The start-up claimed that running queries with DeepSeek R1 is 98% cheaper than using OpenAI’s flagship model. It also purportedly costs much less to train than OpenAI’s GPT models and other rivals.

There was little proof that DeepSeek could do anything claimed on the tin, and some analysts thought that many of the company's promises and finances were unlikely—particularly in the short to medium term.

However, that did not stop a massive share dump by those who don’t know what is happening. AI-related stocks took a hit on Monday, with Nvidia shares tumbling 17 per cent, shedding $600 billion in value and marking the single-biggest one-day loss for a company in stock market history. ASML sank six per cent, while Broadcom slumped 17 per cent and TSMC was down 15 per cent.

Even Cryptocurrency fell for no apparent reason and because the Chinese outfit claimed that its product would take less energy to run, some energy-related stocks also plunged. GE Vernova, which makes wind and gas turbines, fell 21 per cent, while electricity generator Vistra slumped 28 per cent.

By the time investors had calmed down, a trillion dollars had been wiped off the NASDAQ, and nothing had changed in the real world.

Wedbush Securities analyst Dan Ives pointed out that no US Global 2000 is going to use a Chinese startup to launch its AI infrastructure and use cases.

Ives wrote: "At the end of the day, there is only one chip company in the world launching autonomous, robotics, and broader AI use cases, and that is Nvidia."

Some Wall Street analysts think Monday's stock selloff is an overreaction, noting that the enormous demand for AI will continue lifting key players in the sector.

VitalKnowledge analyst Adam Crisafulli said: “It's one thing to train a [large language] model for less money, but accommodating the huge demand for the consumption of all this AI technology is still going to require massive amounts of infrastructure.”

In a statement to CBS News, Nvidia offered praise for DeepSeek.  "DeepSeek is an excellent AI advancement and a perfect example of test-time scaling/ DeepSeek's work illustrates how new models can be created using that technique, leveraging widely available models and compute that is fully export-control compliant."

But, Nvidia added, AI inference, or using AI models to make decisions or predictions, "requires significant numbers of Nvidia GPUs and high-performance networking. We now have three scaling laws: pre-training and post-training, which continue, and new test-time scaling."