Nvidia and Advanced Micro Devices agreed in August 2025 to give the U.S. government 15% of revenue generated from certain artificial-intelligence chip sales to China as part of arrangements tied to export licenses, according to a U.S. official and reporting by Reuters. The arrangement covered Nvidia's H20 and AMD's Instinct MI308 processors.
The distinction between revenue and profit is important. Despite the wording of the existing article title, the reported arrangement concerned 15% of sales revenue, not 15% of the companies' profits. That means the government's potential share would be calculated before accounting for the companies' manufacturing, operating and other costs.
How the 15% Revenue Arrangement Emerged
The agreement followed a rapid shift in U.S. export policy toward AI accelerators designed for the Chinese market. In April 2025, the U.S. government imposed new license requirements on Nvidia's H20 and AMD's MI308, sharply restricting shipments to China.
AMD disclosed at the time that the restrictions applied to its MI308 products and could result in approximately $800 million in inventory, purchase-commitment and related charges. Nvidia likewise faced a substantial financial impact from restrictions on the H20, a processor specifically configured to comply with earlier U.S. export-control limits.
By July, however, Nvidia said the U.S. government had indicated that licenses for H20 sales to China would again be issued. The actual licensing process remained central to whether shipments could resume.
In early August, Nvidia Chief Executive Jensen Huang met President Donald Trump at the White House. Shortly afterward, the Commerce Department began issuing licenses for some AI-chip exports. Reuters reported that Nvidia and AMD agreed to provide the U.S. government with 15% of revenue from licensed H20 and MI308 sales to China.
The structure was unusual because export controls traditionally determine whether sensitive technology may be shipped based on licensing and national-security criteria rather than giving the government a direct percentage of private-sector export revenue.
What the Agreement Could Mean Financially
The sums potentially involved were significant. Bernstein Research estimated in mid-2025 that China's demand for AI accelerators remained extremely large even after the U.S. restrictions disrupted supply. Contemporary estimates cited in coverage of the agreement suggested Nvidia could generate more than $15 billion from H20 sales in China through the remainder of 2025, while AMD could generate roughly $800 million from MI308 sales.
If those sales estimates had been fully realized and all qualifying revenue had been subject to the 15% arrangement, the government's share could have exceeded $2 billion. Those figures were forecasts rather than guaranteed sales, however, and actual revenue depended on licenses, customer demand, supply availability and Chinese regulatory decisions.
The arrangement also mattered because China represented a substantial semiconductor market for both companies. Nvidia reported billions of dollars in revenue associated with China in previous fiscal periods, while AMD said restrictions on its MI308 product had already created a material financial impact.
Export Controls Versus Market Access
The debate surrounding the H20 and MI308 reflected two competing policy considerations. One was the U.S. objective of limiting China's access to computing technology that could accelerate advanced AI development. The other was the argument that shutting American suppliers out of China could strengthen domestic Chinese chipmakers and reduce the long-term influence of U.S. technology platforms.
Huang repeatedly argued that American companies should remain able to compete for legitimate commercial AI demand in China while the United States continues blocking exports of its most advanced processors. Nvidia has said broad adoption of the American computing ecosystem can reinforce U.S. technology leadership by keeping developers and businesses dependent on American hardware and software.
Critics of that approach have argued that even processors below Nvidia's most advanced performance tier could materially expand Chinese AI-computing capacity. Former national-security officials and members of Congress raised concerns that linking export permission to government revenue could weaken the principle that controls on sensitive technology should be based primarily on security considerations.
Those disagreements became more pronounced because the H20 was not Nvidia's highest-end accelerator. The processor used the older Hopper architecture and was designed to meet earlier U.S. restrictions, while Nvidia's more powerful Blackwell-generation products remained subject to tighter controls.
AMD Faced Similar Restrictions on the MI308
AMD's position broadly paralleled Nvidia's. In an April 2025 regulatory filing, AMD said the U.S. government had imposed a new license requirement on exports of its MI308 processors to China and certain other destinations.
The company later recorded about $800 million in inventory and related charges associated with those restrictions. AMD subsequently applied for and received some U.S. licenses allowing shipments to certain China-based customers.
AMD's later regulatory filings added important context to the original announcement. The company said U.S. officials had expressed an expectation in August 2025 that the government would receive 15% of revenue generated by licensed MI308 sales to China. However, AMD also stated that the U.S. government had not published a regulation codifying that requirement.
That disclosure shows why the arrangement should not be treated as a conventional, permanently established export tax. The companies were operating within a licensing framework whose exact implementation and legal structure remained subject to uncertainty.
A New Link Between Export Licensing and Government Revenue
The reported Nvidia and AMD agreements stood out because direct government participation in export revenue is rare in U.S. technology policy. The administration was simultaneously taking a broader role in major industrial and trade transactions, including the government's "golden share" associated with Nippon Steel's investment in U.S. Steel.
Trump was also using tariffs and tariff threats to encourage semiconductor manufacturing in the United States. In August 2025, he said imported semiconductors could face tariffs of around 100% unless companies committed to producing chips domestically, although the eventual structure of U.S. semiconductor tariffs continued to develop afterward.
For Nvidia and AMD, meanwhile, the immediate commercial question was straightforward: access to China could preserve billions of dollars in potential accelerator sales, even if part of that revenue had to be shared with the government and even if individual customers still required export approval.
National-Security Critics Questioned the Precedent
The revenue-sharing arrangement drew criticism from national-security specialists who opposed loosening restrictions on AI accelerators sold to China. Their concern was not only the capability of the H20 or MI308 but also the precedent created by attaching a financial payment to export approval.
Critics argued that export controls could become harder to defend as strictly security-based measures if companies could effectively secure market access while the government received part of the resulting revenue. Supporters of allowing limited sales countered that maintaining U.S. suppliers' presence in China could slow the displacement of American technology by domestic competitors such as Huawei.
That dispute was reflected in congressional reaction as well. Lawmakers from both major parties raised questions about whether the unusual structure could blur the line between national-security regulation and revenue generation.
China Raised Its Own Security Concerns About Nvidia's H20
The regulatory pressure did not come solely from Washington. Chinese authorities also increased scrutiny of Nvidia's H20 after the U.S. licensing reversal.
China's Cyberspace Administration questioned Nvidia about concerns that advanced chips might contain mechanisms capable of tracking or remotely controlling hardware. Chinese state media also warned companies about possible security concerns surrounding the H20.
Nvidia strongly rejected the suggestion that its processors contain hidden remote-control mechanisms. Last week, Nvidia issued a blog post saying its GPUs do not contain backdoors, kill switches or spyware.
The company argued that deliberately embedding secret hardware controls would create cybersecurity vulnerabilities rather than improve security.
Later Disclosures Show the 15% Requirement Remained Unsettled
Subsequent company disclosures provide useful context that was not available when the arrangement was first reported. Nvidia said during its August 2025 earnings update that U.S. officials had expressed an expectation that the government would receive 15% of revenue from licensed H20 sales, but that no regulation had yet been published codifying such a requirement.
AMD later used similar language in its regulatory filings. Its 2026 annual report said it had eventually begun shipping MI308 products to certain China-based customers after receiving licenses, while noting that the government still had not published a regulation establishing the 15% requirement.
That later information does not change what was reported in August 2025: both chipmakers had agreed to an extraordinary revenue-sharing framework connected to China export licenses. It does, however, show that the legal and regulatory implementation was less settled than the initial announcement might have suggested.
Why the Nvidia-AMD Deal Matters
The episode illustrates how tightly commercial strategy, industrial policy and national security have become linked in the global AI-chip market.
For Nvidia and AMD, China represents significant potential demand. For U.S. policymakers, advanced accelerators are also strategic technologies whose availability can affect the pace and cost of AI development. For China, dependence on American processors creates an incentive to accelerate domestic alternatives.
The 15% arrangement therefore involved more than an unusual government revenue stream. It became part of a larger debate over whether the United States can preserve the global reach of its semiconductor industry while restricting access to technologies considered strategically sensitive.
What remained uncertain was how durable the revenue-sharing model would become, how broadly future export licenses would be granted, how Chinese customers would respond, and whether the arrangement would survive legal and regulatory scrutiny. Those factors ultimately determine how much revenue Nvidia, AMD and the U.S. government could actually receive from the policy.
0 Comments