DeepSeek launched its R1 artificial intelligence product late in January just as the United States was announcing its AI super project, Stargate.
Stargate produced a fleeting headline or two before the DeepSeek release blew the doors off Wall Street, wrecking the valuations of U.S. competitors for a day. In fact, its debut set a “one-day record for value destruction in a single name,” wiping out some $589 billion in market capitalization in just one trading session. Watch this CNBC snippet for a taste of the doom mongering.
R1 is ridiculously cheap. It’s open-source (the code is yours to play with). And testing shows that it’s technically equivalent to the pricey American stuff like ChatGPT. Demand for the app has been such that it became “the No. 1 downloaded free app on Apple’s iPhone store.” The buzzword of the day is “Sputnik moment.”
Red China had said that it wanted to supply the undergirding for all future AI worldwide, and this may be a good start.
Magic and miracles
The story of the DeepSeek company itself deserves deep attention. It is a charming fairy tale filled with magic and miracles.
The firm was founded by the Chinese hedge fund High-Flyer in 2023. The hedge fund invested just $4.1 million in DeepSeek—which has been the only funding it received apart from any seed money. The $4.1 million was invested earlier this month, so there must have been a lot of seed. And yet, the seed round only closed in June of 2024. It was magic seed, perhaps.
So R1, with almost no money, took under two years to be developed into a Silicon Valley giant killer. The staff of 39 got the job done at a high-quality level to an open-source standard, documented, tested, and apparently well promoted. And it beat to market the 50 or so other Chinese firms working on AI models.
In the grown-up world, none of this makes sense, of course. Because $4.1 million would hardly cover office rent, telephone service, and coffee supplies for two years, nor the salaries of the small staff. Moreover, the brag going around is that DeepSeek spent only $5.6 million to develop its model. So there is that discrepancy too.
Behind the curtain
Who was funding DeepSeek? This whole scenario feels like the CCP picking a winner behind the scenes. One analyst with Rabobank calls DeepSeek “Chat CCP” and a “Potemkin village.” You won’t need a large staff or a lot of money and equipment if you’re handed another Chinese company’s code.
Anduril founder Palmer Luckey says, regarding DeepSeek’s claim that it took only a few million to develop its AI model: “They put out that number specifically to harm U.S. companies. You had a lot of useful idiots in U.S. media kind of just mindlessly reporting that that’s the case, and neither China nor the media nor DeepSeek has any kind of incentive to correct the record as a lot of U.S. companies like Nvidia crashed to the tunes of hundreds of billions of dollars.”
Of course, you could also steal the work developed by your competition at great cost. IP theft from Open AI seems to be what’s happening. “DeepSeek’s research even admits its R1 model is based on other open-source systems.”
Outside of the Reddit discussion board, independent journalist George Webb may have been the first to suggest that DeepSeek is a Chinese Communist project. Human rights activist Luke de Pulford also noted the scope of user data being collected and stored in China: another tell.
Both men pointed to the end-user agreement (that document we generally agree to without reading), de Pulford noting the unusual data collected by a nominally private firm and Webb that “DeepSeek owns everything you create according to its terms and conditions.”
And yet, some take the fairy tale seriously. Thus, we have been told (in a now-removed Reddit post) that “Chinese Deepseek devs just proved GenAi is a giant scam inflated by capitalists and is actually worth less than $5.5 million. Apparently, these developers have managed to show that training a state of the art AI model is dirt cheap. Some are reporting that 200k requests to Deepseek API only cost them $0.50. And now US-based AI companies are in panic mode.”
A hybrid view
There is a hybrid view worth considering: “Have you been to US universities? Many of the top [computer science] students are Chinese (or Indian). And I’m not talking about Chinese-American, but international Chinese students. One of the main reasons the US is the leader (or co-leader) in AI is Chinese talent. And this isn’t only in AI, China is overtaking the US in many areas of research as well while the US continue showing signs of decline. Plus, the brain drain [to the US] is slowing down; unlike before, Chinese researchers are going back to China after finishing their studies.”
How much, then, is DeepSeek versus ChatGPT a Chinese-versus-Chinese affair? How much of it is IP theft? How much is state-funded?
DeepSeek is already branching out from what might be called an AI operating system to AI applications. George Webb asks: “Do you remember California Governor Gavin Newsom having a press conference about Deep Seek Labs doing an AI fire prevention system in Los Angeles before the breakout of the disastrous fires raging now? Why wouldn’t California AI companies conduct this AI fire prevention exercise with fire prevention pilot products like Google, Meta, Grok, and OpenAI?”
If they want to sell in this country, DeepSeek and its owners should come clean about their ownership, investors, and product development. They should not be given a free pass to wreck the US AI industry. □
James Roth works for a major defense contractor in Virginia.
