Europe's model companies found a position. It is narrower than the rhetoric and more durable than the sceptics expected.

The measure applied to European model companies for three years was the wrong one. Compared against the largest American frontier runs, every European lab looked like a shortfall. Compared against what European buyers were actually procuring, the picture reverses.

European enterprise AI demand is dominated by a specific shape of requirement: strong multilingual performance in European languages, deployment inside the customer's own environment or a European cloud region, contractual clarity about training data, and a per-token price that survives a volume rollout.

That shape does not require a frontier model. It requires a good model with an unusually clean legal and operational story, and that is precisely what the European providers have built.

The on-premise question is not nostalgia

American vendors often read European insistence on local deployment as regulatory theatre. In banking, defence, public health and energy, it is nothing of the sort. These are sectors where the data cannot leave a jurisdiction and where the procurement officer is personally accountable if it does.

A provider that ships weights the customer can run is not offering a worse product to those buyers. It is offering the only admissible one.

Where the position is fragile

Two risks. The first is that open weight releases from the largest labs compress the differentiation: if a globally distributed open model matches European multilingual performance and can also be self-hosted, the local advantage narrows to contract terms alone.

The second is talent concentration. Europe trains excellent researchers and exports a significant share of them. The counterweight this year has been the growth of European offices belonging to non-European labs, which keeps people on the continent without keeping the intellectual property here.

The realistic read

Europe is unlikely to produce the largest model. It is well positioned to produce the models that regulated European industries can actually deploy, and the size of that market has been consistently underestimated by people measuring only the frontier.