Summary — Europe 2031<br>Summary

The current trajectory of AI calls for the most ambitious political agenda in the history of post-war Europe. Unless we embark on it now, Europe will lose the ability to shape its own future. We will end up economically and politically sidelined, with values we cannot defend, social welfare systems we can no longer fund, risks we cannot address, and a Union that cannot hold.<br>Europe 2031 is a five-year scenario about Europe's impending slide into irrelevance: how AI is driving it, and what can still be done to change course.<br>To understand how Europe is at risk of squandering the coming AI revolution, the story first loops back to 2025 - and to three things it got wrong. It misjudged how fast AI would move. It misjudged how much it would change. And it misjudged its own ability to catch up.<br>How we got here — January 2025 to June 2026<br>Europe misreads the speed and scale of AI, and a series of reasonable-looking decisions deepen its dependence.<br>DeepSeek approaches the frontier cheaply, and Europe draws the wrong lesson. In early 2025, the Chinese model R1 is taken as proof that catching up is cheap and that compute barely matters - a comforting idea. But efficiency and compute compound rather than substitute: more chips make the clever shortcuts easier to find, and DeepSeek's own progress is capped by the AI chips China cannot import.<br>The AI Action Summit mostly deals in rhetoric. At the Paris AI Action Summit, the EU announces a €200 billion AI fund that is mostly repackaged money and hoped-for private investment, dwarfed by what the US is actually spending. ‘Sovereignty’ becomes the rallying cry, but actual measures lack teeth and avoid hard trade-offs.<br>GPT-5 disappoints, and the bubble narrative takes hold. When OpenAI's GPT-5 underwhelms, European sceptics read it as confirmation of an ‘AI bubble,’ and momentum stalls. Out of view, the opposite is happening: coding agents in Silicon Valley begin automating software engineering, and the leading labs start using their own models to build the next generation.<br>The governors don't use the tools. Most European civil servants are barred from frontier systems on data-protection grounds, and few of them know how to code. As a result, those meant to regulate and govern the technology often do not properly understand it.<br>Access becomes a favour, not a right. By mid-2026, some European leaders revisit their earlier scepticism. Anthropic's Claude Mythos, withheld from public release, turns out to be capable enough to reshape cybersecurity. Europe is initially left out of the defensive coalition formed around it. Soon after, a US executive order routes new frontier models through a classified review, letting Washington choose which ‘trusted partners’ get them first.<br>Europe realises it holds few of the cards. Controlling just five per cent of the world's AI compute against America's eighty, Europe has little leverage to demand access to frontier AI. Its answer is a positive-sounding tech-sovereignty package that remains too little, too late.<br>What could happen next — August 2026 to March 2031<br>Europe doubles down on sovereignty but forgets to build leverage, while the AI race between the US and China escalates.<br>2027: a Mythos-level open-source model triggers a ransomware wave, and sovereignty policies backfire. Germany and France have just proposed a bill mandating European-only AI for critical public-sector workloads. So when offensive capabilities spread to anyone who wants them, the organisations that switched to European providers in advance - and are thus running defences well behind the frontier - are the ones locked out of their systems and paying the ransoms. The wave only eases when the US and China both ban open-source frontier models, which leaves Europe more dependent on closed American ones than ever.<br>2028: AI stops reasoning in language humans can read, and Washington forces the Dutch to cut ASML's exports to China. The capability jump breaks the oversight tools regulators were relying on, and the EU AI Office - already locked in proceedings against two American developers - has no room to respond. When the Dutch are pressured to halt exports of ASML's older DUV machines to China, other Member States offer little support. The Netherlands caves, and Europe negotiates nothing in return.<br>2029: the US begins rationing frontier AI by country, and economic divergence accelerates. The growing compute shortage reaches a breaking point, and the US rations frontier inference through a tiered, country-based system that reserves most capacity for itself and a few selected allies. Most of Europe lands in Tier 2 and sees its compute allocation from US cloud companies halved. When the EU reaches for the "trade bazooka" to win Tier 1 status, the vote falls short of a qualified majority. European GDP growth begins to diverge sharply from America's: Europe owns little of the AI stack, adopts it more slowly, and gets only limited access to the...