DeepSeek allegedly trained V4 model on banned Nvidia Blackwell chips, denied US chipmakers access

DeepSeek Allegedly Used Nvidia Blackwell Chips Despite US Export Controls

Chinese AI startup DeepSeek has ignited fresh controversy after a senior Trump administration official claimed the company trained its upcoming V4 AI model on Nvidia Blackwell chips, potentially violating U.S. export controls2

2

. The official stated that the Blackwell chips are likely part of a cluster at DeepSeek's data center in Inner Mongolia, an autonomous region of China, and that the company plans to remove technical indicators revealing its use of American AI chips2

2

. Current US policy explicitly bars Blackwell shipments to China, with the official emphasizing: "We're not shipping Blackwells to China"3

3

. The Commerce Department oversees these restrictions as part of broader efforts to prevent advanced semiconductor technology from enhancing China's military capabilities and threatening US dominance in the AI race.

DeepSeek Denied Nvidia and AMD Early Access to V4 Model

In a departure from standard industry practice, DeepSeek withheld its forthcoming flagship model from US chipmakers including Nvidia and AMD, according to two sources familiar with the matter1

1

. Instead, the lab granted early access to domestic suppliers, including Huawei, giving Chinese chipmakers a head start of several weeks to optimize the software for their processors1

1

. AI developers typically share pre-release versions of major models with leading chipmakers to ensure their software performs efficiently on widely used hardware through performance optimization. DeepSeek has previously worked closely with Nvidia's technical staff, making this exclusion particularly notable. Ben Bajarin, CEO of research firm Creative Strategies, suggested the move is likely part of a broader strategy by the Chinese government "to try to keep U.S. hardware and models disadvantaged" in China1

1

.

U.S.-China AI Tensions Escalate Over Access to Advanced AI Chips

The revelations have intensified U.S.-China AI tensions and deepened the divide among Washington policymakers struggling to determine appropriate limits on Chinese access to advanced AI chips2

2

. China hawks fear chips could be diverted from commercial uses to supercharge China's military, while White House AI Czar David Sacks and Nvidia CEO Jensen Huang argue that shipping advanced AI chips to China discourages Chinese competitors like Huawei from redoubling efforts to catch up with American technology2

2

. President Trump's shifting positions reflect this internal conflict. In August, he opened the door to Nvidia selling a scaled-down Blackwell version in China, but later reversed course, suggesting the firm's most advanced chips should be reserved for US companies2

2

. His December decision to allow Chinese firms to buy H200 chips drew sharp criticism from China hawks, though shipments remain stalled over guardrails built into the approvals2

2

.

Model Distillation and Competitive Implications for US AI Companies

The Trump administration official also alleged that DeepSeek's model likely relied on model distillation of models made by leading US AI companies, including Anthropic, Google, OpenAI, and xAI. This technique involves having an older, more established AI model evaluate the quality of answers from a newer model, effectively transferring the older model's learnings. These allegations echo claims previously made by OpenAI and Anthropic. DeepSeek's models have been downloaded more than 75 million times on Hugging Face since the company emerged in January 2025, helping fuel a wave of Chinese open-source models competing with US AI labs1

1

. Among models released in the past year, downloads for Chinese models have surpassed those from any other country on the platform.

Impact on Nvidia and Market Implications

Despite export uncertainty, Nvidia continues to benefit from robust global demand for AI inference and training capabilities. JPMorgan analyst Harlan Sur noted that US export approvals could unlock billions in additional revenue for Nvidia, with the company disclosing a backlog exceeding $500 billion as hyperscalers invest heavily in AI infrastructure4

4

. AMD also reported significant demand, with its MI308 chip generating $390 million in sales in its most recent quarter1

1

. However, the potential for regulatory scrutiny following these revelations could create risks for investor sentiment. The Chinese embassy in Washington responded by stating that Beijing opposes "drawing ideological lines, overstretching the concept of national security, expansive use of export controls and politicizing economic, trade, and technological issues". As DeepSeek prepares to launch its V4 model, the incident highlights how the AI race between the US and China continues to reshape global semiconductor markets and challenge existing export control frameworks.