Worth a Read
Rules per Gigawatt — the Fuel Europe Didn’t Buy
Deeptech investor Michael Jackson argues Europe has already lost the AI race on structure: a fraction of the compute, a fraction of the capital, and a regulatory rulebook finished before the infrastructure was started. The numbers behind the argument, the timeline that makes it sting, and where it meets our own boards.
Source: Sifted Read the original →
Jackson is an American venture partner at Multiple Capital, a Frankfurt-based fund of funds, and one of the loudest contrarian voices on European tech. His argument is structural, not cultural: when Europe's flagship model lab raises a billion, its American competitors raise five to ten times as much in the same season; the United States operates roughly three times Europe's data-centre capacity; and the chips those centres are built from are allocated by an American company. A European founder entering the frontier-model race, in his framing, is under-armed on every input that decides the outcome. He is notably not pessimistic about European founders — his case is that the market structure around them cannot supply the fuel.
The numbers behind that structure are stark, and EUobserver's Brussels reporting in February 2026 laid them out: the EU holds about 5% of the world's AI computing power, against roughly 74% for the United States and 14% for China. Normalise by economic weight and the gap becomes the story — Europe produces close to a fifth of world GDP on a twentieth of the world's AI compute, meaning the US runs roughly an order of magnitude more compute per unit of economic output. The capital pipeline that would close the gap points the other way: about €1.33 trillion of US venture investment across 2020–2025, a third of it into AI, against €252 billion in the EU with a much smaller AI share. Germany — the continent's largest economy — holds only a single-digit slice of Europe's single-digit share.
Then the timeline that gives the argument its edge. In the same window in which that compute gap blew out, Europe's regulatory production line ran at full capacity: the AI Act entered into force in August 2024, its prohibitions took effect in February 2025, the general-purpose-model obligations in August 2025, and the heavy high-risk regime lands this August — on top of the GDPR, DSA, DMA and Data Act stack already in place. Jackson's sharpest jab is aimed exactly here: he calls the AI Act premature, a rulebook completed with visible pride before the thing being regulated had been built at any scale on European soil. The world's most complete AI rulebook now governs five percent of the world's AI compute.
The core idea Compute is the fuel of the 21st-century economy the way fiber was in 2000 — and Europe spent the build-out years writing traffic laws for roads it never built. Five percent of the world's AI compute under the world's most complete AI rulebook is not a policy position; it is a market-structure outcome.
The counter-case deserves its sentence too: Europe's bet is that trust, interoperability and a single rulebook become an advantage once the technology commoditizes — that the race is a marathon in which the early gigawatts matter less than durable adoption. The chart that would confirm it is Europe's compute share and token share rising together. Both are now measured; neither is currently rising.
Where it meets the Closelooknet frameworks
Compute is a 21st-century input the way fiber was in 2000 and 5G spectrum was in 2019 — infrastructure races are decided in the build-out years, and whoever sits out the build-out imports the capability afterwards, on the winner's terms. That is the lens of our Sovereign AI 101 and the reason the Euro-AI Sovereign 50 tracks the companies trying to close the gap from the supply side. And since this week, the demand-side mirror runs daily on our own boards: the Token Split Pulse measures whose models the world's builders actually run — European models currently carry well under one percent of open-market token demand, against roughly two-thirds for Chinese models. Five percent of the fuel, half a percent of the product: the two numbers are the same argument, measured at opposite ends of the pipeline. Our standing cut of the sovereignty build-out — from semis to cloud — is the companion read.
Closelooknet keeps this as a market-diary observation, not a recommendation — the signal is which inputs decide the race, not a call on any single name or policy.
Worth a Read points you to another writer's published work; the synthesis above, and any errors in it, are Closelooknet's, not the source's. Closelooknet publishes a market diary, not investment advice — circumstances differ; consult a licensed advisor before acting.