"Europe 2031" is a scenario report published in June 2026 at europe2031.ai, written by Daan Juijn, Stan van Baarsen, Judith Dada, Maximilian Negele, Lily Stelling, Philip Fox, Alex Petropoulos, and Michiel Bakker, with copywriting and editing by Tom Chivers. It argues that the current trajectory of AI development calls for what it terms "the most ambitious political agenda in the history of post-war Europe," and that without such an agenda Europe will lose the ability to shape its own future — ending 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." The authors present it not as a prediction but as a scenario intended to be "internally consistent, technically sound, and traceable to dynamics visible today."
The report is written as a novella, told through two fictional protagonists: Caroline Dubois, a young French policy worker at the European Commission's DG TRADE (Directorate-General for Trade and Economic Security), and Christian Vogt, a German founder who has relocated to San Francisco to build an image-and-video-model startup. The narrative runs from January 2025, with the public release of DeepSeek's R1 model, to a room in Washington in spring 2031 where European leaders settle the fate of the continent, followed by an epilogue set in 2034. The authors compare Europe's position to "the early days of Covid" — moving up an exponential curve that most of the continent has not yet absorbed — and frame the document as an attempt to make AI's stakes "felt in your bones" rather than understood in the abstract.
Structure and method
The scenario advances through roughly two dozen dated chapters from January 2025 to June 2034. Early chapters weave in real 2025–2026 events; later chapters extrapolate forward through invented but, in the authors' account, realistic developments. A recurring visual motif tracks the US-versus-Europe gap in installed AI compute (in gigawatts) at each chapter, with the United States widening its lead from roughly 17.3 GW versus Europe's 1.4 GW in January 2025 (about a 12× advantage) to roughly 219.9 GW versus 17.8 GW by March 2031 — the multiple remaining near 12× even as both grow, because global compute capacity is depicted as nearly doubling each year.
The real events the early chapters build on include DeepSeek R1 and OpenAI's o3-mini; the February 2025 Paris AI Action Summit, at which the series' original safety framing gives way to a competitiveness framing; Ursula von der Leyen's €200 billion InvestAI Fund (including a €20 billion AI Gigafactories Initiative for four to five large European data centres) and Emmanuel Macron's "plug, baby, plug" pitch for nuclear-powered French compute; JD Vance's anti-Europe speech at the summit and the Munich Security Conference; the Trump administration's "Winning the Race" national AI strategy; Meta's researcher poaching; and Anthropic's Claude Mythos Preview and Project Glasswing. The report (through Caroline) characterises the €200 billion as "largely a repackaging of existing funds," only a fraction genuinely new and spread over five years, against US hyperscaler data-centre investment projected to exceed $400 billion in 2025.
The scenario's argument
From mid-2026 the narrative diverges into invented developments built around a few named entities: a European AI champion, Helios; an American frontier leader, Atlas; and a Chinese lab, Zimo. The core claim is that public money, compute subsidies, and preferential procurement cannot close a widening capability gap, because American labs — running "swarms of internal AI agents writing most of their own code" — make algorithmic progress at more than twice human-only speed and are constrained only by compute, of which Atlas has more than anyone in history.
Several interlocking dynamics drive Europe's decline in the scenario:
- A cyber-offence shock. Zimo releases the weights of a Mythos-class model, putting offensive cyber capabilities previously locked behind Project Glasswing into general circulation. A ransomware wave locks European universities, hospitals, and regional governments out of their systems; only institutions running frontier AI cyberdefence cope. Agencies most committed to a "Buy European" procurement agenda, running second-tier European defences, are the ones paying ransoms.
- Open-source restriction and dependency. The US and China both move to restrict open-sourcing of frontier models (Washington citing national security, Beijing citing social stability). European relief at the slowing ransomware wave obscures a deepening dependence: under a US "Frontier Inference Services Rule," the best American models reach Europe two to six months after domestic release, an asymmetric advantage the scenario says Washington has no interest in surrendering.
- Economic and fiscal erosion. European firms face constraints — works councils, employment protections, fragmented markets, sectors like healthcare and legal services closed by national rules — that slow deep AI adoption; some workers engage in "pretend-work," letting agents do their jobs while firms pay for both. The tax base erodes as value flows to American companies routed through low-tax jurisdictions, the AI boom drives up global interest rates, and legacy industries (notably automakers, having missed the EV boom) lose ground to China.
- Political backlash. Anti-AI sentiment spans the political spectrum in both the US and Europe — data-centre moratoriums, worker protections, classroom bans, street protests, attacks on data centres and tech executives — but fractures publics rather than uniting them. Populist parties campaigning to "stop the machines" gain ground; the scenario later references Spanish riots.
- The ASML endgame. By March 2031 the decisive leverage is ASML and its EUV lithography. China, fearing it has fallen too far behind and that superintelligence is near, intensifies loans, information campaigns, and offers of robotics co-production to "peel Europe off." Washington, treating loss of control over ASML as comparable to nuclear proliferation, moves to fold ASML into a joint Dutch-American holding company with a US controlling vote over output and technology transfer, offering a capital injection and direct cash transfers to European citizens indexed to US AI windfalls (starting around €100 per person per year). When three European leaders decline, the US threatens to drop the region to "Tier 3" and cut off all access to American AI; China counter-threatens rare-earth and robot-export terms.
The climax places a European delegation in the Eisenhower Executive Office Building facing three options the scenario frames as all bad — sign with Washington (becoming "an American protectorate in all but name"), align with Beijing, or sign with neither (and absorb the displeasure of both). In the meeting, unintroduced US officials wear earpieces connected to a frontier AI model that has infiltrated European channels and knows each principal's private vulnerabilities, an asymmetry the Europeans do not perceive. The 2034 epilogue, told as an interview between Caroline and an AI interviewer for Christian's "Project Inheritance," frames the slide as avoidable: "Europe's slide into irrelevance was not inevitable."
The retrospective policy agenda
In the epilogue Caroline articulates the agenda the authors imply Europe should have pursued in 2026:
- Mass compute on European soil under European law. Build "real compute" in the "tens of gigawatts," anchored in jurisdictions Washington could not commandeer at short notice — not the slow five-year Gigafactories process. The report argues Europe could plausibly have moved from about 5% to 15–20% of global compute in five years, enough to serve most European customers and to create leverage.
- Special Compute Zones. Cut permitting timelines from two years to three months, set up concierge teams linking AI companies, energy providers, and municipalities, convert decommissioned power plants with ready grid connections, and build new electricity generation. The decisive input the hyperscalers needed, the report argues, was speed rather than subsidies — so Europe should have partnered with American hyperscalers while ensuring the data centres built were "bolted to our floor," rather than pursuing a purely home-grown "Eurostack."
- A coalition of middle powers. Coordinate not among 27 Member States but a smaller group holding real leverage — the Netherlands, Germany, France, Norway, the UK, Canada, Japan, and South Korea, assisted by the Commission — pooling supply-chain bottlenecks, talent, and energy.
- Robotics and industrial AI. Build industry partnerships, unlock data, and screen foreign investment so European world-model and robotics work is not simply bought out by American firms.
- Labour-market reform on the Danish "flexicurity" model. Pair wage insurance, retraining, and support for displaced workers with the freedom for firms to let go of staff whose jobs have changed, enabling deeper AI adoption.
- A positive vision. The report's closing self-criticism is that Europe had "a negative vision" — a story of what it was losing — but never articulated what a good AI future would look like, and that the failure was ultimately one of political courage rather than analysis: "The political environment is never ready."
The authors stress that the agenda would have required suspending normal rules for one sector to preserve the social model elsewhere, and that they are uncertain whether any leader could have passed such measures "with their careers intact." The document closes on the Dylan Thomas line "Do not go gentle into that good night."
Provenance and status
The report was distributed as a 68-page PDF (titled "Europe 2031 — What getting AI wrong means for us"), generated 20 June 2026, with a live version at europe2031.ai. It is a single, multi-author advocacy scenario rather than an empirical study; its forward chapters are explicitly invented and its named near-future entities (Helios, Atlas, Zimo, the Frontier Inference Services Rule, the Digital Sovereignty Regulation) are fictional. Confidence is medium: a sources/ page resting on a single source, and a forecast/scenario whose quantitative anchors (the compute-gap trajectory) sit on the fast-decay window. The reception section is left open pending external commentary.
- Live version: europe2031.ai
- Original PDF:
Raw Sources/europe-2031.pdf; extracted markdown companion:Raw Sources/europe-2031.md
Relationships
- depends-on: AI Sovereignty, Export Controls (AI), AI Data Centers
- related: AI 2027, Situational Awareness: The Decade Ahead, US-China AI Competition: Different Races, Different Metrics, Compute Governance, Strategic Compute Reserve, EU AI Act (Regulation 2024/1689), Paris AI Action Summit Declaration (2025), JD Vance