China Bans The Use Of Foreign AI Chips In State Funded Data Centers

Nvidia, deeply affected by China's government policy (photo: x @MarioNawfal)

JAKARTA - The Chinese government has issued a new directive requiring all data center projects to receive state funds to only use domestically made artificial intelligence (AI) chips. This is known according to two sources familiar with the policy to Reuters.

In recent weeks, Chinese regulatory authorities have ordered data centers whose construction is still below 30 percent to remove all foreign chips that have been installed or canceled their purchase plans. While projects that are already at a further stage will be assessed case by case, the source said.

This move could be one of China's most aggressive efforts so far to remove its dependence on foreign technology in its key infrastructure, amid a pause in trade tensions between Beijing and Washington, as well as accelerate its ambition to achieve self-sufficiency in AI chips.

China's access to advanced AI chips including those made by Nvidia has become one of the main friction points with the United States, as the two countries struggle for dominance in high-level computing power and artificial intelligence.+

US President Donald Trump said in an interview that aired last Sunday, after his meeting with Chinese President Xi Jinping that Washington would "care for them (China) to deal with Nvidia, but not for the most advanced chips."

However, Beijing's new policy seems to be dashing Nvidia's hopes of reclaiming its market share in China, as well as providing great opportunities for local competitors such as Huawei to expand their chip sales.

It is not yet clear whether this directive applies nationally or only in several provinces, and the source did not mention which regulatory agency issued the order due to the sensitivity of this issue.

Apart from Nvidia, other foreign chip manufacturers that have the potential to be affected include AMD and Intel. Meanwhile, China's Cyber Administration and the National Development and Reform Commission have yet to comment on this report.

China's AI-based data center project has withdrawn state funding of more than 100 billion US dollars since 2021. Most of the projects do receive public funding, although it is not yet known how many will be directly affected by this new policy.

Several projects have even been suspended before their start, including facilities in the northwest region that originally planned to use Nvidia chips. The project was developed by a private technology company that received state funds, said one source.

Beijing has long been upset over Washington's export restrictions to slow China's technological advances and has taken a series of countermeasures to reduce its dependence on US technology.

The United States argued that the restrictions were needed for fear the chips could be used by the Chinese military to improve its defense capabilities.

This year, the Chinese government has also appealed to domestic tech giants not to buy advanced Nvidia chips for safety reasons, as well as show off new data centers that are fully powered by domestically made AI chips.

In 2023, Beijing will also ban the use of Micron products in its critical infrastructure, a move that will then see the US-largest memory chipmaker withdraw from the server market in China.

Nvidia CEO Jensen Huang has been lobbying President Trump and his cabinet to allow the sale of more AI chips to China, arguing that China's dependence on US hardware actually benefits America's interests. But now, China's Nvidia market share has fallen to zero percent, from the previous 95 percent in 2022.

This new policy includes the most advanced H20 series Nvidia chip AI which is still allowed to be sold to China as well as the stronger B200 and H200 models. Although the last two models were officially banned from being officially exported by the US, both are still circulating on China's gray market.

With these new directives, the Chinese government has the potential to expand market share for local chip manufacturers such as Huawei Technologies, Cambricon, as well as startups such as MetaX, Moore Threads, and Enflame.

Products from these companies are starting to compete with Nvidia's chips, although adoption is still limited because developers are more accustomed to the reliable Nvidia software ecosystem.

The move will increase sales of domestic chips, but also risk widening the AI's computing power gap between China and the US.

Big US tech companies such as Microsoft, Meta, and OpenAI have poured hundreds of billions of dollars into building the most advanced Nvidia chip-powered data centers. Meanwhile, major Chinese chipmakers such as SMIC are still facing supply constraints due to US sanctions on semiconductor production equipment that limits their ability to produce advanced chips.