China Bans US AI Chips in Data Centers
Mandate Aims to Reduce U.S. Dependence as AI Competition Intensifies
As the U.S. has stated it will not supply NVIDIA’s latest AI chips to China in the future, Beijing has imposed a ban on the use of American AI chips in its domestic AI data centers. Analysts suggest this marks a signal that China, which has relied on NVIDIA’s GPUs, is effectively declaring a move away from U.S. dependence and igniting the AI hegemony competition.
Reuters reported on the 5th (local time) that the Chinese government announced guidelines mandating the use of only domestically produced AI chips in new AI data centers funded by state resources. It added that data centers with a completion rate below 30% since construction began have been instructed to replace U.S.-made AI chips with Chinese ones. This effectively signals the expulsion of U.S. AI chips from Chinese data centers.
AI chips include GPUs and AI accelerators connected to high-bandwidth memory (HBM). NVIDIA dominates over 90% of the global market. The U.S. has restricted NVIDIA to exporting only low-performance AI chips to China. The tech industry views Beijing’s mandatory use of domestic AI chips as a display of confidence that high-performance AI can be developed without NVIDIA’s technology. Some analysts argue it is a retaliatory card to pressure the U.S. to ease export restrictions.
Meanwhile, NVIDIA CEO Jensen Huang stated at an AI event in London on the same day, “China will surpass the U.S. in the AI race.” He assessed that the current gap in AI technology between the two nations is “a mere split second,” nearly caught up.
China’s Confidence in Technological Self-Reliance: ‘AI Development Sufficient Without NVIDIA’
This time, Chinese regulators issued guidelines requiring data centers receiving government funding to use only domestically produced AI chips. New data centers, as well as those less than 30% complete, must remove all foreign AI chips already installed and replace them with Chinese alternatives. Plans to purchase U.S. AI chips must also be canceled. Previously, the requirement was to use at least 50% domestic AI chips, but this has been strengthened to 100%. This sends a strong signal to the AI industry to reduce reliance on foreign technology, aligning with plans to operate AI data centers solely through domestic capabilities.
Chinese data center construction companies are under pressure. Currently, 400 to 500 data centers are being built across the country, with $100 billion (approximately 144 trillion Korean won) in government funding distributed. Most data centers receiving state support are now required to remove already installed NVIDIA chips and replace them with Chinese ones.
China’s Move to Diverge from NVIDIA
Analysts suggest Beijing’s recent measures were a premeditated step. Earlier in July, when the U.S. government reapproved exports of NVIDIA’s H20 to China, Chinese authorities instead advised domestic companies to avoid using the H20 due to security concerns. In September, they demanded ByteDance, Alibaba, and others halt orders and testing of NVIDIA’s RTX Pro 6000D, a chip exclusively exported to China. Customs authorities also blocked imports of NVIDIA chips at major ports nationwide. This is interpreted as Beijing judging that its AI chip manufacturing technology, despite U.S. pressure, has reached a viable stage.
A semiconductor industry source said,
China, once defenseless against U.S. regulations, now appears to have shifted to an offensive stance driven by self-reliance and confidence.
To encourage the use of less power-efficient domestic chips, Beijing introduced an electricity cost support system, offering up to 50% reductions in power fees for data centers using locally made chips.
After being cornered in 2022 by U.S. restrictions on NVIDIA’s A100 and H100 exports, analysts assess that China has strengthened its position over three years. Unable to secure even NVIDIA’s older models, Beijing turned to developing its own AI chips. It established a 344 billion yuan (approximately 70 trillion Korean won) investment fund for the semiconductor industry and aggressively supported promising companies. While early Chinese AI chips had modest performance, sustained investment has elevated them to levels targeting NVIDIA replacements. For example, Huawei developed the Ascend 910C in October of last year, achieving around 60% of the NVIDIA H100’s performance. To overcome lower performance, it bundled 384 chips to approach the efficiency of NVIDIA’s top-tier systems. Though less cost-effective, the strategy focuses on rapid performance gains through volume.
Cambricon, dubbed “China’s NVIDIA,” developed the MLU 590, which delivers 80% of the NVIDIA A100’s performance. Its AI chips are supplied to Chinese companies like ByteDance. The rise of Chinese AI chips and U.S. export restrictions have caused NVIDIA’s market share in China to plummet.
Jensen Huang, NVIDIA’s CEO, stated last month,
Our share of the advanced chip market in China fell from 95% in 2022 to 0%.
Will U.S.-China Semiconductor Conflict Intensify?
Some in the tech industry argue that while China’s “AI chip independence declaration” reflects a desire to fully break U.S. dependence, it remains difficult to match American technological levels immediately. Morgan Stanley previously projected that China’s AI chip self-sufficiency rate, 34% last year, would rise to 82% by 2027. However, experts now predict the 100% self-sufficiency timeline will accelerate significantly due to Beijing’s recent measures.
As China effectively declares war by expelling U.S. AI chips, the U.S.-China semiconductor hegemony conflict is expected to intensify. U.S. President Donald Trump stated on the 2nd that the latest NVIDIA AI chips would not be allowed in China, and Beijing’s recent measures carry an aggressive undertone. With Chinese AI semiconductor self-reliance on the horizon, the U.S. is likely to impose stricter controls.
Professor Lee Jong-hwan of Sangmyung University’s System Semiconductor Engineering Department said,
China’s retaliatory measures and the U.S.’s efforts to suppress them will escalate the rivalry.
READ the latest news shaping the AI Chips market at AI Chips News
China Bans US AI Chips in Data Centers, source
Add comment