Summary
- Global installed capacity could reach as much as 340GW by 2035 as hyperscalers expand AI infrastructure.
- Europe’s share is forecast to decline from approximately 13% to 10% by 2030.
- Grid waits of up to seven years are pushing developers towards brownfield energy sites and less constrained markets.
Europe’s share of installed global data centre capacity is forecast to fall from approximately 13% to 10% by 2030 as the US and China convert AI demand into new infrastructure more quickly.
A study from Roland Berger projects that global capacity could reach as much as 340GW by 2035, driven by investment in AI training, inference, cloud services, and increasingly automated workloads.
European developers face a combination of slow grid connections, higher electricity prices, limited powered land, permitting complexity, and constrained equipment supply. New connections can take up to seven years in some markets, placing utility delivery well outside the construction timetable sought by large cloud customers.
Power rewrites the location map
Fibre, latency, customer density, and cloud interconnection have traditionally guided data centre location. They remain essential, although electricity availability now determines whether many proposed campuses can proceed at all.
Established markets offer deep network ecosystems and concentrations of customers, but their grid and land constraints are pushing development into secondary regions. Madrid, Milan, Warsaw, the Nordics, and parts of France are attracting schemes that might previously have defaulted to Frankfurt, London, Amsterdam, Paris, or Dublin.
Roland Berger identifies former power stations as one route through the power constraint. Retired generation sites can provide industrial land, transmission infrastructure, roads, and water systems, allowing developers to reuse parts of an existing energy footprint.
The conversion is rarely simple. Generation connections export power, while data centres import it, requiring new protection studies, network analysis, and transformer arrangements. Environmental remediation and the physical condition of legacy assets can reduce the schedule advantage.
The Roland Berger study examines capacity growth, finance, supply chains, and Europe’s position.
Capital spending strains the supply chain
Hyperscaler capital expenditure could reach approximately US$750 billion in 2026, according to the study, with much of the spending directed towards data centres and AI infrastructure. Construction at that scale is placing simultaneous demand on semiconductors, memory, transformers, switchgear, turbines, generators, cooling systems, and specialist contractors.
Long lead times have encouraged customers to reserve manufacturing slots and place overlapping equipment orders. That behaviour can secure delivery for individual projects while making the wider market harder to read, because nominal demand may exceed the quantity ultimately installed.
A slowdown in AI revenue growth could produce abrupt changes in orders even where long-term computing demand remains strong. Contracted, powered projects are better insulated than speculative land positions, while suppliers exposed to a small number of hyperscale customers may experience greater volatility.
Europe’s dependence on grid electricity gives its projects a different supply-chain profile from US campuses considering large amounts of on-site gas generation. European facilities remain exposed to transformer, switchgear, cable, generator, and cooling shortages, with network reinforcement adding another layer of procurement.
A smaller share of global installed capacity does not exclude European companies from the resulting investment. The region has established capability in electrical distribution, automation, cooling, turbines, generators, engineering, construction, and infrastructure finance, allowing suppliers to serve projects built elsewhere.
Export success does not replace domestic computing capacity. The location of data centres influences latency, data governance, access to AI compute, and the degree to which critical services depend on infrastructure operated under other jurisdictions.
Domestic facilities alone cannot create European technology sovereignty because processors, cloud platforms, software, and AI models remain concentrated among a small number of global companies. A shortage of local capacity nevertheless adds another dependency and can constrain the development of European services.
Permitting reform, transmission planning, connection-queue management, and brownfield reuse will all influence the region’s position through 2030. Individual project subsidies have limited value where a campus cannot obtain power or spend years moving between uncoordinated approvals.
Local opposition is also influencing delivery. Large campuses consume visible amounts of land and electricity while creating relatively small permanent workforces, leading communities to seek clearer benefits through heat reuse, grid investment, industrial regeneration, or local supply-chain participation.
Those benefits require functioning infrastructure rather than headline commitments. A district-heating connection needs an off-taker and funded pipe network; a grid upgrade needs an agreed delivery programme; and a brownfield redevelopment needs remediation and a viable technical design.
Europe’s declining share reflects a gap in delivery speed rather than a lack of announced demand. The region’s position will be decided by how quickly grid applications, permits, equipment orders, construction, and commissioning can be converted into tested capacity.
