In a significant shift in U.S. technology export policy, the Trump administration has announced a partial relaxation of restrictions on advanced AI chips bound for China. The move allows Nvidia to resume sales of its H200 AI accelerator to approved Chinese customers, marking a potential thaw in a high-stakes tech trade dispute that has reshaped the global semiconductor landscape for over a year.
The Policy Shift and Its Conditions
On December 8, 2025, U.S. President Donald Trump announced via his Truth Social platform that Nvidia would be permitted to deliver its H200 chips to "qualified customers" in China and other countries, provided U.S. national security is assured. This decision represents a notable, albeit limited, reversal of the stringent export controls imposed earlier in the year. However, the approval comes with significant caveats. The newer, more powerful Blackwell and upcoming Rubin architectures remain strictly off-limits for export to China. Furthermore, the policy extends to other major chipmakers like AMD and Intel, applying similar rules across the industry. A particularly novel condition requires Nvidia to remit 25% of the sales revenue from these H200 exports to the U.S. government, a measure Trump stated would "support U.S. jobs, enhance American manufacturing strength, and benefit U.S. taxpayers."
Key Chip Models in the U.S.-China Export Context:
| Chip Model (Nvidia) | Status for Export to China (as of Dec 2025) | Key Note |
|---|---|---|
| H200 | Approved for "qualified customers" with conditions | Launched Q4 2023. Performance superior to H20. May be restricted to ~18-month-old versions. |
| H20 | Previously approved (special China variant) | A purposefully downgraded chip created to comply with earlier U.S. restrictions. |
| Blackwell | Not Approved for export | Nvidia's next-generation architecture after Hopper (H200). |
| Rubin | Not Approved for export | Nvidia's announced future architecture following Blackwell. |
| The same export rules are stated to apply to competitors like AMD and Intel. |
Nvidia's Strategic and Financial Imperative
For Nvidia, this policy adjustment is a critical, hard-won concession. The company's financial performance has been stellar, yet its inability to sell in the massive Chinese market has been a persistent headache. Following the April 2025 export restrictions, CEO Jensen Huang noted the company's sales in China plummeted from representing 95% of its data center revenue in the region to virtually zero. By the third quarter of 2025, AI chip sales in China had dwindled to a mere USD 50 million, accounting for just 0.09% of Nvidia's total revenue. Huang has been a vocal critic of the broad export bans, arguing that they cede the market to competitors and ultimately spur China to develop its own alternatives faster, thereby undermining long-term U.S. leadership in AI. The approval for H200 sales, even with restrictions, offers a vital pathway to recapture a portion of this crucial market.
Financial Impact on Nvidia's China Business:
- Q3 2025 China AI Chip Sales: ~USD 50 million
- Share of Total Nvidia Revenue (Q3 2025): 0.09%
- Historical Context (per CEO Jensen Huang): China sales fell from ~95% of regional data center revenue to "virtually 0" after April 2025 restrictions.
- New Export Condition: Nvidia must pay 25% of H200 sales revenue from approved Chinese exports to the U.S. government.
The H200's Position in a Competitive Landscape
The H200, launched in late 2023, occupies a unique position under the new rules. It is a generation ahead of the purpose-built, downgraded H20 chip that Nvidia previously created for the Chinese market, offering superior performance for both AI training and inference tasks. While reports from Semafor suggest the U.S. may only permit the export of H200 chips that are approximately 18 months old, its technical capabilities still represent a significant offering. Its re-entry comes at a time when Chinese domestic AI chips have made substantial progress. The gap in single-card computing power between local alternatives and the H200 persists, but it has narrowed considerably due to intense R&D efforts spurred by U.S. sanctions. The final adoption of the H200 in China remains uncertain, as Huang himself has expressed doubt about whether Chinese clients would accept it, emphasizing that they would not accept a deliberately degraded product.
Geopolitical and Industry Ramifications
The decision is more than a commercial adjustment; it is a carefully calibrated geopolitical move. The Chinese Foreign Ministry responded to the announcement by reiterating its stance that China and the U.S. should "achieve mutual benefit and win-win results through cooperation." The policy appears to strike a balance between applying economic pressure, supporting a key U.S. industry, and managing the risk of accelerating China's indigenous chip capabilities. It follows Nvidia's successful lobbying to block the inclusion of the more restrictive GAIN AI Act in the annual U.S. National Defense Authorization Act just last week. The requirement for a 25% revenue share is unprecedented and sets a new template for how the U.S. might seek to directly monetize and control strategic technology exports in the future.
The Path Forward and Market Uncertainty
While Trump's social media post sets the direction, the practical implementation falls to the U.S. Department of Commerce, which must formalize the export licenses. Even with U.S. approval, each sale will still require validation from Chinese authorities. The timeline for actual shipments and the scale of demand are open questions. The AI chip market is dynamic, and 18 months is a long time in semiconductor development. Chinese tech firms, having adapted to life without the latest U.S. chips, must now weigh the benefits of integrating the H200 against their ongoing investments in domestic supply chains. For Nvidia and its shareholders, the reopening of the Chinese market, even partially, is a welcome development, but one shrouded in regulatory complexity and competitive uncertainty that will define the AI hardware race for years to come.