WASHINGTON — President Donald Trump’s administration said it will let Nvidia and Advanced Micro Devices resume shipping China-compliant artificial intelligence chips — Nvidia’s H20 and AMD’s MI308 — after the companies agreed to give the U.S. government a 15% cut of the China sales, Aug. 11, 2025.
The unusual revenue-share — often referred to as the Trump Nvidia China deal — is drawing bipartisan criticism because it ties export-control approvals, typically framed as a national security tool, to direct federal revenue.
Inside the Trump Nvidia China deal and the 15% “cut”
Under the arrangement, the Commerce Department can issue export licenses for the H20 and MI308 while Nvidia and AMD remit 15% of the revenue from those China sales to the U.S. government, according to an Associated Press report.
Trump has defended the Trump Nvidia China deal as a way to keep U.S. firms competing in China while restricting the most advanced chips that Washington fears could boost Chinese military or surveillance capabilities. Trump has also described the H20 as “obsolete” compared with Nvidia’s latest offerings — language critics argue can obscure a basic point: Large volumes of “good-enough” accelerators can still be clustered into powerful systems.
What remains unclear is the plumbing. Neither the administration nor the companies have fully detailed how the 15% cut will be calculated, audited and enforced, or whether the same model could be applied to other products or exporters.
Why critics call it a dangerous precedent
Critics say the Trump Nvidia China deal risks turning export controls into a pay-to-export system, effectively creating a financial incentive to approve shipments that may otherwise be judged too risky. In an Aug. 15 letter to Trump, Rep. Raja Krishnamoorthi, the top Democrat on the House Select Committee focused on China, argued the arrangement violates the Constitution and federal statute and signals that U.S. national security is for sale.
A Tax Policy Center analysis made a similar argument, calling the 15% cut an export tax “by any name” and warning that tying revenue to licensing decisions could invite lobbying pressure, uneven enforcement and evasive routing through third countries.
What’s shipping — and why the chips matter
Nvidia’s H20 is a China-focused inference chip built to comply with U.S. limits, while AMD’s MI308 is a downgraded version of its Instinct MI300X line designed for the same purpose. Both were placed under tighter restrictions in April, and Reuters reported in December that AMD CEO Lisa Su said the company has licenses to ship some MI308 chips to China and is prepared to pay the 15% fee if it makes those shipments — a sign the Trump Nvidia China deal is moving from headline to compliance reality.
For critics, that reality is the concern: Even if the H20 and MI308 are less capable than the top-tier chips blocked from China, they can still support meaningful AI development at scale, especially when combined in data centers.
Background: a multiyear export-control chess match
The debate over the Trump Nvidia China deal sits on top of a longer pattern: Washington tightens rules, and companies redesign products. After the U.S. began restricting advanced chip exports in 2022, Nvidia moved quickly to offer China-specific variants such as the A800, as described in a 2022 Reuters report.
By 2023, U.S. officials were explicitly trying to close loopholes they believed allowed slightly downgraded chips to slip through, according to a 2023 Reuters report on the updated crackdown.
And by early 2024, Nvidia was preparing mass production of the H20 as its next China-focused option, Reuters reported in a 2024 story on the H20 rollout.
For now, the Trump Nvidia China deal leaves two questions hanging over every license: whether the arrangement withstands legal scrutiny, and whether a government “cut” becomes the new price of doing business in geopolitically sensitive markets.
