The U.S. government has confirmed that Chinese AI startup DeepSeek trained its latest artificial intelligence model using Nvidia’s most advanced chip, the Blackwell, in what officials say could represent a violation of American export controls.
Blackwell Chips in Inner Mongolia
A senior Trump administration official said on Monday, February 23, 2026, that the Blackwell chips are likely clustered at DeepSeek’s data center in Inner Mongolia. The official added that the company may attempt to remove technical indicators that would reveal its use of American hardware.
The U.S. maintains a strict policy barring shipments of Blackwell chips to China. Nvidia declined to comment, while the Commerce Department and DeepSeek did not respond to requests for comment.
Beijing’s Response
The Chinese embassy in Washington criticized the U.S. stance, saying Beijing opposes “drawing ideological lines, overstretching the concept of national security, expansive use of export controls and politicizing economic, trade, and technological issues.”
Policy Divide in Washington
Confirmation of DeepSeek’s access to Blackwell chips, first reported by Reuters, has deepened divisions among U.S. policymakers.
- White House AI Czar David Sacks and Nvidia CEO Jensen Huang argue that allowing limited shipments of advanced chips discourages Chinese firms like Huawei from accelerating domestic alternatives.
- China hawks, however, warn that chips could be diverted to military use, undermining U.S. dominance in AI.
“This shows why exporting any AI chips to China is so dangerous,” said Chris McGuire, a former National Security Council official.
Export Controls and Chip Access
U.S. export controls overseen by the Commerce Department currently bar Blackwell shipments to China. President Donald Trump initially opened the door to selling a scaled-down version of the chip in August 2025 but later reversed course, reserving the most advanced chips for U.S. firms.
In December, Trump allowed Chinese companies to buy Nvidia’s H200 chips, the second most advanced, though shipments remain stalled due to regulatory guardrails.
DeepSeek’s AI Development
The official said DeepSeek’s model likely relied on “distillation” techniques, drawing on learnings from leading U.S. AI models developed by Anthropic, Google, OpenAI, and xAI.
DeepSeek, based in Hangzhou, shook markets last year with AI models rivaling top U.S. offerings, raising concerns in Washington that China could close the gap despite restrictions.
Conclusion
The revelation that DeepSeek trained its new model on Nvidia’s Blackwell chips underscores the challenges of enforcing export controls in the global AI race. As Washington debates how far restrictions should go, the case highlights both China’s reliance on U.S. technology and the geopolitical stakes of advanced semiconductor access.
