Quid of the energy race behind data centres?

We like to talk about artificial intelligence as a matter of software, researchers and chips. We forget that it is first a matter of electricity. Every model trained, every query served, every image generated starts with a kilowatt-hour drawn from a grid somewhere.

For a long time, only engineers cared about that detail. It has become a strategic, financial and political question.

The bottleneck has moved

Not long ago, the limiting factor for AI was the chip. Today, in many countries, it is the grid connection. Building a data centre takes a few years; reinforcing a high-voltage line or adding new generating capacity can take far longer. Projects end up in a queue, and access to power becomes an asset in its own right.

The ideal site for a data centre is no longer the one close to its customers. It is the one close to a substation.

What it means for investors

The consequence shows in the strategies of the large tech companies: long-term power purchase agreements, renewed interest in nuclear, small modular reactors included, and investment in storage and grids. Companies once filed under dull income stocks find themselves at the heart of the technology story.

For investors, the AI value chain is stretching upstream: power producers, grid operators, transformer makers, cooling specialists. These are slow, capital-intensive, regulated businesses. The very opposite of how we picture tech.

Europe has a card to play

The continent does not have the largest AI labs. It does have, in some countries, abundant low-carbon electricity, interconnected grids and an industrial tradition in energy. If the AI race becomes partly an energy race, the geography of the competition shifts.

What remains is to turn that advantage into projects: faster connection procedures, clear rules on the use of resources, water in particular, and long-term financing. It is not the kind of subject that makes headlines. It is the kind of subject that picks the winners.