ContextLecture in Vienna to an audience of policymakers and economists; the ECB keeps prices stable in the euro area and faces weak investment and growing strategic dependence.
Why it mattersTreats AI as a capital-intensive buildout and warns that Europe needs its own models, infrastructure and financing to avoid dependence.
- Europe depends on foreign AI models and infrastructure and risks being cut off.
- Closing the data-centre gap could cost hundreds of billions of euros.
- Savers are exposed to US tech giants and a sharp correction.
- Develop good-enough models running on European infrastructure.
- Draw the whole capital market into the buildout, from bonds to securitisation.
- Let companies adopt AI on its merits through a world of many, including open, models.
Read the speech summary
1. A lesson from the 1870s
Speaking in Vienna, Christine Lagarde recalled the Gründerzeit before 1873, when European savers financed the technologies of the day, the Suez Canal and, in far larger sums, American railways, in what Hobsbawm called the age of capital. After the 1873 crash, the assets and the growth stayed on American soil, and America overtook Europe while Europe stagnated for two decades. Today European savers are again funding the defining technology, AI, mostly in the US: about 440 billion euros of American tech equity sits in euro area household portfolios. She asks whether Europe will capture a fair part of the gains this time.
2. Why Europe must adopt fast
Europe is doing better than its reputation: euro area firms will put about 10% of investment into AI in 2026, AI-related borrowing explains about a quarter of business credit growth in the first quarter, and worker use has doubled in two years to over half. Yet US digital investment grew twice as fast, and US workers use AI two to three times more. Speed matters because ageing strains the social model, over 100 billion euros a year of public investment needs are unfunded, and the working-age population is set to fall by over a million people annually, so only faster productivity growth can pay for it; ECB estimates say quick adoption could raise productivity by up to 4% within a decade. Europe also bears the cost of the boom regardless of whether it captures any gain: US hyperscalers issued over 100 billion dollars of bonds last year, lifting long-term yields that euro area rates follow.
3. Why importing is not enough
Unlike in 1873, the companies are now built elsewhere: the US produced 59 notable AI models last year and China 35, France and the UK one each, and the US hosts three quarters of AI computing capacity against Europe's 5%. AI dependence differs from earlier technologies for three reasons. Data: a model reads a firm's data, so its supplier could learn what an industry knows and compete with it, and nearly half of firms that rejected AI cited data protection. Access: AI will soon screen goods, audit tax returns, dispatch trains and clear payments, so cutting or changing access would hit every sector at once, a leverage no trade partner has held. Frontier: in pharma, finance and defence a small edge in capability brings outsized gains and winner-takes-all dynamics. Europe thus faces a choice between holding back and forgoing growth, or adopting quickly and losing economic freedom.
4. Building sovereign capacity
Her answer has three steps. Build far more computing capacity, since Europe's data-centre gap could grow more than sixfold in a decade and the new gigafactories will fill only a fraction. Develop good-enough open models running on European infrastructure, building on Mistral and a San Sebastián firm's compressed model, because the best open models today are Chinese and their terms can change. And secure access to the frontier by being indispensable in the supply chain, as ASML is, while innovating to find the next link, since rivals will try to replicate it. She also warned that a correction in the US AI boom would reach Europe through investment funds exposed to US tech, a task for supervisors, though banks are well capitalised.
5. Capital: more of it, and a different kind
All this comes back to capital. Closing the data-centre gap could cost up to 600 billion euros, and the US has drawn on bonds, private credit and securitisation while Europe leans on banks. New models need patient equity; Mistral's record raise last week is one round of the many needed. Europe's households save about 1.4 trillion euros a year, so the savings and investments union is what lets those savings be deployed at home. AI, she argues, can be the project that drives European capital markets forward, as railways did in the US.