Why China isn’t rushing to buy Nvidia’s H200 AI chip

US President Donald Trump’s announcement on 8 December easing restrictions on Nvidia’s H200 chip exports to China elicited a muted response from Chinese authorities. Whether Nvidia can reopen the Chinese market remains uncertain.

In a Bloomberg article published on 13 December, White House AI czar David Sacks was quoted as saying that China has figured out the US strategy for allowing it to buy Nvidia’s H200 chips and is rejecting the AI chip in favour of domestically developed semiconductors. He said, “They’re rejecting our chips. Apparently they don’t want them, and I think the reason for that is they want semiconductor independence.”

Bloomberg also cited people familiar with the matter as saying that China plans to roll out a new funding programme of up to 500 billion RMB (US$70.9 billion) to support domestic chip manufacturing, underscoring Beijing’s determination to reduce reliance on foreign suppliers such as Nvidia and to advance autonomy and control in semiconductors.

Meanwhile, China may not necessarily rule out the return of Nvidia chips to its market. According to Trump, he has communicated with Chinese President Xi Jinping on the issue of opening up Nvidia chip sales to China and received a positive response from Xi.

At a regular press conference on 9 December, in response to reports that the US had approved sales of Nvidia’s H200 chips to China, Chinese Foreign Ministry spokesperson Guo Jiakun said, “We have noticed the reports. China always advocates that China and the United States achieve mutual benefit through cooperation.”

Although China’s domestically produced chips have made significant progress over the past two years, there remains a clear performance gap compared with the H200. The H200 is Nvidia’s second-fastest AI chip in terms of computing power and is currently the most powerful chip product accessible to Chinese companies.

It was also reported that major Chinese firms such as Alibaba and ByteDance have contacted Nvidia this week to discuss purchases of the H200, with both companies said to be considering large orders. Although Chinese authorities have not yet approved any procurement licenses for the H200, they may permit certain Nvidia chips to re-enter the Chinese market to support the development of domestic AI technologies.

From a broader perspective, China welcomes greater openness by the US in high-tech exports to China, as this would expand shared interests between both countries. However, China’s condition is that any chips exported by the US must not only meet security standards and align with China’s AI development needs, but also must not disrupt China’s drive towards domestic chip self-sufficiency.

From a technical perspective, although the H200 is not the most advanced chip in the US, for China, it is still a high-end AI chip suitable for large model training and inference. It remains highly valuable for Chinese AI companies such as Alibaba and ByteDance.

China is concerned about the extent to which the entry of the H200 chip into the Chinese market would impact the development of domestic Chinese chips. Chinese public opinion viewed the US decision to allow Nvidia’s high-end chips back into China — just as domestic chips are making continuous breakthroughs — as a “sugar-coated bullet” aimed at the country. The ultimate goal is to reclaim Nvidia’s dominant position in China’s chip market, hindering the growth and strengthening of domestic Chinese chips.

Over the past two years, amid strict US controls on the export of high-end chips to China, China’s AI computing power landscape has undergone a huge transformation.

In 2024, Chinese companies purchased roughly 1 million Nvidia H20 chips — customised for the local market — far outpacing domestic shipments of Huawei’s Ascend 910B, which reached about 450,000 units. The contrast highlights China’s continued reliance on foreign high-end AI hardware, even as its own chip industry makes significant advances.

However, in 2025, Huawei began large-scale shipments of the Ascend 910C, which features a “dual 910B” design — combining two 910B chips into one — and is considered a key product in the Chinese market aimed at rivalling Nvidia’s high-end chips.

At the same time, Chinese chip companies such as Cambricon and Moore Threads have been focusing on different aspects of AI workloads, including model training and inference, and are aligning themselves with domestic cloud providers and industry customers.

In other words, China has already started charting a new path towards diversified computing power and chip localisation in response to sanctions — and has seen significant results.

Yet high-end chips like the H200 remain highly attractive to China’s leading large-model enterprises because, for top-tier training clusters, the H200 still outperforms domestic chips. Once they can acquire the H200, companies such as Alibaba Cloud and Tencent Cloud are likely to use it for their flagship models. This is especially true for firms with overseas operations, which need to stay aligned with the international ecosystem; deploying new training clusters on the H200 allows them to share toolchains and models more easily with global partners.

This could put domestic Chinese chips under severe pressure — potentially threatening their very survival if left unchecked, a scenario Chinese authorities are keen to avoid.

Ultimately, the extent to which Nvidia can re-enter the Chinese market will hinge on how Chinese authorities manage the delicate balance between importing advanced foreign chips and promoting domestic alternatives in critical systems — ensuring they are not left vulnerable to a strategic chokehold by the US.

This article was first published in Lianhe Zaobao as “英伟达重新打开中国市场的关键”.

Related: Why Beijing thinks Nvidia’s H20 chip could be a spy tool | US and Chinese chipmakers tread different paths in AI gold rush