Summary
- Structure Research forecasts AI-focused data centre capacity rising from about 2.3GW to 150GW by 2030.
- The forecast places power availability and infrastructure delivery at the centre of AI capacity growth.
- Europe’s challenge is turning sovereign AI ambition into grid-connected, cooled, permitted capacity.
Structure Research has forecast that AI-focused data centre capacity could rise from about 2.3GW today to 150GW by 2030, a projection that shifts the discussion from compute demand to the physical infrastructure needed to support it.
The forecast appears in the firm’s AI Infrastructure Report, with a public release framing the market around rapid capacity growth, power availability, capital concentration, and changing workload patterns.
Even if the final number changes as the market develops, the direction is clear. AI infrastructure demand is now being measured in gigawatts, which means power systems, land, substations, cooling plant, equipment manufacturing, and permitting are shaping the pace of digital infrastructure growth.
Forecasts do not build substations
Capacity forecasts can move faster than infrastructure. GPU roadmaps, cloud contracts, and AI model demand can change within months. Grid connections, transformers, switchgear, mechanical plant, planning consent, and construction programmes move on longer cycles.
A 150GW scenario would require a vast expansion of electrical and mechanical infrastructure. Data centres would need secure grid capacity or credible behind-the-meter generation, plus redundancy systems, cooling equipment, fibre routes, skilled labour, commissioning teams, and operations staff. The industry’s shortage is not limited to chips.
Europe’s position is mixed. The region has strong enterprise demand, cloud regions, research capability, and political interest in sovereign AI. It also has constrained data centre hubs, fragmented permitting, high energy costs in some markets, and increasing scrutiny over water, heat, emissions, and local grid impact.
AI training and inference will pull capacity in different directions. Training campuses may favour large power-advantaged sites with enough land and energy to host dense clusters. Inference capacity may need to sit closer to users, applications, and regulated data sets. Europe is likely to need both, which makes the infrastructure map more complicated than a simple move to cheaper power regions.
Power becomes the commercial filter
The forecast strengthens the value of secured electricity. Land without a credible connection is losing appeal. A smaller site with signed grid capacity, a route to substation delivery, and a clear cooling design may be more valuable than a larger speculative plot with uncertain energisation.
Investors are already trying to distinguish between theoretical pipeline and deliverable megawatts. AI demand makes that distinction more severe because customers want capacity quickly and at high density. Announced pipelines can inflate confidence while actual commissioned supply remains blocked by utilities, equipment lead times, and permits.
Cooling is the second filter. Many AI deployments will need direct-to-chip liquid cooling, rear-door heat exchangers, or other high-density approaches. Those systems change white-space design, commissioning, maintenance, leak detection, fluid management, warranties, and staff training. Capacity measured only in megawatts can hide a substantial mechanical transition.
The European policy environment will also shape where AI capacity lands. Energy reporting, water scrutiny, heat-reuse expectations, NIS2, critical infrastructure rules, and local planning controls all affect whether capacity can be delivered and defended.
Structure Research’s forecast should be read as a stress test for the infrastructure system behind AI. If demand rises anywhere near that level, markets will be judged less by ambition than by their ability to deliver power, cooling, permits, supply chains, and operational resilience.
