Howard Lutnick, the candidate for the US Commerce Secretary, has accused DeepSeek, an AI company based in Hangzhou, of stealing US technology and bypassing US export controls to receive expensive Nvidia chips.  ,
Lutnick claimed that because it had access to a sizable supply of Nvidia chips and could take data from Meta’s open system, DeepSeek may make its AI models “dirt inexpensive” during the election hearing with the US Senate on January 29.  ,
” I take a really jaded view of China”, he said. They only consider themselves and attempt to harm us, so we must defend ourselves. We need to pull our creativity, and we need to cease helping them. DeepSeek can rely on Meta’s empty platform because it is available. Nvidia’s cards – which they bought lots of, and they found their way around ]export handles ] – drive their DeepSeek design. It’s got to stop”.
Lutnick promised to coordinate and give the Bureau of Industry and Security ( BIS )’s export controls authority to stop China from using American tools in a US market place.
Beijing is celebrating the Chinese New Year from January 28 to February 4, but it has so far no responded to Lutnick’s responses.  ,
Former PayPal CEO David Sacks, an expert on AI and cryptocurrency issues for the White House, claimed there was” substantial information” that DeepSeek drew data from Microsoft’s OpenAI designs for its own use.  ,
IT specialists said “distillation” or , “knowledge distillation” is widely used in Artificial education. It is a method by which output from a larger AI unit are used to teach and enhance a smaller one.
DeepSeek, in this process, can be understood as a student who keeps asking questions to a competent teacher, for instance ChatGPT, and uses the responses to fine-tune its logic. At some point, DeepSeek will be as bright as ChatGPT.  ,
The “distillation” process requires far less processing power than what OpenAI has used to teach ChatGPT.  ,
According to OpenAI, it had seen some proof that DeepSeek may have accessed its information through “distillation,” according to the Financial Times. It criticized DeepSeek for breaking its intellectual property.
Some Chinese IT experts agree that DeepSeek was created through “distillation” . ,
Wang Zhiyuan, a Beijing-based IT journalist, writes in an essay that it’s clear DeepSeek V-3, released on December 26, 2024, had used the “distillation” approach in education. He claims that he came to that conclusion after examining the DeepSeek features.  ,
He claims that many other Chinese AI types have used distilled information from the released on September 12th, 2013, ChatGPT o1. He claims that a group of Chinese experts published an academic report on November 25, 2024, that has already thoroughly described the extraction process and its performance.
He claims that a distilled data AI model may not be able to answer quite challenging questions but is sufficient to solve high school-level issues. In his opinion, all little AI models may develop distilled information before entering the market.
” Don’t laugh at those who took a short-cut”! Wang says. ” DeepSeek employed a unique approach to conserve computing power. After all, its education value is just US$ 5.58 million, 1.1 % of US$ 500 million of Meta’s Llama 3.1″.
A group of DeepSeek experts published a report on January 22 after the release of the DeepSeek-R1 on January 20, 2025, in which they claimed its most recent AI design achieves performance similar to ChatGPT-o1.
They said the training of DeepSeek-R1 used the distilled data from Alibaba’s Tongyi Qianwen ( Qwen ) and Meta’s Llama. They said the DeepSeek-R1-Distill-Qwen type outperforms ChatGPT-4o.  ,
50, 000 H100 cards?  ,
DeepSeek said it used just 2, 000 models of Nvidia’s H800 cards to teach its AI design. Before the US outlawed the export of the cards to China in October 2022, its caregiver High Flyer, a Taiwanese hedge fund, claimed it had amassed a swarm of 10,000 A100 cards.  ,
But then Lutnick suspects that DeepSeek bypassed the US trade controls by importing high-end Nvidia cards via third-countries, such as Singapore.  ,
The Wall Street Journal reported last July that some shady establishments in Singapore paid Chinese pupils to take the A100 bits back to China.  ,
Without providing any evidence, Alexandr Wang, chief executive of the US-based Scale AI, told CNBC that DeepSeek has 50, 000 models of H100 bits, the most advanced Nvidia cards on the market.
Xiang Zhiping, a Hubei-based IT poet, finds that realistic. ” It’s no surprise if DeepSeek has 50, 000 H100 cards. Xiang claims that any Taiwanese internet company may have accumulated a lot of Nvidia cards.  ,
Even if DeepSeek has a lot of chips, he claims, it will still use technology and modern IT frameworks to win the game rather than relying on outdated hardware to increase technology power indefinitely.  ,
The US banned the export of A100 and H100 cards to China in October 2022, and then the slower A800 and H800 cards in October 2023. After this, Nvidia tailor-made the even-slower H20 cards for the Foreign businesses.
The Biden presidency made a regulatory framework known on January 14, 2025, that would enact a ban on imports of American AI devices and designs. The foundation became effective on January 31.  ,
Some observers claim that the device ban’s gradual increasing gave China too much time to build premium AI cards.
Yong Jian contributes to the Asia Times. He is a Chinese columnist who specializes in Chinese technologies, economy and politics.  ,