Who in China Doesn't Want Nvidia Processors? Surprising Reactions to the US Move

• The authorities have already encouraged IT companies several times to replace Nvidia chips with domestic equivalents.
• However, Chinese companies prefer American solutions and try to circumvent the bans.
• This discrepancy is most likely due to differences between the goals of the government and the IT industry, with Huawei being the common denominator.

Companies on both sides of the Pacific are being very creative in circumventing the government's constraints. Last year, Nvidia introduced the H20 processor specifically for the Chinese market. The AI ​​semiconductor giant aimed to create a chip that would appeal to Chinese customers while also complying with Washington's restrictions. The H20 is therefore less powerful than the Blackwell processor family, the company's current flagship product.

Initially, it seemed that both goals had been achieved. Despite the H20's limited computing power, Chinese companies lined up to order it. However, the Donald Trump administration blocked H20 exports to China in April 2025. Nvidia CEO Jensen Huang leveraged his new position as the president's favorite from Silicon Valley and managed to convince the White House president to change his mind.

It's unclear what arguments Huang used or what President Donald Trump's calculations are. It's likely that, faced with the development of sophisticated methods for smuggling AI processors into China, Washington may have concluded that tilting at windmills is pointless and that it would be better to allow the legal sale of chips with reduced capabilities, while also reaping the financial benefits. The price for changing the restrictions will likely be that Nvidia will hand over 15% of its profits from H20 sales to China to the government .

This hypothesis is supported by the apparent shift in the center of gravity of American pressure on the Chinese semiconductor industry, through, on the one hand, a ban on the sale of cutting-edge chips and, on the other, a blockade on the export of chipmaking tools. This strategy was initiated by the Biden administration under the CHIPS Act and accompanying policies, and is being continued by the Trump administration.

Beijing turns up its nose at American chips

Huang's joy was short-lived. In early August, Bloomberg reported that the China Cyberspace Administration—the powerful institution regulating China's digital space —had summoned tech companies, including ByteDance, Alibaba, Tencent, and Baidu, to a meeting and "recommended" them to abandon Nvidia chips, at least in projects deemed important for national security. This information was later confirmed by the Japanese economic website Nikkei Asia.

This wasn't the end. In mid-September, the China Cyberspace Administration reportedly ordered technology companies to cease testing and order another Nvidia processor, the RTX Pro 6000D. This GPU has been modified for AI applications. Bloomberg and the Financial Times independently reported the information.

Reports have also begun to emerge from China about alleged backdoors—intentional security flaws installed in the American company's processors. Similar accusations have been leveled at Huawei for years. Nvidia, like the Chinese conglomerate, has vehemently denied them.

The matter, however, appears more complex. There is no formal ban, although in the Chinese political system, "recommendation" and "ban" are often synonymous terms. However, according to Forbes, the RTX Pro 6000D, a new addition to Nvidia's lineup, failed to gain traction with Chinese users. The processor is more expensive and has weaker performance than Nvidia's RTX5090 graphics cards, which are banned from export to China. This version of events was provided by the Chinese Ministry of Foreign Affairs. This raises the question: if the test results were truly unsatisfactory and Chinese companies are voluntarily canceling orders, why the government's intervention?

There are also claims that Beijing's actions are part of a negotiating strategy. By restricting Nvidia's access to its market, China aims to force Washington and Nvidia to make concessions, specifically, permission to export Blackwell processors.

The authorities say: buy Chinese!

Such tactics shouldn't be ruled out. However, discouraging companies from purchasing Nvidia equipment perfectly aligns with Beijing's desire to achieve self-sufficiency in semiconductors. The recommendations from the China Cyberspace Administration also coincide with city officials' declarations of a transition to domestically produced integrated circuits. The capital, Beijing, aims to switch exclusively to domestically produced chips by 2027. The economic capital, Shanghai, is setting a slower pace. By 2027, the city is expected to meet 70% of its national demand in this regard, and that's only for data centers.

Equally interesting is the case of Guiyang, the capital of Guizhou Province in the south of the country. Local authorities aim to achieve a 90% utilization rate for domestic semiconductors by 2027. Why is this so important? Guiyang is one of the places in China where the largest number of new data centers are being built, also used by foreign companies such as Apple.

Are these plans realistic? It's difficult to say for sure. According to data from the Chinese financial firm Shanxi Securities, Nvidia controlled as much as 80% of the local AI processor market at the beginning of 2024. Despite the grandiose announcements and clear guidance from the authorities, the company's analysts estimate that by 2030, the American giant will maintain a 50-60% market share and remain the largest player. The remaining shares will be taken by Chinese entities.

Why do Chinese companies prefer Nvidia products for now?

Despite Beijing and Washington's policies, Chinese technology companies are reluctant to abandon American-designed chips, mostly manufactured in Taiwan by TSMC . There are several reasons. The first is the advanced level of Chinese chips, manufactured with AI in mind. The Huawei-designed Ascend 910B is expected to be comparable to Nvidia's H100. However, it falls short when compared to the more modern H20. The Chinese processor is expected to have 85% of the processing power of the American chip. The Chinese Ascend 920, scheduled to enter mass production later this year, is expected to achieve parameters similar to the H20.

Huawei's success in the technological pursuit of Nvidia cannot be denied. However, the Americans are not standing still .

Restrictions on access to machinery for producing advanced semiconductors are also having an impact. This is evident in the roadmap for AI processor development for the next three years, announced on September 18th. Chris McGuire, a former US State Department employee and expert in high-tech and national security, pointed out several surprising details of this plan.

The AI ​​processors Huawei plans to produce next year have less processing power and memory bandwidth than the top chips currently produced in China. A chip comparable to Nvidia's B30A isn't expected to enter production until late 2028.

Published forecasts indicate that by 2027, Nvidia's top-of-the-line AI chips will have up to 27 times more processing power than Huawei's best counterparts. The Chinese company also indicates that while the Americans will adopt increasingly modern solutions, its integrated circuits will remain focused on silicon wafer thicknesses of 5-7 nm. These are modern, but they lag behind the industry leaders.

Are chip clusters the solution?

To compensate for the weaker performance of its own chips, Huawei is relying on clusters. The most popular currently is the Nvidia NVL72 cluster, which combines 72 GB200 processors. Huawei's CloudMatrix 384 model was supposed to combine as many as 384 chips. Future plans are even more ambitious. The Atlas 950 SuperPoD and Atlas 960 SuperPoD clusters, scheduled for launch in 2027, are expected to combine 8,192 and 15,488 NPUs, respectively.

However, everything comes at a price. More chips mean higher power consumption and require more IT work. The CloudMaster 384's operating costs are estimated to be three to five times higher than those of the NVL72.

And that's not all. The weaker performance of domestic chips is impacting the development of Chinese artificial intelligence . According to the Financial Times, the use of Ascend processors was said to have led to a delay in work on the next version of DeepSeek. Despite the help of a team of specialists sent by Huawei, training the R2 model on Ascend processors was reportedly so frustrating that DeepSeek returned to Nvidia. The two companies continue to collaborate, but the focus has shifted from training new AI models to developing chips supporting AI applications. Figuratively speaking, instead of searching for gold, both companies were working on panning devices. It's unclear when R2 will be available to users, and users are switching to Alibaba's competing Qwen3.

If the newspaper's reports are true, then reliance on domestic chips poses a real problem for Chinese companies. It's not even about the advantages of foreign competitors; much fiercer competition is taking place within China itself. It's possible that by using Chinese chips, Chinese companies could lose out to domestic competitors working with American solutions .

This isn't the only problem facing Chinese technology companies. The authorities have likely designated Huawei as the national champion of the semiconductor industry. However, the giant is still searching for a new form of leadership after US sanctions curtailed its role in the telecommunications market. Hence the company's commitment to developing artificial intelligence, offering cloud services, and more. Huawei has thus found itself on a collision course with Alibaba, Baidu, ByteDance, and Tencent. Using the national champion's semiconductors means financing a competitor for these companies.

Chinese technology companies are increasingly separating their operations. Where national security requirements arise, Huawei processors are used, while Nvidia remains in use everywhere else. It could be argued that, in an attempt to solve one problem, Beijing has created another. As Kyle Chan of Princeton University points out, the issue is much more serious from the perspective of its semiconductor ambitions. China faces a difficult choice: should it prioritize increasing the performance of AI models today or developing more efficient chips for artificial intelligence in the future?