Originally published in Svenska Dagbladet by Björn Jeffery, November 12, 2025
Tech giants are expected to burn through 4,600 billion kronor on chips and data centres next year. Many are warning of a new bubble. But what if it’s actually too little?
The numbers quickly become dizzying — difficult to read and difficult to comprehend.
26,500,000,000,000 kronor is what investment bank Citigroup estimates tech giants will spend on AI through to 2029. To put that figure in perspective: 26,500 billion kronor is roughly four times Sweden’s GDP.
It is easy to dismiss all of this as bubble economics and a prolonged replay of the dot-com era. I have made these comparisons myself in previous analyses. The intellectually honest thing to do is therefore also to pose the opposite question — what if it is actually too little?
We can start with someone close to the matter: Julian Schrittwieser, a researcher at AI company Anthropic. He believes people are poor at understanding true exponential development and what it looks like in practice. One example he gives is Covid, and how it took the outside world quite a while to truly grasp how extensive the spread of infection would become.
In the field of AI, according to Schrittwieser, we are early in what could be an exponential era. Already, the length of tasks that AI can perform is doubling every seven months. Difficult questions — and conceivable work assignments — are becoming progressively easier. In purely mathematical terms, Schrittwieser says that an AI model by the end of 2026 will be as good as a human expert in a particular field of knowledge. By the end of 2027, models will often be better than experts across the board.
That “a computer” could manage something like that would have been completely unthinkable just a few years ago. Now some believe it is within sight. The exponential moves fast.
With this as a starting point, the large investments in AI infrastructure become easier to understand. There are arguments that we are currently in a bottleneck that is constraining the pace of AI development. Had we had more electricity, more chips, and more data centres, we could have accelerated even further and reached better results sooner.
And it is precisely that underlying infrastructure that all of this is about.
What is called AGI — artificial general intelligence, the term referring to when AI becomes as smart as, or smarter than, a human — will be difficult to achieve with the capacity we have today. If you believe we are heading towards this future, we need to invest here and now. The AI companies’ advance orders for electricity from small modular reactors — SMRs — can be seen in this light. How else will there be enough electricity when we reach the point where we need it?
This does, admittedly, sound a little bubbly.
But to take a more down-to-earth comparison, one can think back to when the underground railway was being planned. This is not a historical account — those can be found elsewhere — but simply a metaphor for what it can sound like.
Imagine someone proposing to dig deep tunnels beneath Stockholm and lay rails in them. Trains would run on those rails every few minutes, and over time the network would become the most important component of how people in Sweden’s capital get from A to B. It is a large and radical idea, viewed from that moment in time.
A sceptic would likely have asked who would even want to sit in these tube-trains. They would also wonder whether it would not be enormously expensive to dig all these tunnels before we even know whether it will work, or become popular. These are reasonable objections.
But such is the nature of new infrastructure. It is uncertain, expensive — and it must precede its actual purpose by many years. If you therefore believe that a radically different society, driven by AI, is on the horizon — well, then the time to build its infrastructure is here and now.
One can hold different views on whether this AI future is as imminent as the tech giants appear to believe. One can also wonder whether such a future would be something desirable for all people on earth. But spending trillions over the coming five years “laying the rails” for such a world is at least a consistent line of reasoning from their worldview.
If you feel that humanity is in a bottleneck when it comes to solving chronic disease, climate threats, space exploration, and poverty — what exactly are we waiting for?
