A limitation on exports of AI processors made in the US that was set to go into effect on May 15 has now been rescinded after an order from the Department of Commerce (PDF).
The AI Diffusion Rule was set in motion at the end of the Biden administration and would have limited the number of AI processors that could be sold to foreign countries. It was meant to give the US an edge in the race for global AI technology, but critics, including tech companies, said it would hurt the companies making the processors by capping their sales.
The Trump administration’s move now lifts restrictions on companies such as Nvidia from selling processors meant for AI applications to other countries, including Mexico, China and Russia.
“These new requirements would have stifled American innovation and saddled companies with burdensome new regulatory requirements,” the Department of Commerce said in a statement. “The AI Diffusion Rule also would have undermined US diplomatic relations with dozens of countries by downgrading them to second-tier status.”
The Department of Commerce said it is still warning companies and the public about allowing US-made AI chips to be used for Chinese AI models and is discouraging the use of Huawei Ascend chips specifically for exported products.
‘Cautious relief’ among chipmakers
Some in the industry worried that the rule could have hurt startups or mid-tier players, but now, the mood is “cautious relief,” said one tech-industry executive.
“While most companies support the need for guardrails around advanced AI capabilities, the AI Diffusion Rule as originally framed created a lot of ambiguity — especially around how ‘diffusion’ would be interpreted in deployment, licensing or technical architecture,” said Dmitry Zakharchenko, chief software officer for Blaize, a California-based AI chip company.
Zakharchenko says the concern among companies like his wasn’t just sales but also cross-country research collaboration, AI model deployment and cloud-infrastructure partners.
“The risk was that overbroad policy could inadvertently restrict benign use cases, including public safety, industrial inspection, and transportation,” he said.
As for what comes next, Zakharchenko said that some of the companies that would have been affected are taking their own steps to mitigate risks from overseas. That, he said, is happening “both at the hardware level and through partner ecosystems.”
“That includes secure boot, encrypted pipelines, regional deployment controls and tighter collaboration with integrators who understand where and how inference workloads are run,” Zakharchenko said.
