese market under tightly controlled conditions. And for the broader industry, it is a reminder that the global AI supply chain remains as much a matter of geopolitics as it is of engineering.

Stay Ahead of AI

For ongoing coverage of AI chips, export controls, and the companies shaping the future of computing, follow the AI industry coverage at AI Buzz Wire.

Read more AI news →