Originally published in Svenska Dagbladet by Björn Jeffery, October 20, 2025
The AI boom has made data centres the world economy’s new growth engine. Many are now worried that diminishing interest could lead to economic crisis.
A long, narrow grey building sits along Malmövägen, just before you drive into Staffanstorp. To the untrained eye it could be some kind of sports hall. On the inside, however, you find something entirely different. This is the home of “Microsoft Cloud Operations” — one of the tech giant’s data centres in Sweden.
The ongoing AI boom has put a laser focus on this otherwise rather sleepy category. The expansion of data centres in the United States has become so important to the American economy that some analysts believe it can barely sustain its growth without them. But what happens if demand for data centres suddenly falls?
Harvard professor Jason Furman is one of those who has asked this question and done some rough calculations. Around 4 percent of American GDP — roughly 11,000 billion kronor — is being invested in what is called intellectual property and physical computing infrastructure. A significant portion of that is data centres.
The reality is a little more nuanced than that. Had the money not been invested in data centres, it would likely have ended up somewhere else instead, which could also have contributed to GDP. Furman’s example is therefore hypothetical, but illustratively interesting.
Is the United States — and through it the entire world with its interconnected economy — relying far too much on a single category? And if so, how would the world economy fare if interest and enthusiasm were to diminish?
Such a reversal could take different forms. Today, AI companies are building out capacity for the demand they believe will emerge in the near future. The idea is that once we get there, it will be too late to build. Some individual enthusiasts argue that the lack of chips and data centres has already created a bottleneck for AI development, and that it would have progressed considerably faster had we had access to more resources.
Regardless of whether it is a bottleneck or not, data centres take a long time to build. If you believe demand of some kind will increase going forward, you have to build here and now. The long series of deals that OpenAI recently made with companies like Oracle, AMD, and Broadcom are examples of this. If it turns out that demand does not materialise — or only arrives at a different time — these billion-kronor deals could turn out to be even more expensive than they first appear.
There are also other reasons why we might not end up needing as many data centres. One simple reason is new types of AI development that do not require as much computing capacity. Last week brought one such example from Chinese tech giant Tencent. They demonstrated a new method that, in simplified terms, allows AI models to learn from themselves. The result is that the amount of computing capacity required decreases radically. The example shows how models that cost over 10,000 dollars to train are beaten by this new model — at a total cost in the region of three dollars.
Over the weekend came similar news from Alibaba, where their new system Aegaeon reduces the need for Nvidia’s H20 chips by as much as 82 percent.
Regular readers of these analyses may recall the Chinese AI model DeepSeek from January of this year. It was described at the time as a “Sputnik moment” by venture capitalist Marc Andreessen — the moment when the world realised there were other players in the AI race, China in particular. DeepSeek, like the Tencent and Alibaba examples above, also used fewer data resources than the American models. The mere thought of that caused financial markets to tremble significantly.
Companies like OpenAI and Anthropic have painted themselves into a corner in several ways. To justify the companies’ high valuations, they need to sell a vision of the future with extraordinary optimism — a vision of AI becoming smarter, faster, and more accessible to more people. To meet this vision in practice, enormous investment is required here and now, including in data centre expansion. And to finance this expansion, more money is needed — preferably at a high valuation of the company.
Should any part of this loop snag — whatever the reason — it could become difficult for the AI companies to continue in the same way. The business model is built on a belief about the future commensurate with a new kind of industrial revolution. As a reminder, OpenAI’s valuation as of this writing is over 4,700 billion kronor.
For those who have invested at that level — and for the global world economy as a whole — we must hope that the money being spent on data centres is well spent. The interest — and dependence — from the outside world in this anonymous and dull industry has never been greater than it is right now.
