Former European Central Bank President Mario Draghi wrote on Friday that Europe needs a substantial expansion of artificial intelligence data centers if it is to preserve economic sovereignty and stimulate productivity growth.
Draghi pointed to a widening productivity differential between the euro area and the United States, saying the gap grew from $9 per hour in 2018 to $21 in 2025. He cited ECB scenarios that indicate rapid AI adoption has the potential to add 0.3 to 0.4 percentage points annually to total factor productivity growth - a metric that has been roughly flat since 2022.
In his piece for the Financial Times, Draghi noted the European Union currently hosts under 5% of the worlds AI compute capacity, while the United States accounts for around 75% of global compute. He described an existing shortfall in Europe of roughly 3GW between demand and installed supply, equivalent to about one quarter of current capacity, and warned that the shortfall could expand to 14GW by 2030.
"Being cut off from AI, once the economy runs on it, would be more like being cut off from the US financial system. The effects would be catastrophic," Draghi wrote.
Although Europe controls only a small portion of the AI value chain, Draghi argued it can retain sovereignty over data. He referenced European Commission estimates that place the data economy at over 200 billion, or more than 5% of GDP, by 2030.
Draghi highlighted practical obstacles that slow capacity expansion. He noted that a data center typically takes 24 months to build in the United States but about 42 months in Germany, largely because of longer permit timelines and delays in connecting to the power grid. He also observed that the cost of building an AI data center in Sweden is only about 10% higher than building one in China.
To bridge the gap between demand and investment, Draghi proposed that European firms combine their commitments into contracts large enough to finance new data centers. He cited an example of European companies, including ASML, Capgemini and Amadeus, which have pledged multiyear purchases of Mistral's European Compute Units. Those commitments are intended to support roughly 1GW of capacity by 2030.
The proposals put forward focus on creating commercially backed demand signals and addressing regulatory and grid-related bottlenecks. They center on increasing installed AI compute in Europe and shortening the time and cost barriers that currently disadvantage on-continent build-outs.
Key Points
- Europes productivity gap with the U.S. widened from $9 per hour in 2018 to $21 in 2025; rapid AI uptake could add 0.3 to 0.4 percentage points per year to total factor productivity, according to ECB scenarios.
- The EU hosts under 5% of global AI compute capacity versus roughly 75% for the U.S.; Europe faces a current demand-installation shortfall of ~3GW, potentially rising to 14GW by 2030.
- Draghi recommends pooling corporate commitments to create financeable contracts for new data centers; a group including ASML, Capgemini and Amadeus has pledged multiyear purchases to support about 1GW of capacity by 2030.
Risks and Uncertainties
- Regulatory and permitting delays: Building times can be significantly longer in some European countries - for example 42 months in Germany versus 24 months in the U.S. - which may slow capacity expansion and affect infrastructure and utilities sectors.
- Supply-demand divergence: The gap between AI compute demand and installed capacity could grow to 14GW by 2030, creating potential constraints for industries depending on large-scale AI compute.
- Concentration of compute: With under 5% of global AI compute in the EU and 75% in the U.S., European firms and markets may face strategic vulnerabilities tied to external compute concentration.
These points underscore the policy and industrial choices Europe must confront if it seeks to host more of the AI ecosystem onshore and to ensure data-related economic value remains within the continent.