The semiconductor behemoth Nvidia is grappling with unanticipated new U.S. export restrictions targeting its H20 chips.

Nvidia stated in a filing on Tuesday that the U.S. government has informed them of a perpetual license requirement for exporting their H20 AI chips to China, citing the potential risk of the chips being employed in a supercomputer within the country.

Nvidia foresees incurring approximately $5.5 billion in related costs during its first fiscal quarter of 2026, concluding on April 27. Consequently, the company's share price decreased by around 6% in after-hours trading.

Under current and former US export controls, the H20 represents Nvidia's most sophisticated AI chip permitted for export to China. Last week, NPR indicated that CEO Jensen Huang may have successfully negotiated an exemption from new H20 limitations during a dinner at President Donald Trump's Mar-a-Lago establishment, partly by pledging Nvidia's investment in AI data centres within the US.

Perhaps not entirely by chance, Nvidia declared on Monday its intention to invest hundreds of millions of dollars over the forthcoming four years in the domestic production of certain AI chips. Commentators were swift to highlight the company's commitment as lacking in specific particulars.

Several government officials had asked for stricter rules on sending the H20 chip to other countries. This was because the chip was reportedly used to train models by a Chinese AI company called DeepSeek. One of these models was the R1 "reasoning" model, which caused a lot of confusion in the U.S. AI market in January.

Nvidia opted to refrain from making a statement.