US Accusations Mount Against DeepSeek: Chinese AI Firm Allegedly Trained New Model on Banned Nvidia Blackwell Chips
Published Date: 24th Feb, 2026
February 24, 2026
The United States escalated accusations against Chinese artificial intelligence startup DeepSeek today, claiming the company trained its forthcoming advanced AI model on Nvidia's restricted Blackwell series processors, in what Washington describes as a clear breach of longstanding export controls designed to curb China's access to cutting-edge semiconductor technology.
A senior Trump administration official told Reuters that DeepSeek's latest model, expected to launch as early as next week, utilized Nvidia's most powerful Blackwell chips for training. The official asserted that technical markers indicating American chip usage could be stripped out before release to conceal the violation, with the hardware likely clustered at a data center in Inner Mongolia.
Export Bans Under Spotlight
US Commerce Department rules explicitly prohibit shipments of Blackwell GPUs to China, a policy rooted in national security concerns over potential military applications of advanced AI. The administration emphasized that no such chips are officially exported to the country, raising questions about how DeepSeek obtained them. Officials declined to detail intelligence sources or acquisition methods but stressed the incident highlights enforcement gaps in supply chain oversight.
DeepSeek has not responded to requests for comment, and Nvidia declined to address the specific allegations. The company has previously denied knowledge of illicit transfers and affirmed full compliance with US regulations.
Pattern of Scrutiny on DeepSeek
This latest claim adds to mounting pressure on the Hangzhou-based firm. Congressional investigations earlier this year revealed Nvidia provided substantial technical assistance to DeepSeek, including optimizations that boosted training efficiency for models like DeepSeek-V3. Critics argue such support, even if initially commercial, indirectly aided capabilities now raising alarms.
Additional reports suggest DeepSeek employed indirect channels, including shell companies in Southeast Asia and remote access to restricted hardware clusters, to bypass sanctions. The firm's ability to produce high-performing models with seemingly limited resources has fueled speculation about undisclosed compute power, challenging assumptions that massive Western-scale infrastructure is essential for frontier AI.
Broader Geopolitical and Market Ripples
The controversy unfolds against intensifying US-China rivalry in artificial intelligence, where export controls aim to preserve American technological edge. Recent approvals for downgraded Nvidia chips to China sparked internal debates, yet Blackwell restrictions remain firm. Bipartisan lawmakers have pushed measures to block DeepSeek tools from federal systems over data security and intellectual property risks.
Market observers note potential volatility ahead of Nvidia's upcoming earnings, with the allegations underscoring persistent demand for restricted hardware alongside heightened policy risks. DeepSeek's efficient, cost-effective models have already disrupted expectations, prompting fears that sanctions may accelerate Chinese self-reliance rather than contain progress.
Implications for the Global AI Landscape
As DeepSeek prepares its next release, the accusations spotlight vulnerabilities in controlling dual-use technologies. If substantiated, the case could prompt tighter enforcement, expanded blacklists, or new legislative efforts targeting evasion tactics. For now, the episode reinforces the high-stakes nature of the AI race, where breakthroughs in one nation provoke swift countermeasures in another, with profound consequences for innovation, security, and global competition.
Date: 24th Feb, 2026
