VoltgridNews

NVIDIA Chips for China's Military: USCC Report Cites PLA Procurement

The USCC said Nov. 18 that China is trying to evade export controls to reach advanced NVIDIA chips for military AI, but it gives no count of chips delivered to the PLA.

Chips & Power Desk

3 min read

Cover of the 2026 Annual Report to Congress of the U.S.-China Economic and Security Review Commission, with the US Capitol and the commission seal

The cover of the USCC's 2026 Annual Report to Congress, released Nov. 18.

Image: U.S.-China Economic and Security Review Commission

Source · Image from the USCC website, a U.S. congressional commission, used for news reporting.

The US-China Economic and Security Review Commission said on Nov. 18 that China's military is seeking the most advanced NVIDIA chips for AI-enabled weapons, and that Beijing is trying to evade US export controls to get them.

In its 2026 annual report to Congress, the congressional advisory commission wrote that "China is seeking to evade U.S. export controls to gain the most advanced semiconductors that underpin its intelligentization drive." It defines the term as the People's Liberation Army's push to build AI into weapons, kill chains, command and control, and logistics. Reuters, which reported the findings, said domestic chips such as Huawei's Ascend and Cambricon's Siyuan series still trail Nvidia and AMD products.

The commission's evidence is a procurement trail, not a delivery record. It cites an April 2026 assessment by the Center for Security and Emerging Technology, which concluded from PLA procurement documents of the past several years that the army is actively soliciting Nvidia's advanced chips. The passage does not say how many chips reached the PLA or which models. Voltgrid has tracked that gap in earlier reporting: Reuters found on Nov. 11 that a PLA-linked university bought two H200 chips from a reseller, and a Nov. 13 Reuters review counted $3.2 billion of China AI projects built on restricted Nvidia parts.

How Far Huawei's Ascend Chips Trail NVIDIA

The report puts numbers on the domestic gap. Its Table 3 lists the Huawei Ascend 910C at 12,800 total processing performance (TPP) against 40,000 for Nvidia's B200, and 3.2 TB/s of memory bandwidth against 8 TB/s. The commission labels the sources as various public estimates and rounds the gap to about 3x on compute and 2.5x on bandwidth.

The USCC's Table 3 puts the Ascend 910C about 3x behind the B200 on compute and about 2.5x on memory bandwidth, from various public estimates.

Graphic: Voltgrid News

The power consequence is spelled out in the text. Because Huawei's chips are weaker, the report says, Huawei needs more chips, more power, more physical infrastructure and more sophisticated networking to match US performance. It cites estimates that Huawei made 200,000 to 400,000 Ascend units in 2025, against four to five million Nvidia units. For a Chinese data-center operator, matching a US cluster's output with domestic parts implies a larger power and networking bill, which is one reason access to Nvidia parts stays valuable.

Remote access and the policy ask

The commission also points to remote access: Chinese AI companies reportedly reach advanced Nvidia chips through data centers in Southeast Asia, and it says existing rules have "evidently not fully prevented" this. It notes that the Commerce Department's Bureau of Industry and Security clarified in May 2026 that controls bar exports to companies headquartered in China even when they operate abroad.

Its recommendation targets supply rather than smuggling. It asks Congress to amend the Export Control Reform Act of 2018 so that exporters of advanced AI chips to China must certify that the sales will not divert wafer fabrication, high-bandwidth memory or advanced packaging capacity from the US market, with auditable data and criminal penalties for false certification. The commission's recommendations are advice to Congress and do not bind anyone.

What is not confirmed yet

Nvidia has said US controls have effectively shut it out of China's AI data-center market. Reuters reported Wednesday that Washington said on Monday it would further restrict China's access to advanced chips from companies including Nvidia, and Voltgrid reported on Nov. 18 that Nvidia has stopped shipping licensed Hopper chips to China. The reports Voltgrid read carry no response from Nvidia or Beijing to the findings, and the commission does not name which PLA units sought which chips.

Sources

Published
Sources
2 cited