Google is securing nuclear capacity in Finland to power AI data centres. It is the direct follow-on from the movement we described on Wednesday: operators are no longer asking the grid for electricity but going to get it themselves. This version is the best possible on emissions, and it raises the same questions.
We wrote two days ago that data centres were leaving the electricity grid, and that the main risk was a choice of sources decided by availability rather than by emissions. Here is the counter-example, and it deserves to be welcomed before it is discussed.
Why this is the right version
Three reasons make this deal preferable to what we feared.
Emissions. Electricity of nuclear origin has a very low carbon intensity, comparable to that of wind. For a load that runs continuously, that is hard to beat.
Consistency. A data centre consumes day and night, without variation. That is exactly the profile nuclear serves best, unlike solar, which requires massive storage to cover the night.
Location. The Nordic countries offer a climate that reduces cooling needs, and therefore the water consumption we documented as a sticking point in India.
The question that remains
Despite all this, the underlying problem we raised on Wednesday remains whole, and it is not technical.
Decarbonised generating capacity is a limited resource. Finnish nuclear electricity reserved for data centres is not available to decarbonise the country's industry, heating or transport.
In other words, the question is not whether this electricity is clean. It is. The question is whether its best collective use is to power computation, rather than to replace fossil uses elsewhere.
This is not a technical question, it is a trade-off. And it is being settled by the ability to sign a long-term contract rather than by deliberation.
It must be put honestly. An operator that finances the construction or the continued operation of decarbonised capacity adds generation rather than diverting it. If the deal allows a plant to keep running or a new one to be built, the collective balance improves. The whole difference lies there: a contract that buys existing electricity shifts a use; a contract that finances new capacity creates one. These two situations are very different, and press releases rarely distinguish between them.
What it says about the trajectory
This movement confirms what we described: the sector is resolving its energy constraints through private long-term agreements, and that produces three effects.
Concentration. Signing for twenty years with an electricity producer requires capital that only a few players possess. Barriers to entry close a little further.
Geography. Locations now follow available electricity, which shifts activity towards the countries that have it, irrespective of any other consideration.
Statistical invisibility. This is the point we highlighted on Wednesday. Consumption that goes through bilateral agreements partly escapes public monitoring mechanisms, at the very moment when measurement is becoming essential.
What to take away
This deal is far preferable to a gas-powered data centre. It must be said clearly, because we regularly criticise this sector over its consumption and it would be dishonest not to acknowledge a good choice.
It does not, however, settle the central question, the one we formulated on Wednesday: who decides how these facilities are powered. Here, a private player decided, with the means to secure a rare resource for two decades.
The right response would not be to prevent this kind of deal, which improves the balance. It would be to require that these commitments be declared, comparable and integrated into the energy planning of the countries concerned. That is what exists for heavy industry. AI has become one, and its energy deals are still treated as private matters.