Bruegel proposes redirecting funding for AI gigafactories toward faster permitting and grid connections for data centers

The European Union should redirect funding for artificial intelligence gigafactories toward grants that reward faster permitting and grid connections for data centers, researchers Bertin Martens and Tillman Schenk recommend in a Bruegel report. They argue that the main obstacle to expanding Europe’s computing capacity is the time required to bring facilities online, and that public funding should help authorities address this bottleneck.

In brief The report recommends grants that reward faster permitting and the coordination required for grid connections. The authors challenge broad subsidies for gigafactory construction, while retaining an exception for a small group of strategic activities. The data used indicate that the EU will account for approximately 5% of global AI computing capacity in 2026 and 5.6% in 2031. However, the inventory excludes AI capacity in the existing cloud centers of major US providers in the EU, for which complete data are unavailable. The CADA proposal provides for acceleration zones and a permitting deadline of no more than 12 months from the submission of a complete application. It includes planning for energy needs, but the aggregated basic permit does not cover grid connection. These provisions are not yet binding obligations. The Commission justifies gigafactories by citing access to advanced computing for research, companies and public services. The authors also propose subsidizing capacity rental, as well as a voluntary label for services processed in the EU, supported by external audits.

The gap they describe is large, even under a scenario of rapid growth. The Europe2031.ai data used in the report indicate that the EU will have approximately 2 gigawatts of AI capacity in 2026, nearly 5% of the global total, compared with about 35 GW in the United States and 5 GW in China. The projects considered would bring European capacity to almost 21 GW in 2031, but its global share would reach only 5.6%, because infrastructure is also expanding elsewhere in the world.

The estimate has an important limitation: the inventory tracks projects of at least 10 megawatts and excludes AI capacity in the existing cloud centers of major US providers in the EU, for which complete data are unavailable. The 5% figure is therefore not an exhaustive measurement of all European infrastructure. The gigawatts here express the power associated with computing capacity; they do not represent annual electricity consumption or a measure that fully captures differences in chip performance.

The argument against large-scale subsidies starts with investments already planned. In the inventory analyzed by the authors, 76 of 101 projects are fully privately financed and account for 84% of planned capacity. The five gigafactories included in the report’s scenario would add approximately 750 MW, nearly 4% of Europe’s projected capacity for 2031. These are estimates of future contributions, not capacity already built. “Private capital is not the decisive constraint on expanding Europe’s AI computing capacity,” Martens and Schenk write, translated from the report.

The European Commission also justifies support for gigafactories by citing access for researchers, small and medium-sized enterprises and the public sector to the infrastructure needed for advanced models. The initiative involves privately led projects and blended financing. InvestAI, launched in February 2025, aims to mobilize €200 billion for AI, including a €20 billion European fund for gigafactories. These amounts represent investment-mobilization targets and do not prove that payments have already been made.

The authors challenge the need to finance the construction of centers to ensure this access. Researchers and public services could receive subsidies to rent capacity from private operators, they argue. They nevertheless retain an exception for a small group of strategic activities. The disagreement therefore also concerns how access to computing is purchased: by supporting infrastructure or by financing its use.

The cost of delays provides the economic rationale for their proposal. The report cites a Carnegie financial model for a hypothetical 100 MW center, in which a one-year delay in starting operations reduces lifetime value by more than 5.5%. In the same model, doubling energy prices reduces value by 4.5%. The comparison illustrates how much lost time before operations begin can matter, but it is not a loss measured across all European centers: the report notes that the results are sensitive to the model’s assumptions.

In practice, a project depends on different authorities for land, environmental, construction and electricity matters. The Commission’s proposal on the development of cloud and AI infrastructure, known as CADA, seeks to coordinate these procedures through data-center acceleration zones and one-stop information points. States would issue an aggregated basic permit for each zone, while projects would obtain the additional permits required separately. The text proposes a deadline of no more than 12 months for permitting projects in these zones, calculated from the submission of a complete application.

Access to electricity nevertheless requires a clarification regarding the report’s criticism. Martens and Schenk consider that the proposal does not resolve the grid-connection bottleneck. CADA does, however, contain provisions on assessing energy needs, integrating them into grid-development plans and coordinating authorities with transmission and distribution system operators. One-stop points could also facilitate connection applications. The aggregated basic permit nevertheless excludes grid-connection permits, according to the proposal’s recitals. Administrative simplification for the center does not automatically secure its electricity connection.

The solution proposed by the researchers is for regions to compete for grants based on the speed of permitting, with larger sums for those that move faster. The funds could come from budgets earmarked for gigafactories, while investments in grids should be prepared in advance. Public money would thus support coordination among the institutions on which projects depend. This is a mechanism recommended by the authors, not a change in EU funding that has been adopted.

Even if permits were issued more quickly, Europe would still need to obtain the chips. The report describes a global supply that is limited in the short and medium term by production capacity and recommends attracting a larger share of major US companies’ equipment to centers located in the EU. From this perspective, overly strict restrictions on the origin of components could make projects more expensive and discourage investment. This is the risk anticipated by the researchers, not an effect already demonstrated by CADA.

Their argument for locating computing in Europe starts from the vulnerability of digital services to external intervention. Installing infrastructure in the EU would give authorities more regulatory and negotiating options, but would not eliminate all dependencies: access to models could be restricted even if the equipment is located in Europe. The Commission’s proposal specifically seeks to assess such risks through four levels of cloud-sovereignty assurance, with progressively stricter requirements concerning data location, provider independence and control.

Martens and Schenk also recommend a voluntary label for AI services processed in the EU, accompanied by evidence verifiable through external audits. Providers should show where computing operations are performed, to prevent services processed partly outside the Union from being promoted as European. They also call for greater transparency about chips and capacity already installed in Europe, precisely to assess the gap between European demand and the infrastructure serving it.

The Bruegel report is dated September 2026. CADA was proposed by the Commission in June and requires adoption by the European Parliament and the Council of the EU; the measures described are not yet applicable obligations. Separately, the Commission’s page dedicated to gigafactories indicates that the official call was launched in July 2026 and estimates that construction of the first facility will begin in 2027. The timeline remains preliminary, and announced projects do not equate to computing power available to users.