Negotiations reshape export policy for AI chips
President Donald Trump said Monday that the U.S. government and Nvidia reached a revenue sharing arrangement tied to sales of the company’s China specific AI chip. According to the president, the White House first floated a 20% take of Nvidia’s China revenue, then settled at 15 percent after in person talks with CEO Jensen Huang. The agreement unlocks export control licenses that allow Nvidia to resume shipments of the H20 processor to Chinese customers, reversing a halt that began in April when licensing became mandatory. A White House official said Advanced Micro Devices will operate under the same 15 percent structure for its Instinct MI308 chip.
The deal signals a shift from blanket prohibitions toward conditional market access. Nvidia had designed the H20 to comply with the Biden era controls of 2023 by reducing performance relative to its high end H100 and H200 products. Those controls were intended to limit China’s access to the most advanced accelerators. The new license based approach lets the U.S. capture a share of revenue from approved shipments while keeping tighter restrictions on the latest hardware.
Security trade offs and the status of Blackwell
National security concerns remain central. U.S. officials have argued that unfettered access to cutting edge accelerators could speed Chinese military and AI capabilities. Trump contrasted the H20 with Nvidia’s next generation Blackwell family, describing Blackwell level performance as out of scope for unmodified export. He said any China bound variant would need significant performance reductions, on the order of one third to one half, before licenses would be considered. Nvidia has maintained that keeping Chinese AI developers tied to U.S. technology is strategically safer than forcing a rapid substitution with domestic silicon. Critics counter that any resumed shipments risk narrowing the technology gap the controls are meant to preserve.
The H20 sits in the middle of that debate. It is related to the H100 and H200 that power many U.S. data centers, but was tuned down for the China market to fit within earlier rules. Trump said China already has roughly comparable capability from local vendors, citing Huawei’s progress, which he argued diminishes the incremental risk of permitting licensed H20 exports while still drawing a bright line around Blackwell.
Market impact and corporate calculations
The license pathway restores access to a large and fast growing buyer base for AI chips in China. Before licenses were required this spring, Nvidia had expected about 8 billion dollars in H20 revenue for the July quarter. That forecast evaporated when sales were paused in May, and the company said the China market was effectively closed. Nvidia’s spokesperson told CNBC the company hopes clear rules will allow U.S. firms to compete in China and globally. For AMD, the 15 percent structure offers a similar mechanism to ship its MI308 while remaining within U.S. policy.
Investors will now assess how the 15 percent levy interacts with margins and volume. Semiconductor programs are planned years ahead, and the shift from prohibition to a royalty style model introduces a new variable for forecasting. The policy also creates a template that could be extended. Trump said Huang is likely to return to the White House to discuss licensing for a reduced performance version of Blackwell, which would test how far the framework can stretch without undermining the strategic aim of export controls.
What the deal signals for U.S.–China tech relations
The arrangement blends export licensing with a revenue capture tool that converts some China sales into U.S. government proceeds. Supporters see a pragmatic balance that preserves leverage, funds enforcement, and keeps U.S. suppliers embedded in the Chinese ecosystem. Skeptics view the policy as an opaque tax that may be hard to administer and could erode the deterrent value of controls. Either way, the agreement marks a notable evolution in semiconductor policy between the two countries, and its execution on the H20 and AMD MI308 will set expectations for any future talks on Blackwell or other advanced accelerators.
