Trump approves NVIDIA H200 chip sales to China with 25% US revenue cut
Dec 9, 2025
Key Points
- Trump approves NVIDIA H200 chip exports to China with the U.S. government claiming 25% of sales revenue, a major shift that permits higher-performance processors than previously allowed.
- NVIDIA could ship $2 billion to $5 billion in chips quarterly to China if volumes materialize, though stock reaction was muted, suggesting markets already priced in the approval.
- The deal assumes NVIDIA's software ecosystem creates lasting vendor lock-in, but the Tesla and solar panel precedents suggest China can leapfrog U.S. technology once it gains market access.
Summary
Trump has approved NVIDIA's H200 chip exports to China with the U.S. government taking a 25% cut of sales, opening the world's second-largest economy to higher-performance chips than previously permitted.
The H200 is NVIDIA's second-best processor, more powerful than the H20 that was previously approved for export but still a generation behind Blackwell, NVIDIA's flagship line. NVIDIA had earlier agreed to give the U.S. 15% of China sales from the lower-performing H20 before those plans collapsed during trade negotiations. This new deal emerged after a meeting between Trump and NVIDIA CEO Jensen Huang.
If volumes materialize, the move is substantial. NVIDIA's CFO said in August the company could ship between $2 billion and $5 billion in chips to China per quarter if geopolitical barriers lifted. At NVIDIA's current 73.4% gross margins, even a 25% government cut leaves the company with strong unit economics. Yet the market has largely priced in the development—NVIDIA stock was down slightly on the news despite the approval.
NVIDIA has argued the company should compete in China because the country hosts many top AI researchers and the U.S. benefits from their using American technology. The idea, as NVIDIA's founder articulated months ago, is to make developers "addicted" to the NVIDIA software ecosystem so they stay locked in even if China develops domestic alternatives.
But this dependency assumption faces a real test. The Tesla precedent shows the risk: the U.S. exported hundreds of thousands of Teslas to China, and Chinese automakers including BYD, Huawei, and Xiaomi have since built superior EVs and are now competitive globally. If China leapfrogged in cars, why assume semiconductors will be different? The same pattern held for solar panels. The U.S. flooded its own market with cheap Chinese panels and killed domestic manufacturing.
One counterargument exists: exporting cheap NVIDIA chips raises the cost for China to fund uneconomical domestic alternatives like Huawei's chips. If NVIDIA supplies dominate the market, Beijing has to spend more to keep a local industry alive. But this assumes Beijing will hesitate to spend money on strategic independence, an assumption that collapses if the Chinese government views chip sovereignty as a national priority worth the cost. It does.
U.S. chip export policy now hinges on whether fast-takeoff AGI is imminent and requires controlling compute flows. If that framing is correct, denying chips to China delays a rival's path to artificial general intelligence. If AGI is less certain, the cost-benefit of export restrictions becomes murkier. Trump's approval suggests the administration is skeptical of the Manhattan Project framing, at least for now.
The deal also applies to AMD and Intel, which now has a 10% government stake, signaling a broader opening. It comes weeks after officials including Secretary of State Marco Rubio blocked NVIDIA's push to export a stripped-down Blackwell chip, showing internal debate persists over where to draw the line.