In a major shift in U.S.–China tech policy, President Trump has announced a new framework allowing limited sales of Nvidia’s H200 AI chips to China. The decision blends commercial access with unprecedented government oversight, marking one of the most unusual arrangements ever imposed on U.S. technology exports.
The policy introduces a 25% revenue-sharing fee, mandatory buyer vetting, and tight restrictions on which chip models qualify for export. The move is widely viewed as a compromise between safeguarding national security and supporting U.S. semiconductor competitiveness in the world’s largest AI market.
1. Key Details of the New Policy
Nvidia H200 GPUs Approved — But Not the Newest Models
The authorization applies only to Nvidia’s H200 GPU, a high-performance chip from the earlier Hopper generation.
It does not include Nvidia’s latest and most advanced AI processors, such as:
- Blackwell (B-series)
- Rubin (R-series, upcoming)
This ensures China gains access to capable AI hardware without reaching the cutting edge of U.S. innovation.
25% Revenue Cut Collected by the U.S. Government
A defining feature of the policy is a 25% U.S. government cut of total sales revenue for these chips.
A White House official clarified that the fee will function as an import tax, charged when chips arrive in the U.S. from manufacturing hubs (like Taiwan) for mandatory security inspection before export.
This creates a direct financial link between semiconductor sales to China and the U.S. Treasury—a rare mechanism in export controls.
Sales Allowed Only to “Approved Customers”
The U.S. Department of Commerce will vet all Chinese buyers.
A license will be required for each transaction, ensuring:
- No military or restricted entities receive the chips
- All exports comply with national-security frameworks
- The U.S. retains oversight of AI-relevant technology inflow into China
AMD and Intel Included
President Trump confirmed that the same 25% rule and vetting process will extend to other U.S. chipmakers, including:
- AMD (MI series AI accelerators)
- Intel (Gaudi and future AI processors)
This harmonizes restrictions across the U.S. semiconductor sector.
2. Strategic and Political Implications
Higher Revenue Share Than Previous Agreements
The 25% share is significantly higher than the 15% revenue split that Nvidia and AMD had previously agreed to for their China-specific chips, such as:
- Nvidia H20
- AMD MI309X (China edition)
The new policy represents a stronger assertion of U.S. control over advanced chip exports.
A Delicate U.S. Compromise
The decision attempts to balance two competing priorities:
✔ National Security
Restrict access to America’s most advanced AI hardware.
✔ Economic Competitiveness
Allow U.S. chipmakers access to the multi-billion-dollar Chinese AI market.
Limiting exports to H200-level chips gives China powerful but not cutting-edge hardware—intended to preserve a multi-generation technology gap.
China’s Reaction
President Trump stated on Truth Social that Chinese President Xi Jinping “responded positively.”
However, analysts note:
- China may limit purchases to avoid long-term dependence on U.S. tech
- Beijing may accelerate support for domestic chipmakers like Huawei Ascend and Cambricon
- Increased performance capacity could boost China’s near-term AI development
Performance Gap: H200 vs. H20
Reports indicate the H200 is nearly six times more powerful than the H20, the previously most advanced chip legally exportable to China.
This marks a substantial upgrade in capability though it still remains several generations behind Blackwell and Rubin.
3. A Highly Unusual Policy in U.S. Trade History
The 25% revenue-sharing tax represents a groundbreaking mechanism in foreign technology sales:
- It resembles royalty sharing more than a traditional export control
- It ensures the U.S. directly profits from China’s AI hardware demand
- It adds a new economic dimension to tech-national-security policy
Experts say this hybrid model could influence future tech regulations involving key industries like quantum computing, synthetic biology, and cybersecurity hardware.
0 Comments