Nvidia confirmed to Nikkei Asia that two of its premium AI computing chips and one type of its powerful AI computing system were affected by a new US regulation. The company said it will seek export control licenses and will also talk to customers in China about the issue.
AMD meanwhile alerted its China operations on Wednesday that the US chipmaker will suspend some shipments of high-end GPUs to China, according to two people familiar with the matter. AMD confirmed to Nikkei Asia that it has received notification of new licensing requirements from the US Department of Commerce that prevent it from shipping MI250 integrated circuits to China and Russia.
“The only current products that the new licensing requirement applies to are A100, H100 and systems such as DGX that include them,” a Nvidia spokesperson said, referring to the company’s premium AI accelerators, which can be used to develop cutting-edge supercomputers.
AMD’s MI250 chip is another such AI accelerator.
An official from the Commerce Department told Nikkei Asia that the agency cannot comment on specific policy changes but said it is taking a comprehensive approach to implementing additional actions necessary to protect US national security and foreign policy interests and avoid any advanced US technologies ending up being used in Chinese military applications.
Washington’s move marks the latest escalation in the US-China tech battle.
In August, the US restricted exports of the most advanced chip design software to curb Chinese efforts to produce cutting-edge chips locally. The CEOs of two leading US semiconductor equipment makers, Lam Research and KLA, also confirmed in earnings conferences that the US had further restricted the shipment of machines that can be used to build chips more advanced than those at the 14-nanometer level.
Currently, the world’s top chipmakers – Intel, TSMC and Samsung – are racing to produce 3-nm chips using the latest chip production technologies. Chips with smaller nanometer sizes are more powerful but also more challenging to develop.
The US found that the Chinese military is still heavily relying on American technologies and advanced chip production capacity in Taiwan and South Korea for procuring its AI chips, according to a research report in June from Georgetown University’s Center for Security and Emerging Technology that analysed thousands of public purchasing records for the People’s Liberation Army.
Of the 97 individual AI chips that it could identify in the records, nearly all were designed by Nvidia, Xilinx (now an AMD company), Intel or Microsemi, the report said.
By comparison, the Georgetown team could not find any public records of PLA units or state-owned defence enterprises placing orders for high-end AI chips designed by Chinese companies such as HiSilicon (Huawei), Sugon, Sunway, Hygon or Phytium.