The US and China have won the data race. Europe's advantage is that we never needed to enter it.

Europe's answer to the AI race is seven "gigafactories": €10 billion of public money and €20 billion of private money for compute clusters of 100,000 chips each. The tender has slipped five times since it was announced. Seventy-six companies expressed interest; about ten are expected to bid. Last week the deadline for four of the seven sites was pushed to February 2027, because the member states have not finished approving their share of the money. Most of the EU money arrives in 2028 and 2030.

This is not a programme that is behind schedule. It is a programme for a race that is already over.

The US and China are so far ahead in AI training that Europe cannot catch up. We will not build the facilities, and we cannot gather the data, needed to compete on that frontier. Simple as that.

And that is the best news European tech has had in a decade.

Our biggest advantage is that we don't need to covet our users' data. We are not going to train models on it. Our privacy rules make it hard to use for anything else. What looks like a constraint is the only durable moat in this industry.

In Europe, we can build the privacy-respecting AI infrastructure the world actually needs. The AI that doesn't leak your data. The AI that doesn't copy your knowledge. The AI that does not make a business out of learning your secrets.

Breaking a habit is harder than writing a rule

Building that will not be easy. Today's tech infrastructure assumes data is the product. Billion-dollar funding rounds have gone into systems designed to turn terabytes of data into future profits. The pipes, the platforms, the business models and the investment theses are all built on that one assumption.

When I go around the European startup ecosystem, as I have done for most of my career, I meet the same assumption everywhere: we must copy the data-driven business models of the Americans, regulatory obstacles notwithstanding. That might have been true when the obstacles were purely regulatory. Now that they are systemic, it is no longer true.

A regulator can be argued with. Physics and capital cannot. If the frontier requires data volumes and compute clusters that no European company can assemble, then "we'll do what they did" is not a strategy. It's nostalgia.

What Europe can build instead

Let's build the AI infrastructure of the future: infrastructure that does not rely on amassing the world's data, and that does not make data the raw material of the industry.

Concretely, that means infrastructure where the data stays where it is. Where the model does the work without taking a copy. Where your memory of a conversation belongs to you and your machine, not to a database somewhere that your questions slowly enrich. Privacy as architecture, not as a promise in a policy document.

That is not a smaller ambition than the frontier labs'. It is a harder engineering problem, because you cannot solve it by adding more data. You have to design systems that are useful precisely because they are constrained.

What the world is starting to ask

In doing so, we may deliver what the entire world is starting to seek. AI that does not snoop on you. AI that answers to nobody but you. AI that answers your questions with no ulterior sales or political motives. AI that does not centralise all human knowledge in the hands of a few governments and companies.

The demand for that is not European. It is human. It just happens that Europe is the place where the constraint is hardest, the rules are clearest, and therefore the incentive to get the architecture right is strongest.

What we're doing about it at Rool

I don't want to end on a manifesto. So here is the part of this that isn't an argument but a product.

Rool builds personal AI machines. Your conversations and your memory live on your machine, not in our database. There is no second customer: you pay for the machine, nobody pays for you. We run the compute in Finland and Denmark, for the clean energy, and we keep everything else small enough that subscriptions cover it. That is what makes it possible to never need your data. Not as a policy, but as a business model that has no use for it.

That is one company's version of the European opportunity. There should be a hundred.

Let's get started.