Draghi Urges Europe to Build Large-Scale AI Data Centres to Defend Data Sovereignty

Mario Draghi, former Italian prime minister and former European Central Bank president, published a long article arguing that Europe needs to build large-scale artificial intelligence data centres to safeguard European data sovereignty. In 2024, Draghi released the report The Future of European Competitiveness, which set out a series of major reform measures, including increased investment in strategic areas such as artificial intelligence and defence. The report was welcomed at the time by the European Commission and member states, which said raising Europe's competitiveness was the EU's top priority.

In the article, Draghi said the EU accounts for less than 5% of global AI computing power, while the United States accounts for as much as 75%. Citing estimates, the article said the productivity gap between the euro area and the United States will widen from 9 US dollars per hour in 2018 to 21 US dollars per hour in 2025. The European Central Bank forecasts that rapid adoption of artificial intelligence will raise total factor productivity growth by 0.3 to 0.4 percentage points a year, while since 2022 Europe's total factor productivity growth has been close to zero.

The article argues that data is an area where Europe can still maintain sovereignty. The European Commission estimates that by 2030 Europe's data economy will exceed 800 billion euros, accounting for more than 5% of gross domestic product. To make full use of public-sector data, health records, statistics agency data and industrial information generated by manufacturing, Europe needs to control the storage and processing of data, which requires building large-scale AI data centres. The article notes that data centre construction has sparked debate, with local communities concerned about environmental costs and rising energy bills, while most of the data centres being built in Europe serve US technology giants. In its Cloud and AI Development Act, the European Commission has proposed a tiered regulatory approach under which the most sensitive uses must operate under European control, while US operators can provide most computing power through "sovereign cloud" services operated in Europe.

Amazon, for example, charges 15% more for its sovereign cloud service in Germany than for its standard service. On the supply side, building a data centre takes 24 months in the United States but 42 months in Germany, mainly because approval and grid connection processes take longer. High electricity prices make the cost of building and operating data centres in parts of Europe twice as high as in China, while the cost of building an AI data centre in Sweden is only about one-tenth higher than in China.

The article says Europe's main obstacle is fragmented demand. Building data centres requires huge capital investment, and investors are only willing to commit funds once they have reliable long-term offtake contracts. In the United States, hyperscale cloud service providers and frontier AI laboratories can provide such contracts; in Europe, demand is spread across millions of enterprises, and even large companies tend to buy computing power on an annual basis. Nscale, a new-generation cloud company operating in Europe, sells most of its computing power to US buyers. European companies including ASML, Capgemini and Amadeus have committed to multi-year purchases of Mistral AI's "European computing units", providing funding support for reaching 1 gigawatt of computing capacity by 2030. The article argues that a large and concentrated group of buyers needs to be cultivated, pooling medium-sized procurement commitments into large contracts sufficient to support financing.