President Donald Trump announced this week that U.S. regulators will allow Nvidia to ship its H200 artificial‑intelligence chips to “approved customers” in China — provided the United States receives a 25% cut of the revenue. The move, disclosed on the president’s Truth Social feed and confirmed by multiple outlets, is a sharp turn from the tighter export curbs put in place under the previous administration and marks an attempt to thread a narrow economic and national‑security needle.
The deal, in plain terms
Under the outline Trump shared, Nvidia will be permitted to sell H200 units — more capable than the previously offered H20 — to vetted buyers abroad, including in China. The Commerce Department is said to be finalizing the rules, and the same framework would apply to other American chipmakers such as AMD and Intel.
Trump said Chinese leader Xi Jinping “responded positively” to the proposal. The White House frames the arrangement as a way to bolster U.S. manufacturing and taxpayers while keeping American technology at the center of global AI development. Exactly how the U.S. will collect the 25% — whether via an export levy, an enforced revenue-sharing mechanism, or another fiscal instrument — has not been published.
Markets reacted quickly: Nvidia shares ticked up on the news before settling, while investors and Washington policymakers began parsing the security and industrial implications.
Why this matters — and why it’s controversial
Semiconductors are the lever in the global AI race. Access to high‑end chips like the H200 can determine which companies and countries lead large‑scale model training and deployment. Washington’s earlier export restrictions were aimed at slowing China’s ability to field million‑chip AI clusters; proponents argued the limits bought time for U.S. firms and allies.
But the curbs also had side effects. Beijing doubled down on indigenous alternatives, and Chinese firms — from cloud groups to chip designers — have made notable advances. Companies such as DeepSeek and Alibaba, and hardware makers like Huawei, have narrowed parts of the gap. Meanwhile, dependence on foundries like Taiwan Semiconductor Manufacturing Company for advanced nodes left U.S. strategy with limited levers.
Some in the White House now argue the restrictions have largely failed to halt progress and instead ceded market share and influence over standards. Allowing controlled exports with a U.S. revenue share is being pitched as a pragmatic compromise: U.S. products stay in the market and U.S. authorities maintain a financial and regulatory role.
Roadblocks and realities
Approval on paper does not guarantee broad Chinese acceptance. Beijing earlier signaled resistance: when Washington initially pursued a softer compromise that would have let Nvidia ship the H20, China instructed local buyers to stop purchases over security concerns. Financial Times reporting since the announcement suggests Beijing may still limit access to foreign chips even if Washington authorizes sales — a reminder that political signaling and domestic industrial policy in China can outweigh commercial incentives.
There are also legal and political hurdles at home. Some lawmakers from both parties have pushed legislation to curb or delay sales of advanced AI hardware to China; national‑security hawks argue any loosening risks accelerating Chinese military or surveillance applications. Other officials see economic upside in letting American firms compete for revenue and shape global standards.
Industry mechanics matter too. Nvidia previously negotiated a 15% revenue‑share approach with Washington for China sales; the president’s 25% figure would be a steeper levy. How companies will price products, decide which customers qualify as “approved,” and manage supply‑chain constraints (including chip fabrication and rare‑earth elements) remains to be seen.
What this means for the AI ecosystem
Even as geopolitical friction plays out, major AI platforms and tools continue to expand rapidly. That intensifies the stakes of who controls the underlying compute. New product pushes and platform updates from cloud and AI providers underscore how integrated software and hardware battles have become; for example, Google is building deeper agentic features into its services, and generative models are spreading into productivity tools — trends that shape demand for high‑end accelerators like the H200. See coverage of Google’s AI Mode and how model integrations are reshaping workflows in products like Gemini Deep Research and other consumer-facing AI releases.
If China accepts H200 imports under tight vetting, U.S. companies could gain market share and influence over hardware‑software compatibility worldwide. If Beijing restricts uptake anyway, the policy risks becoming a symbolic concession with limited commercial payoff.
Where this could go next
Expect intense detail work at the Commerce Department: lists of approved customers, auditing and enforcement mechanisms, technical whitelists and blacklists, and perhaps exceptions for sensitive applications. Congress may attempt to legislate constraints or oversight. Chinese industrial policy will be decisive — Beijing can blunt the policy’s impact by tightening procurement rules, favoring domestic suppliers, or insisting on local testing and controls.
The announcement is a reminder that in the AI era, chips are as much geopolitical instruments as commercial products. The precise contours of how revenue‑sharing, export vetting, and industrial strategy will mesh are still unknown — and the answer will shape who gets to train the next generation of large models.
(Reporting synthesized from multiple outlets and public statements.)