Mr Lane's Speed Limit and the Clause Nobody Enforces

Philip R. Lane, chief economist of the European Central Bank, published a text on Monday that reads less like a speech and more like an inventory. The AI opportunity for Europe, as he titled it, lists the actors who must play their part: frontier model developers, infrastructure builders, entrepreneurs, researchers, energy operators, financiers, educators, managers, workers. The list is deliberately long. The implicit question is which of these actors the European Union has already legislated into or out of the game.

Mr Lane's number worth retaining is this: euro area firms are set to devote around 10 per cent of their total investment to AI this year. That is not a forecast of future transformation. That is current activity. Business investment, digital services expansion, global demand for European components and capital goods — the ECB sees AI already moving the quarterly dials.

Against that figure, place a second one that reached European inboxes this week from across the Atlantic. According to local press reporting out of San Francisco, the developer of the Claude model generated 16.8 billion dollars in revenue in the three months to the end of September, a 45 per cent quarterly jump, en route to a stock market listing. Its revenues this year, by the same reporting, progressed from 4.7 billion in the first quarter to 11.6 billion in the second to the latest figure. The ChatGPT developer is tracking toward a valuation above one trillion dollars.

Two numbers, one problem. European firms are putting ten cents of every investment euro into AI. The frontier models they are integrating into their products are overwhelmingly built by American companies whose quarterly revenues now exceed the annual research budgets of several member states. This is the competitiveness gap Mr Lane is gesturing at without naming.

The regulatory question underneath is sharper than the usual complaint that Brussels over-regulates. The AI Act and the Digital Markets Act were drafted by different Directorates-General, pushed through by different Council presidencies, and amended by different Parliament committees. They do not share an enforcement philosophy. The AI Act polices the model layer: risk categories, general-purpose model obligations, systemic risk designations. The DMA polices the platform layer: gatekeeper duties, interoperability, self-preferencing. The AI firms Mr Lane is implicitly worried about operate across both layers simultaneously.

Who wrote this split? France pushed, during the AI Act trilogue, for lighter treatment of foundation models, with an eye to its national champion. Germany joined that push. Several smaller member states, backed by the Parliament's civil liberties voices, pushed the other way. The compromise that emerged — tiered obligations, codes of practice, a European AI Office housed within the Commission — satisfied nobody and now has to be enforced by a body that is still staffing up. The DMA, meanwhile, has a Commission enforcement team already running its first gatekeeper designations, and its procedural muscle is visibly stronger.

The transmission to Monday morning matters here. A European firm integrating an American frontier model into a customised product is making three compliance calculations at once. Is the upstream model a general-purpose AI model with systemic risk under the AI Act, triggering obligations the firm partly inherits? Is the upstream provider a gatekeeper, or close to becoming one, under the DMA? And is the data flow underneath compliant with the General Data Protection Regulation, which neither of the newer regulations displaces? The answer to each question is handled by a different Commission unit, with a different enforcement timeline.

Mr Lane's speech does not say this. Central bank chief economists do not criticise sister institutions in public. What he does say — repeatedly, carefully — is that the net impact depends on how firms and labour market institutions navigate the re-orientation. That is code for: the regulation is not the only variable, but it is a variable, and it is one Europe controls.

The Gulf dimension is where this gets quietly interesting. Sovereign AI investments announced in the region over the past year treat EU regulation as a parameter, not an obstacle. If a Gulf-backed model wants European enterprise customers, it will build to AI Act specifications voluntarily, because the alternative is losing the market. Extraterritorial reach is already working. The question is whether it is working in the direction Mr Lane would want: toward European firms capturing the productivity gains, or toward European firms becoming compliant buyers of other people's models.

A regulation is what its enforcement is. The AI Act's enforcement apparatus is being assembled while the technology it governs doubles in commercial scale every quarter. Mr Lane has given the ECB's view of the opportunity. The Commission has not yet given its view of the race.