ECB's Lagarde to Europe: build your own AI or risk being 'cut off' by US or China

Christine Lagarde, president of the European Central Bank, said in Vienna on Monday that Europe must build its own AI capacity or accept a future in which foreign powers, chiefly the United States, which hosts 75% of the world's AI compute, can pull the plug on services the continent now depends on. The warning came at a Hofburg forum on European resilience, and the prescription that followed was deliberately modest: not a moonshot, but a buildable response that Lagarde is asking Europe to fund before the choice closes.

The asymmetry Lagarde named is stark. Last year, the US produced 59 notable AI models and China 35; France and the UK each produced one. The US holds three-quarters of the world's AI computing capacity; Europe holds 5%. On current trends, the gap between Europe's datacentre capacity and its own demand is projected to grow more than sixfold within a decade. Euro area firms are already planning to devote around 10% of total investment to AI in 2026, and the ECB estimates AI-related borrowing accounted for about a quarter of the growth in corporate credit in the first quarter of the year.

For Lagarde, this is not a competitiveness problem. It is a sovereignty problem, and the reason has three parts.

The first is data. A model trained on, or run against, a European firm's data lets its supplier learn that industry. Outsourced AI is, in this reading, a continuing transfer of know-how to whoever runs the model, an inversion of the usual argument that buying foreign tech is cheaper than building it. The second is access. Withdrawal of AI services would hit every sector at once, because the technology is now embedded in border screening, tax audits, train networks, hospital wards, and bank payments. The third is leverage. "No trade partner has ever held over Europe" the kind of leverage that AI dependence creates, Lagarde said in her speech, usable in any negotiation over tariffs, digital taxes, or export controls. She added that euro area households hold about €440 billion, roughly $480 billion at recent exchange rates, in US technology firms, and that large US hyperscalers issued more than $100 billion in bonds last year, most of it long-dated, now accounting for close to a tenth of new euro bond issuance by non-financial corporations. European savings are financing the very infrastructure the continent is growing dependent on.

The choice Lagarde framed is "awkward." Europe can throttle adoption to protect its data and lose a decade of productivity, or adopt fast and become dependent on powers that may weaponise access. Almost half of euro area firms that considered AI and decided against it cited data protection concerns, evidence that the throttling path is already in use. The ECB estimates AI could lift euro-area productivity by up to 4% over a decade; a separate Draghi report estimated that a 2% rise over a similar horizon would cover up to one-third of the fiscal cost of Europe's strategic investments. The workforce will shrink by more than a million people a year over the next 25 years, and existing budgets and EU instruments leave more than €100 billion a year, roughly $110 billion at recent exchange rates, of the public share of strategic investment needs uncovered.

Lagarde's prescription is not to match US frontier labs. It is to build domestic models that are "good enough" to carry out most tasks, and to back them with a sixfold expansion of European datacentre capacity. She anchored the argument in a historical parallel: the 1873 Vienna stock-exchange collapse, and the Hobsbawm-era "age of capital" in which European savings financed American railroads and industry, and the growth those investments produced stayed on the other side of the Atlantic. The structural difference now, she argued, is that the technology being financed is also a lever that can be turned against the financier.

The geopolitical backdrop is not abstract. Lagarde named Trump-administration tariffs, demands over Greenland, and US troop withdrawals from Europe as having shaken trust, even though the EU and US remain key allies. She also acknowledged, in separate remarks covered by the same Vienna press, that "there is much debate about whether the expansion of AI in the United States is a bubble," and that a bubble would also hit Europe.

The next test is whether the build-out happens at the speed the gap requires. European household savings sit in US tech equity, and US hyperscalers sit in the European bond market; the money is already flowing the wrong way. Lagarde's argument is that Europe does not need to win the frontier model race to avoid the worst version of the awkward choice. It needs to build the domestic compute and the domestic models before the sixfold gap forecloses the option to do so.