How can Europe build sovereign AI computing capacity?

The European Union needs to massively scale up its artificial intelligence computing infrastructure if it wants to remain competitive and achieve some degree of AI sovereignty. Even with all data centre projects currently in the pipeline, its share of global AI infrastructure will barely exceed 5% by 2031, compared to over 70% for the United States and 15% for China. Local capacity is necessary for AI sovereignty, and this is of pressing importance as geopolitical tensions over control of and access to AI infrastructure and products grow. The EU should therefore maximise domestic compute capacity, regardless of the origin of these resources.

Private capital is not a constraint on Europe’s AI compute buildout. More than 70% of currently planned investments are privately financed. Speed-to-operation and grid access are more important constraints. Instead of subsidising private investments, the EU subsidies could be redirected to competitive time-to-permit grants that should be allocated to regions that move faster. Policy should also prioritise anticipatory grid investment and faster grid connections.

To circumvent global AI chip production capacity constraints, the EU could incentivise US hyperscalers – that buy the bulk of global chip production – to install a larger share of their supply in EU data centres. Stringent local content requirements for AI infrastructure risk, such as in the proposed Cloud and AI Development Act (CADA), will increase costs, reduce competition and may reduce investment. Rather than putting sovereignty conditions on AI infrastructure inputs, it should put these on AI services outputs, for example with voluntary ‘computed in the EU’ labels on services sold to EU customers, in line with the least stringent CADA sovereignty criteria. Hyperscalers’ disclosure of AI chips installed in the EU, and AI workloads computed in the EU, could also contribute.