Europe's AI push hits power limits

Europe’s AI problem is no longer just about models, chips, or regulation. It’s about substations, grid queues, and whether the rules Europe wrote for AI still fit the systems companies are actually building. That tension snapped into focus this week. CEPS launched a new task force on May 6 to help make the European Commission’s Apply AI Strategy real, just as fresh warnings landed on privacy, autonomous AI agents, and the power crunch behind data-centre growth. (ceps.eu) ### What changed this week? The immediate news is institutional. CEPS, a Brussels think tank with a lot of policy gravity, said it is setting up a task force to translate the Commission’s Apply AI Strategy into practical sector plans for healthcare and pharma, automotive and mobility, and government and public services. The group runs from June 2026 (ceps.eu) wrestling with implementation. (ceps.eu) ### What is the Apply AI Strategy trying to do? The strategy is the Commission’s push to get AI used at scale across key industries instead of leaving Europe stuck as the place that regulates AI while others deploy it. CEPS describes it as a turn toward adoption across strategic sectors, not just oversight. But the catch is that adoption depends on t(ceps.eu)overing that all three are contested. (ceps.eu) ### Why are people saying the AI Act is behind the curve? Because a lot of the newest AI products are not simple chatbots or fixed-purpose models. They are agents — systems that can plan, call tools, trigger actions, and operate with more autonomy across software and business processes. A Tech Policy Press essay published May 5 argues the EU AI Act w(ceps.eu)ons by risk categories that get fuzzier when an AI system becomes more open-ended and action-taking. That does not mean the Act is useless. It means the edge cases are arriving faster than the legal categories. (techpolicy.press) ### Where does privacy enter the picture? Right in the middle of Europe’s competition push. Reuters reported on May 5 that Google scientist Sergei Vassilvitskii warned EU regulators that proposals forcing Google to share search data with rivals such as OpenAI could expose private user information. One especially sharp detail came from coverage of his argument: he said the anonymisation appro(techpolicy.press) two hours. Whether regulators accept that claim or not, it shows the trade-off clearly — more competition can require more data access, but more data access can collide with privacy. (msn.com) ### Why is electricity suddenly central to AI policy? Because AI runs in data centres, and data centres run on huge, continuous power loads. Euronews highlighted a new study showing Europe’s hottest data-centre markets now face grid-connection waits of 7 to 10 years, stretching to 13 years in the (msn.com)jumped from around 13 MW in 2019 to roughly 280 to 300 MW in 2025 — around the demand of 250,000 European households. (euronews.com) ### Why is that such a hard bottleneck? Because grid capacity is not like cloud software — you cannot patch it overnight. New transmission lines, substations, and permitting take years. In places like Dublin, Frankfurt, Amsterdam, Paris, and London, the queue itself has become a barrier to building new AI infras(euronews.com) not either. (euronews.com) ### So what is Europe really balancing? Three goals that do not naturally align. Europe wants more AI competition, stronger privacy protections, and faster industrial deployment. But forcing more data sharing can weaken privacy, tightening safeguards can slow competition remedies, and scaling compute can smash into the grid. None of those trade-offs are abstract anymore — they are showing up in live policy fights this week. (msn.com) ### Bottom line? Europe’s AI push is entering its least glamorous phase — the part where legal definitions, market design, and electrical infrastructure all have to work at once. That is usually where big strategies get real. And where they stall.