Europe could continue building datacentres throughout the rest of the decade while losing ground in the global market, exposing the distance between political ambitions for artificial intelligence sovereignty and the infrastructure available to deliver them.

A study from Roland Berger forecasts that Europe’s share of installed global datacentre capacity will fall from roughly 13% to 10% by 2030 as construction proceeds faster in the United States, China, and several emerging markets. Global capacity could reach around 340GW by 2035 under the consultancy’s modelling, driven heavily by AI training and inference.

The estimate does not indicate that European capacity is shrinking in absolute terms. Instead, it suggests that familiar barriers involving grid connections, electricity prices, planning, and construction will compound while investment accelerates elsewhere.

That relative decline reaches beyond property and hosting markets because computing has become an input into research, industrial production, healthcare, finance, public administration, and software development. A region with limited capacity may still buy services from abroad, although it gains less control over investment, data location, energy use, and the businesses forming around the infrastructure.

Electricity decides which projects are built

Datacentre proposals can attract capital and public announcements long before they have a deliverable electricity connection. In several European markets, developers may wait years for sufficient power, while substations, transmission lines, and generating projects move through separate approval processes.

Roland Berger identifies grid access as the leading obstacle reported by European industry participants, followed by energy cost and regulatory complexity. Connection periods can extend towards seven years in constrained areas, which is poorly matched to demand for new AI infrastructure and the financial expectations of investors.

Cloud providers can direct spending towards regions offering faster delivery, while AI companies often locate engineering and model development near available computing capacity. Suppliers of cooling, switchgear, backup systems, and construction services then follow the largest projects, carrying economic activity beyond the datacentre itself.

Europe retains valuable advantages, including skilled engineering, dense fibre networks, established enterprise demand, strong renewable resources in some regions, and a large market for locally hosted or regulated services. None of those assets can compensate for unavailable electricity at the site where a customer needs it.

Sovereignty cannot be purchased through subsidy alone

The European Commission wants to triple EU datacentre capacity within five to seven years and meet projected demand by 2035. Its proposed cloud and AI legislation would improve access to sites and accelerate development, while AI factories and gigafactories are intended to create publicly supported computing capacity.

Public funding may reduce financing risk, but it cannot bypass network engineering or shorten the manufacturing time for electrical equipment. A credible programme requires coordinated investment in generation, transmission, distribution, fibre, water, planning teams, and specialist skills.

There is also a risk in treating maximum capacity as the sole measure of success. Roland Berger notes that hyperscaler capital expenditure is approaching or exceeding operating cash flow in some cases, leaving scope for a correction if AI demand, customer revenue, or hardware utilisation disappoints. European governments could approve weakly located projects just as the wider market becomes more selective.

Facilities designed around flexible workloads, heat reuse, modular expansion, and low carbon power may create more durable value than speculative campuses built around headline capacity. European manufacturers can also benefit from global construction through exports of electrical systems, cooling equipment, power electronics, and industrial automation.

However, supplier exports are not a substitute for regional compute. Companies and public bodies requiring European jurisdiction, low latency, or control over sensitive data need capacity within the continent. A persistent gap would increase dependence on foreign cloud platforms even as European regulation demands tighter governance of those services.

The projected fall from 13% to 10% remains a forecast rather than an inevitable outcome, while market share alone does not measure resilience. It nevertheless exposes the difference between declaring computing strategically important and building the power, planning, and industrial base required to support it.