Unlock the Editor’s Digest for free
Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.
The most influential person in tech today is a dead economist: William Stanley Jevons. The paradox bearing his name has become a meme in Silicon Valley as companies try to keep their share prices afloat.
Jevons is best remembered for saying that the arrival of efficient steam engines might not make the world use less coal — and could do the opposite. It’s a handy narrative for Silicon Valley bosses after Chinese artificial intelligence company DeepSeek claimed to have found a cheaper, easier way to train AI models than their own.
As a result, neo-Jevonsianism is suddenly in fashion. Microsoft’s Satya Nadella, Nvidia founder Jensen Huang and ASML chief Christophe Fouquet all seem relaxed about DeepSeek. What’s bad is good, they argue: cheaper AI will mean more demand for AI.
They are right, directionally. Think of chips and data centres as today’s coal. If AI applications became cheaper, through using less computing power, customers ought to deploy them more. Demand for compute could, as coal did, rise. An example of this in the real world: air conditioning.
But even if AI proves a Jevonsian good, not everyone stands to benefit equally. The impact on big tech firms depends in each case on whether they’re a taker of AI, a maker, a broker or an enabler.
Takers should be winners whether Jevons is relevant or not. Think of software companies like Salesforce, ServiceNow or Adobe, that use AI as a building block and might now be able to get it at lower cost.
For makers of AI, the rough outweighs the smooth. Meta Platforms and OpenAI have spent billions on power, data, human effort and microchips to generate models like Llama and o1 that — if DeepSeek is the real deal — could be aped for a fraction of that. They may turn out to have overinvested relative to their future returns.
Brokers are in a happier place. Alphabet, Microsoft and Amazon all rent out computing power, for example. Leaner AI models mean customers need less of that, but they’ll eventually want more, making this really a timing problem. They would, though, pass the pain on to enablers — chipmakers like Nvidia, whose shares have fallen 13 per cent since Friday.
Over time, the lines have blurred. Microsoft is a broker — it reported $40.9bn of quarterly cloud revenue on Wednesday — but also an investor in OpenAI, a maker. Meta both makes and takes. Alphabet, owner of Google, is all three — it sells cloud computing, builds AI models and uses them to improve its search and ad sales offerings.
In any case, financial returns may already be an afterthought. Companies focused on artificial general intelligence — that which matches humans — are likely to go for broke regardless. Meta boss Mark Zuckerberg talked on Wednesday of investing “hundreds of billions of dollars” in AI infrastructure.
While Jevons is the flavour of the minute, he is therefore of limited help in 2025. Besides, he couldn’t have foreseen the pace at which changing expectations get priced into markets. His 1866 work “The Coal Question” forecast UK coal demand would double every 20 years. Silicon Valley and its investors won’t wait that long.
john.foley@ft.com
