What DeepSeek’s AI really means for the market

Unlock the Editor’s Digest for free

Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.

Good morning. Donald Trump’s Treasury secretary Scott Bessent supports universal tariffs starting at 2.5 per cent and rising as as high as 20 per cent over time. The idea is to give countries a chance to negotiate with the US before the real pain starts. It’s unclear, however, if Bessent has convinced anyone — crucially, Trump — of his plan. Email us: robert.armstrong@ft.com and aiden.reiter@ft.com. 

That sound you just heard was not a bubble popping

The most important trend in finance in the past 40 years is the emergence of winner-take-all industry structures in technology. We live in a world of network effects and increasing returns to scale. In industries from PC software (Microsoft), internet search (Alphabet), online retail (Amazon) and social media (Meta) to high-end consumer hardware (Apple), most of the profits and market share have gone to one company. It is only a moderate exaggeration to say all that has mattered for investors in the past few decades was being on the right side of one or more of these winner-take-all stories.

So it is natural that everyone either has assumed, desired or feared that the next great technology trend, artificial intelligence, would have roughly this economic structure. What happened in markets yesterday was not an out-of-nowhere surprise, nor a panic, nor a bubble popping. It was the pricing in of slightly higher odds that AI is not a winner-take-all game.

It was always conceivable that the AI revolution would be like the invention of the automobile or the aeroplane. Those revolutions did lead to the creation of huge, durably profitable companies, but also lots of competition, ensuring that much of the value created went to consumers rather than shareholders. Now that looks like a more significant possibility for AI. 

Most readers will know the news by now. DeepSeek, a Chinese AI company, has released an AI model called R1, that is comparable in ability to the best models from companies such as OpenAI, Anthropic and Meta, but was trained at a radically lower cost and using less than state-of-the art GPU chips. DeepSeek also made enough of the details of the model public that others can run it on their own computers (but not necessarily enough that others could re-create it). 

This puts three punctures in the winner-take-all theory of AI. It deflates the idea that the best AI results could only be achieved with Nvidia’s best chips and software; the idea that only the biggest tech companies could afford to build and run high-quality AI models; and the idea that only companies with their own AI models could offer great AI applications.

What kept Nvidia’s stock rocketing for the past two years was sheer computational superiority. Its chips were miles better for AI. DeepSeek built R1 with Nvidia’s older, slower chips, which US sanctions allow to be exported to China. This suggests the door is open to other competitors at the silicon level. 

Next, cost. The biggest tech companies looked more or less insulated from competition on AI because, it was assumed, the only way to build a better AI model was to train it with more inputs using more computing power, at huge expense. The finished model would then be under the control of the big company that built it. The barrier to entry was money — that is why Big Tech has gone on a data centre spending spree. But R1 suggests that companies with less money can soon operate competitive models.

Finally, applications. “Inference” is where the user really meets AI: questions answered, tasks completed. It was assumed that any application would depend on the bigger companies’ models and servers for inference. But models like R1 might work well enough, and do inference efficiently enough, to run on the servers of all sorts of companies that will not need to “rent” from a company such as OpenAI.

All this has big implications. But other than Nvidia, the big fallers yesterday were not the Magnificent Seven Big Techs. Those companies’ AI halos remain squarely in place. The real damage was done to the stocks behind the emerging data centre economy. Constellation, the utility company that covers the mid-Atlantic data centre hub, Vistra, Washington state’s big power provider, and NRG, the Northeast’s and Texas’ main player, all fell hard:

GE Vernova, Eaton and Quanta Services build power systems for data centres. Oracle just announced a big data centre investment. Broadcom and Arista provide non-GPU technology to data centres. All were whacked:

Amazon, Alphabet, Meta and Microsoft have all invested very heavily in AI data centres. Perhaps they have wasted some of that money — but they could be free to spend less now. Apple and Microsoft, whose natural strength might have been in building applications for AI models rather than the models themselves, could even be in a better position.

Nvidia itself is hardly on its knees. Its proprietary coding language, Cuda, is still the industry standard. And just because the DeepSeek model is more efficient does not mean that leaner models will not benefit from the higher computational power offered by Nvidia’s best chips. While its shares dropped nearly 17 per cent yesterday, that only brings it back to the (very, very high) level of September:

We should not overstate the market’s reaction to R1. The Nasdaq fell 3 per cent. That was a bad day, not a panic. The winner-take-all view of AI is wounded, not dead. The AI bubble, if that’s what it is, may yet pop, but it didn’t pop yesterday. Not even close.

One good read

Misplaced hope.

FT Unhedged podcast

Can’t get enough of Unhedged? Listen to our new podcast, for a 15-minute dive into the latest markets news and financial headlines, twice a week. Catch up on past editions of the newsletter here.