Europe is losing AI-era jobs to its data rules, and the AI Act delay will not bring them back
A large US bank now grades countries red, amber or green before deciding where to add AI-related headcount, and most of continental Europe reads red. One test is whether data rules let a firm run AI on its own client records. Brussels has already pushed the AI Act's toughest rules to December 2027, so the obstacle now is data law. The desk expects London to keep winning the hires.
A large US bank has started ranking countries on a red-amber-green scale before deciding where to add staff, and most of continental Europe is coming up red. An executive at the bank told the Financial Times it is "less inclined to add headcount in countries we consider red for AI adoption," and marked the UK "greenish." That is a hiring plan. It moves the jobs AI is meant to make more productive to places where companies are allowed to use AI on their own data.
The bank's traffic-light criteria include whether a government prioritises AI and whether data-protection rules let a company use its own commercial and client data in AI applications. Continental Europe fails mostly on the second test, which has little to do with the AI Act that dominates coverage of European AI policy.
Brussels has already delayed the rules executives worried about most. Brussels' Digital Omnibus pushed the high-risk obligations for stand-alone systems under Annex III back to December 2, 2027, and delayed the equivalent rules for AI embedded in other products to August 2, 2028, with the Council signing off on June 29. A rulebook whose sharpest teeth arrive more than a year from now cannot explain what this bank's hiring committee is doing today.
Some obligations remain on schedule. The transparency obligations under Article 50, requiring disclosure when content is AI-generated, stayed close to their original schedule and took effect in August 2026, and the watermarking rules that go with them start in December. Executives can point to those as evidence Brussels is still regulating on a live clock. Transparency labelling is a one-off compliance cost. A restriction on using client data is permanent, and it decides whether a job function can exist in a country at all.
Beyond the AI Act, the barriers are narrower and less visible: data-use limits that predate the Act by years, higher energy costs, slower data-centre build-out, and, per the CERRE think-tank's Zach Meyers, weaker customer uptake of AI tools than in the US. Slaughter and May's Laura Houston states the regulatory half of that plainly: "On paper, the EU has gone further than any other major economy in regulating AI." Those costs and complexities now shape where a company runs a model on live data and where it trains one.
The jobs at stake are in back-office operations and analytics, roles AI is supposed to compress. That is the same layer of staff that moved to Warsaw, Lisbon and Bangalore during the offshoring wave two decades ago. This time, access to usable data determines the location. Wherever a bank or industrial group can legally point a model at its own client files, it can also put the resulting jobs there; wherever it cannot, the same function gets staffed somewhere else.
London is the visible winner so far. Novo Nordisk picked the city for a new AI co-innovation hub with AWS, and its chief AI officer, Anja Leth Zimmer, put the logic in almost the bank's own words: "We go where the AI talent is and where we can operate responsibly." Prysmian's chief strategy officer, Maria Cristina Bifulco, frames the same choice from the industrial side: companies "have choices about where they locate their next R&D programmes, manufacturing capacity and operational investments," and those choices increasingly track AI-friendly policy and regulatory certainty, a shift from the older calculus of wages and logistics.
The European Commission's answer is that the bloc still holds the better raw material. A spokesperson points to "world-class industries with the highest-quality data to power industrial AI models" and the bloc's planned AI gigafactories. It is also true that companies complain about EU rules on a predictable cycle, and that GDPR's arrival in 2018 did not stop foreign investment from flowing into Europe.
GDPR is an imperfect precedent. GDPR changed how companies stored and moved data, and operations teams stayed where they were. This round is different because the decision is being made at the margin, inside internal traffic-light memos, well before any gigafactory groundbreaking or industrial-data pitch reaches the executives now drafting hiring plans. Aveva's chief executive, Caspar Herzberg, says his company is still "evaluating the right AI criteria," including "AI maturity," and calls it "an issue we see impacting investment planning going forward." Those decisions are being taken now, on a faster clock than any Brussels strategy.
The Commission's gigafactories are a different kind of investment from the one the bank describes. Building a model on pooled industrial data inside a purpose-built facility is compatible with strict rules on personal data, since the data in question was never a customer's private file to begin with. The headcount question is about ordinary commercial operations, the analysts and case handlers who would otherwise process a bank's own client records through an AI system, and that is the use case continental data-protection law makes hardest.
The losers are continental service hubs and the European IT-services and telecoms firms that staff them, businesses whose model is renting out exactly the kind of headcount now being redirected to London and US cities. The shift appears in hiring plans rather than factory closures or headline layoffs: the next round of hires lands elsewhere, one traffic-light review at a time, while European energy costs are also rising because of the war in Iran.
The desk expects the Commission's next digital-omnibus proposal to focus on GDPR's rules for using personal data in AI training and processing, since that is the constraint executives are actually naming, and expects that proposal before the middle of 2027. We also expect at least one more large multinational to publicly choose the UK over the EU for an AI hub before the end of this year. Brussels' next fix will show whether it is responding with a data-protection carve-out or another AI Act delay; only the first addresses what the banks are measuring.