US officials are looking into whether Chinese AI company DeepSeek might have trained its R1 chatbot on Nvidia GPUs acquired through third-party companies in Singapore.
As reported by Bloomberg, the investigations looking into whether the AI company may have circumvented US restrictions on the sales of chips to train its model which has thrown the US AI market into flux.
Earlier this week, DeepSeek claimed that it had trained its R1 chatbot on its -V3 LLM which cost the company just $5.6 million and using far less advanced chips (Nvidia H800 chips) than US companies have access to. Following the announcement, shares in the likes of Nvidia, Microsoft and Meta rapidly fell, as their comparative training costs were reaching the billions.
The validity of DeepSeek's claims has been brought into question, however. OpenAI has claimed that it has evidence that DeepSeek used its models to help train the cheaper, more efficient -V3 model.
Furthermore, the $5.6m cost was based on estimated GPU rental figures and did not include the company's own compute, and only looked at the one training run.
Anthropic's CEO Dario Amodei said that DeepSeek-V3 was not as cheap as thought, and was not something that "fundamentally changes the economics of LLMs."
Now, officials in the White House and Federal Bureau of Investigation in the US are reportedly investigating whether the company actually used more expensive and powerful GPUs of the ilk that are not permitted for use by Chinese companies. by buying or leasing them from companies in Singapore.
An Nvidia spokesperson told Bloomberg in a statement that “we insist that our partners comply with all applicable laws, and if we receive any information to the contrary, act accordingly.”
DCD has contacted DeepSeek for comment.
Howard Lutnick, President Donald Trump’s choice to lead the Commerce Department, agreed with the suggestion that the company may have found a way around the restrictions. "Nvidia’s chips, which they bought tons of, and they found their ways around it, drive their DeepSeek model,” he said.
Singapore represents a significant portion of Nvidia's revenue - around 20 percent according to regulatory filings - but the majority of shipments "associated with Singapore revenue were to locations other than Singapore, and shipments to Singapore were insignificant.”
