Chinese AI startup DeepSeek reportedly trained its latest AI model on Nvidia’s top Blackwell chips, raising U.S. export control concerns.
![]() |
| DeepSeek’s reported training of AI models on Nvidia’s most advanced chips highlights tensions over U.S.-China technology access and national security concerns. Image: CH |
Tech Desk — February 24, 2026:
Chinese AI startup DeepSeek reportedly trained its latest AI model using Nvidia’s most advanced AI chip, the Blackwell, in what a senior Trump administration official said could constitute a violation of U.S. export controls. The model is expected to be released as soon as next week, and the development comes amid heightened scrutiny over China’s access to American AI semiconductor technology.
According to the official, DeepSeek likely removed technical indicators that could reveal the use of American chips and may have clustered the Blackwells at its data center in Inner Mongolia. The source declined to specify how the U.S. obtained the information or how DeepSeek acquired the chips but emphasized that U.S. policy prohibits shipping Blackwells to China. Nvidia, the Commerce Department, and DeepSeek did not comment on the report. The Chinese embassy in Washington, however, condemned “drawing ideological lines” and criticized the politicization of trade and technology issues.
The news could intensify debates in Washington over Chinese access to advanced AI chips. Policymakers who favor stricter controls worry that high-end chips could be diverted from commercial uses to military applications, potentially challenging U.S. dominance in AI. Others, including White House AI Czar David Sacks and Nvidia CEO Jensen Huang, argue that allowing access to advanced chips may discourage Chinese firms like Huawei from accelerating efforts to compete with U.S. technology.
U.S. export controls currently bar Blackwell shipments to China. Last August, former President Donald Trump briefly allowed Nvidia to sell a scaled-down version of Blackwell chips in China but reversed the decision, reserving the most advanced technology for U.S. companies. A December decision permitted sales of Nvidia’s H200 chips to Chinese firms, but shipments remain stalled due to guardrails embedded in the approvals process.
Experts say DeepSeek’s reliance on smuggled Blackwells underscores China’s limited domestic AI chip production and explains why H200 approvals could provide a lifeline. Saif Khan, former director of technology and national security at the White House’s National Security Council under President Joe Biden, noted that the incident highlights China’s shortfall in homegrown AI chips.
The DeepSeek model reportedly incorporated “distillation” techniques using established AI models from U.S. companies such as Anthropic, Google, OpenAI, and xAI. Distillation allows newer models to learn from older, more powerful AI models by evaluating their outputs, effectively transferring knowledge.
Hangzhou-based DeepSeek previously shook markets early last year with AI models that rivaled top U.S. offerings, sparking concerns that China could narrow the AI gap despite restrictions. The reported use of Nvidia’s Blackwell chips adds urgency to ongoing debates over export controls, U.S.-China technology competition, and the strategic risks posed by advanced AI capabilities.
This development underscores the growing tension between fostering global AI innovation and protecting sensitive U.S. technologies from potential misuse abroad. Policymakers are now faced with balancing economic, technological, and national security considerations as China continues to expand its AI capabilities.
