China may ease its NVIDIA H200 restrictions for Alibaba, ByteDance and DeepSeek

China is weighing a limited reversal of its own restrictions on NVIDIA’s H200 AI chips, a move that could give major local technology companies some access to hardware they have been waiting to buy.

According to The Information, Chinese officials are considering approvals that would allow companies such as Alibaba, ByteDance and DeepSeek to purchase a capped number of NVIDIA H200 chips. The figure being discussed could be under 200,000 units, which would still be less than half of what Chinese firms reportedly asked for.

The possible shift matters because it would soften Beijing’s earlier stance toward NVIDIA’s China compliant AI hardware. The United States had already placed export controls on NVIDIA’s most advanced chips, forcing the company to design products that could still be sold into China while meeting American rules. The H200 was part of that effort, but China later discouraged or limited domestic companies from buying it as part of its broader push to reduce reliance on foreign technology.

Beijing is balancing AI demand with its push for local chip independence

The reported change suggests China may be trying to manage two competing needs. On one side, it wants local companies to rely more on domestic chips and reduce exposure to American restrictions. On the other side, Chinese AI firms still need large amounts of high performance computing power to train and run advanced models.

That pressure is becoming harder to ignore. Companies such as Alibaba, ByteDance and DeepSeek are central to China’s AI race, and limited access to NVIDIA hardware could help them meet near term demand while domestic alternatives continue to develop.

The policy shift has not been finalized, and the exact number of chips remains under discussion. The Information reported that Chinese officials have told some companies they may soon be allowed to buy H200 chips, but the final approval level may remain tightly controlled.

NVIDIA CEO Jensen Huang has also said the company had secured approval from China to sell chips to local companies. Earlier this year, Huang said NVIDIA had not included China sales guidance in its financial estimates, a sign of how uncertain the market had become for the company.

Even if the approvals move forward, the H200 is no longer NVIDIA’s newest AI product. The chip belongs to the Hopper generation, while NVIDIA has already moved ahead with newer AI platforms. The company is preparing Rubin shipments for later this year, followed by Rubin Ultra in 2027, although some reports have suggested that production constraints could push parts of Rubin Ultra into 2028.

For Chinese companies, that gap still may not reduce the H200’s importance. Access to older NVIDIA chips can still be valuable when the alternative is a shortage of high end AI accelerators. For NVIDIA, even a limited reopening in China would matter because the country remains one of the most important AI hardware markets, despite ongoing political and regulatory pressure.

The broader picture is clear. China is not abandoning its goal of building a stronger domestic chip industry, but it may be willing to use NVIDIA hardware where it helps local AI companies stay competitive. That makes the reported H200 approvals less like a full policy reversal and more like a controlled compromise.