Share prices for some of the biggest American tech brands that crested the AI hype waves crashed this morning on the rocks of DeepSeek, a Chinese startup that last week released LLMs that challenges US dominance.

As The Register revealed at the weekend, DeepSeek launched some openly available machine-learning models which perform favorably against US competitors OpenAI and Meta, according to benchmarks. And what’s really set the cat amongst the investors is that they were trained using fewer Nvidia chips, or so DeepSeek claims.

DeepSeek, founded in 2023, by Liang Wenfeng and financially supported by his quantitive hedge fund High Flyer, released DeepSeek V3 at the close of 2024, and has now rolled out R1, classified as a reasoning model optimized from V3.

This led investors to question the viability of spending tens of billions of dollars on AI training, especially when the returns on investment still seem uncertain. This existential crisis hit stock valuations, with Nvidia down by as much as 14 percent in pre-market trading today, Microsoft falling 7 percent, and Meta down 5 percent.

The Nasdaq 100 futures was down 3.51 percent at the time of writing amid the searching questions that DeepSeek is seemingly posing.

“Deepseek R1 is AI’s Sputnik moment,” said Marc Andreessen, partner in venture capitalist Andreessen Horowitz, which raised billions last year to spend on AI startups. As Reg readers no doubt know, this effectively indicates that something has to change so US big tech can compete without spending so much.

Microsoft, AWS, Google and others are ploughing billions into AI in the expectation that those bets come off. Meta said last week it is spending $60 billion on AI and there is also the Stargate initiative that is betting a collective $500 billion.

As of September last year, the hyperscalers had spent $200 billion in capital expenditure since the start of 2023, and yet AI licenses were around a tenth of that. The bubble was expanding and expanding.

Former Microsoft exec Steven Sinofsky said on X:

The “big scale solutions” are “consuming too much capital,” he added on X. He reckons that beyond this, the current path is not sustainable.

“It is a path that works against the history of computing, which is that resources needed become free, not more expensive. The market for computing simply doesn’t accept solutions that cost more, especially consumption based pricing. We’ve seen Microsoft and Google do a bit of resetting with respect to pricing in the hopes of turning these massive CapEx efforts into direct revenue.”

The arrival of DeepSeek – which claims it built a scale-out architecture to train its V3 model on 14.8 trillion tokens using 2,048 Nvidia H800s, amassing 2.788 million GPU hours, and costing circa $5.58 million – has been impactful in the financial world over the past few hours.

“It seems as if there is a bit of reality dawning that China has not been sitting idle, even as these tariffs and investment restrictions on tech companies have been put in place,” said Mitul Kotecha, Asia head of emerging markets macro and foreign exchange strategy at Barclays, as quoted by the Financial Times. ®