So-called “AI-diffusion” restrictions, set to take effect within weeks, will limit GPU sales not just to adversaries but to friendly countries like Israel, Switzerland, India, and Saudi Arabia.

The restrictions also curb the ability of US cloud giants—Alphabet, Amazon, and Microsoft—to build out data centre capacity in these “Tier 2” nations. Washington’s thinking is that AI compute could be laundered through allies to reach blacklisted countries like China.

Nvidia isn’t pleased. A spokesNvida branded the rules “misguided” when they were introduced in the dying days of the Biden administration. Since then, the new White House has shown no signs of pulling back, doubling down this month by banning the firm’s China-focused H20 processor.

Analysts reckon these moves will cut deep. Bank of America’s Vivek Arya estimates that nearly a quarter of Nvidia’s revenue last year came from non-China countries that are now under restriction. Combined with the earlier curbs on China itself, that could translate into a 14 per cent revenue hit—$28 billion if Nvidia hits its projected $201 billion for the current fiscal year.

The shock hasn’t derailed Nvidia yet. The firm is sitting on a fat cash pile and still feasting on AI demand from US tech giants. But the longer-term fallout could be brutal. Limiting the firm’s overseas reach hands a free pass to foreign rivals like Huawei, as well as chipmakers in South Korea, Japan, and Europe.

With Nvidia still lobbying hard, some think a compromise is possible. TD Cowen analyst Paul Gallant reckons the US could broaden Tier 1 eligibility—allowing more countries to buy GPUs freely. But that’s far from assured.

Nvidia’s chief executive Jensen Huang has been lobbying behind closed doors while dangling a $500 billion AI infrastructure investment in the US.

UBS’s Tim Arcuri said: “We struggle to see Nvidia agreeing to make $500 billion in AI infrastructure investment in the US if both the H20 would be banned AND that it would be unsuccessful in killing the AI Diffusion Rule.”