The Republican chair of the US House Select Committee on China has protested the Trump administration's decision this week to lift restrictions on the sale of Nvidia H20 GPUs and similar processors, warning the chips could be used to advance Chinese AI and military interests.

"We must not allow US companies to sell these vital artificial intelligence (AI) assets to Chinese entities," Representative John Moolenaar (R-MI), who chairs the committee, wrote in a letter to Secretary of Commerce Howard Lutnick.

The Trump administration had previously moved to block the sale of Nvidia's H20 and similar silicon back in April for fear that they could find their way into Chinese supercomputers. Nvidia's H20, alongside AMD's MI308 hardware, were originally released as cut-down versions of their more powerful siblings in order to comply with US export controls governing the sale of AI accelerators to China.

Following the decision, Nvidia estimated the decision would cost the GPU giant nearly $10.5 billion in the first half of its 2026 fiscal year. AMD reported the export ban would cost the firm $1.5 billion in lost revenues in 2025.

This week, the effective ban was lifted, with the chipmakers announcing they'd applied for licenses to resume sales of the accelerators, with Nvidia reporting that the "U.S. government has assured NVIDIA that licenses will be granted, and NVIDIA hopes to start deliveries soon."

Executives from both Nvidia and AMD have argued that if the US makes it too difficult to obtain US semiconductors, they'll seek out alternatives.

In his letter to Lutnick on Friday, Moolenaar rejected this assertion. "Nvidia can produce these chips at a scale that their Chinese competitors cannot rival, making it unlikely that restricting H20 exports to China would simply shift demand to domestic Chinese alternatives."

Moolenaar's comments reflect recent findings from Canadian research outlet TechInsights, which suggested SMIC, the Middle Kingdom's top chip manufacturer, remains generations behind the rest of the world. 

Beyond their use in training Chinese AI models, Moolenaar also warned that the chips could be used by Beijing to build supercomputers in violation of the Supercomputer end-use rule.

"Tencent reportedly used H20s to train its Hunyuan-Large model a project that almost certainly used one or more computing clusters requiring over 200 [petaFLOPS] of computing power (roughly 4,550 H20 GPUs), which meets the US definition of a 'supercomputer.'"

In case you're wondering, the US Export Administration Regulations defines a supercomputer as:  "A computing 'system' having a collective maximum theoretical compute capacity of 100 or more double-precision (64-bit) petaFLOPS or 200 or more single-precision (32-bit) petaFLOPS within a 41,600 ft3 or smaller envelope."

Based on what little information there is out there on the H20's higher precision performance, the chips support up to 44 teraFLOPS at single-precision, which roughly aligns with the figures cited in Moolenaar's letter to Lutnick.

Moolenaar closed his letter by demanding a briefing with Lutnick no later than August 8 to assess how the department plans to evaluate license applications for H20 and similar chips.

The Register reached out to Moolenaar's office for comment, but had not heard back at the time of publication. �