By Björn Jeffery, SvD Tech Brief. Published in Svenska Dagbladet on 27 January 2025.
Hundreds of billions have been invested in AI chips, making Nvidia the world’s most highly valued company. Now a small Chinese player is challenging the need for those investments — and the markets are beginning to shake.
On 27 March 2000, Cisco became the world’s most highly valued company. The American networking giant was riding high on dotcom waves and the internet economy promised growth and success. Cisco was selling the picks and shovels of the gold rush — networking equipment. As everyone around the world scrambled to get connected, it looked like a sure winner.
Everyone remembers how it ended. Less than a year later, Cisco’s share price had fallen by 80 percent and the crash was a fact. Now the markets are shaking and the anxiety is clear: are we facing a similar scenario once again?
The Chinese AI model DeepSeek R1 is giving markets a sense of déjà vu. Over the weekend, DeepSeek reached the top spot in the American App Store, overtaking rival ChatGPT. Venture capitalist Marc Andreessen called it the “Sputnik moment” of the AI world — the moment when the world understood that there were other players in the game.
The appeal for consumers is straightforward: DeepSeek is good and free. But it is how DeepSeek was built that is truly disruptive. It is said to have used far fewer resources than comparable competitors while still delivering equivalent results. Should that prove to be true, it could be bad news for companies like Nvidia, which has prospered handsomely from tech giants’ investments in chips and data centres. Will as many really be needed going forward?
Nvidia — today the world’s most highly valued company — carries expectations of growing, not shrinking, demand. When markets opened on Monday, the company’s share fell by around 13 percent — representing a loss of value of approximately 465 billion dollars, the largest single-day loss in stock market history.
Making matters worse, China faces strict restrictions on what type of chips it can import from Nvidia. DeepSeek therefore appears to have been developed using inferior chips compared to those available to Meta and Google.
A great many uncertainties remain. These range from suspicions that DeepSeek stole data and information from ChatGPT, to claims that the company behind it does in fact have access to the best chips but cannot admit it for understandable reasons. Meta has assigned extra staff to try to work out how it succeeded. But the uncertainty is sufficient to rattle the markets. Dutch company ASML, which makes machines for manufacturing chips, fell around nine percent when European exchanges opened.
If it turns out that the Chinese have found a more efficient way to develop AI models, it can be interpreted in two different ways.
The first is what underlies the anxiety described above. Do we really need more and better chips if we can achieve the same type of results with far fewer? Is this a potential paradigm shift in how AI models are trained? It could mean a reduced need for chips and data centres, and lower investment required to remain competitive. It would also mean that high-level AI development could take place in many more countries, moving beyond today’s heavy concentration in Silicon Valley.
The second perspective is what is known as Jevons paradox. Named after the English economist William Stanley Jevons, who showed in 1865 that increased efficiency in the use of coal did not reduce demand for the raw material — but rather increased it. Applied to AI development, this would mean that the need for capacity and chips will only grow as the stakes get larger and more players can participate. Microsoft CEO Satya Nadella wrote on Monday that he believes this is precisely a Jevons paradox moment.
Looking back at Cisco in 2000, there are further parallels to be found. The direction for the future proved to be correct — the internet economy would indeed become world-changing. But it took considerably longer than anyone had expected. And the immediate need for Cisco’s hardware did not materialise on cue. Only now — 25 years later — has Cisco’s share price caught up and is approaching the same heights it once reached. But more than anything else, Cisco’s dotcom valuation was based on near-infinite growth projections. Enthusiasm overrode reason.
The fact that a chip manufacturer like Nvidia is the world’s most highly valued company speaks to a similar enthusiasm today. It is not sufficient, necessarily, that AI is here to stay, or that it may come to revolutionise society and business life. Timing also matters. And as DeepSeek is reminding markets right now: there may be more ways to reach that vision than buying vast numbers of extraordinarily expensive chips.
