Europe 2031<br>Europe 2031<br>What getting AI wrong means for us<br>Daan Juijn, Stan van Baarsen, Judith Dada, Lily Stelling, Philip Fox, Alex Petropoulos, and Michiel Bakker.Copywriting and editing by Tom Chivers.
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>March 2031 - Washington, D.C.<br>Caroline splashes cold water on her face and looks at herself in the bathroom mirror. Her hands are shaking. She grips the edge of the sink and waits for it to pass. Through the small high window she can see a slice of Washington sky, flat and bright.<br>Down the corridor, six people are deciding the fate of the European continent. She does not know whether anything she said will matter.<br>She suspects it will not.
This is a scenario about Europe's impending slide into irrelevance: how AI is driving it, and what can still be done to change course.<br>It is told through the fictional characters of Caroline Dubois, a French policy worker in Brussels, and Christian Vogt, the German founder of a fast-growing AI company who has just relocated to Silicon Valley. While their personas are made up, the events they experience are grounded in existing trends, and meant to be realistic.<br>To understand how Europe is at risk of squandering the coming AI revolution, we first have to return to early 2025 - because the story we're about to tell is already well underway.<br>Listen to the story SpotifyAppleWeb<br>Summary Read as PDF About & FAQs
Jan 2025 Feb 2025 Aug 2025 Nov 2025 Feb 2026 Mar 2026 Apr 2026 Jun 2026 The Scenario<br>January 2025 - DeepSeek and ye shall find<br>Caroline Dubois’ office is buzzing. This morning, a Chinese firm, DeepSeek, released a new AI model, R1. It is cheap and effective, and – although no European firm has played a part in developing it – Brussels is in a fervour.
Caroline works on digital technology at the Directorate-General for Trade and Economic Security – DG TRADE, the part of the European Commission that deals with things like tariffs and exports controls. At twenty-eight, she has been there for three years, and feels she is earning the respect of her peers and superiors; but she, unlike them, is seriously worried about Europe’s future, and the DeepSeek news has not reassured her. She has recently visited Silicon Valley. It is 9,000 kilometres from Brussels, but it feels further. The idea that AI is sparking a new industrial revolution is a truism in California; in the European Commission offices, it is bordering on science fiction.
R1’s rise has excited her colleagues because they see it as clear proof that it is possible to train cutting-edge AI without the resources of the Silicon Valley giants. European policymakers jump at the idea that it is possible to outsmart the Americans, to be small and nimble and clever, to do better without the hundreds of billions of dollars that have been pouring into indulgent American data centres and gigantic model training runs.
The idea makes sense. DeepSeek has supposedly developed R1 at a fraction of the cost of OpenAI’s ChatGPT, or Anthropic’s Claude, but as far as anyone can tell, it is almost as powerful. It even has the new ‘reasoning’ feature that the American models have started rolling out, in which the model lays out its thought process on a virtual scratchpad. And its weights – the mathematical values that make the model what it is – are publicly available. Anyone can run the model on their own infrastructure, free from American tech dependence.
Though somewhat energised by the hopeful mood in Brussels, Caroline isn’t quite ready to join in. She carefully listens to more cautious voices who point out that big efficiency gains are hardly unheard of – Anthropic, Google DeepMind, and OpenAI find them all the time. DeepSeek has smart researchers and moved fast, but will soon find itself constrained by computing power; China simply doesn’t have enough ‘compute’ to train models ever since America restricted exports on the cutting-edge AI chips that China can’t produce domestically. They argue that finding efficiency gains is much easier if you have lots of compute to help you find them. DeepSeek’s own CEO pointed out that their main problem is the US ban on AI chip exports.
And so Silicon Valley is largely unfazed by R1. The hyperscalers do not slow down their massive AI investments. Days after R1, OpenAI releases o3-mini, a more impressive model than R1, and a sign that American progress is still rapid.
In Brussels, the o3 news barely registers. Acknowledging it would mean accepting that AI acceleration is continuing, driven by huge American companies, led by CEOs that can’t be trusted.
Caroline isn’t sure which side to take. She was in California...