Artificial intelligence is usually presented as something immaterial. We talk about algorithms, models, data, and the cloud, while political debate focuses on regulation, privacy, employment, and whether Europe can compete with the United States and China.

But AI is extraordinarily physical. It needs data centres, and data centres need land, transmission lines, substations, and transformers. They consume large quantities of electricity, while cooling can require substantial amounts of water. Europe’s AI race is therefore rapidly becoming an infrastructure race.

The EU is joining enthusiastically. It is establishing 19 AI Factories across its supercomputing network, with overall European investment in supercomputing infrastructure and AI Factories between 2021 and 2027 expected to reach €10 billion. In July, it launched a call for up to seven AI Gigafactories, backed by as much as €10 billion in EU and national funding and intended to unlock at least €20 billion in private investment.

This is industrial policy. But it is also a massive new claim on public infrastructure.

AI meets the grid

According to the International Energy Agency (IEA), global data-centre electricity consumption is projected to more than double, from around 415 terawatt-hours (TWh) in 2024 to approximately 945 TWh in 2030. European consumption is expected to rise by more than 70 per cent.

The problem is not simply aggregate consumption. Data centres concentrate enormous electricity demand at particular locations, and they can be built much faster than the grids that supply them. The IEA notes that waits for a grid connection in parts of Europe can already stretch for years.

At the same time, Europe must connect renewable generation, build housing, electrify industry, and accommodate electric vehicles and heat pumps. Data centres could account for around 10 per cent of EU electricity-demand growth through 2030.

This creates a political question often disguised as a technical one: when grid capacity is scarce, who gets it first? A housing development? A manufacturer trying to electrify? Or an AI data centre?

The market cannot answer neutrally. Electricity grids are regulated infrastructures built through decades of investment, planning, and political decision-making. Allocating access to them also distributes economic opportunity.

AI is turning electricity policy into industrial policy.

Infrastructure constraints are already reshaping Europe’s digital geography. A Reuters analysis found that hyperscale data centres are moving away from traditional technology hubs in search of cheaper electricity and land and quicker grid connections.

According to data cited by Reuters, the average European data-centre site planned for 2026–28 is 175 kilometres from a major city, compared with 46 kilometres for projects delivered between 2022 and 2025. AI may be imagined as borderless, but its infrastructure is becoming intensely geographical.

That could benefit regions bypassed by previous waves of technology investment. But it raises an old development question: what do communities receive in return for reorganising local resources around externally generated demand?

Northern Germany provides a revealing example. Schwarz Group, the owner of Lidl, recently announced plans to invest as much as €5.6 billion in a data centre near Rostock. Its initial 240-megawatt capacity would be equivalent to the average electricity consumption of approximately 600,000 households. Yet the facility is expected to create only around 120 permanent jobs.

That does not make the project undesirable. It is planned to run on renewable electricity during normal operation and to use closed-loop water cooling, while its excess heat could potentially warm homes in Rostock. These are precisely the kinds of benefits governments should demand.

But the comparison reveals the distinctive political economy of AI infrastructure. Data centres are extraordinarily capital- and resource-intensive but relatively labour-light. The traditional development bargain — provide infrastructure because investment will generate large numbers of jobs — becomes much less straightforward.

Public infrastructure, private rents

Europe has good reasons to expand its computing capacity. Dependence on American cloud and AI companies creates strategic vulnerabilities, while European researchers and businesses need access to advanced computing. The question is not whether governments should intervene. They already are. It is on whose terms.

The EU describes AI infrastructure in the language of sovereignty and strategic autonomy. Yet its economic value can still accrue disproportionately to technology companies, cloud providers, data-centre operators, and investors. The state reduces risk and expands capacity; society supplies much of the underlying infrastructure; private actors are positioned to capture much of the resulting value.

The pattern is familiar. Railways, telecommunications, and energy systems have all combined public infrastructure with private accumulation. AI intensifies the issue because resource demand is expanding so quickly. The Commission now aims to triple EU data-centre capacity by 2035, while acknowledging the resulting pressures on energy and water.

Europe therefore needs a different AI bargain. Where large facilities require grid expansion, developers should contribute appropriately to its cost. Projects should be encouraged to add renewable generation rather than simply compete for existing clean electricity, and waste heat should be captured wherever feasible.

Public financing should also carry public-interest conditions. Universities, researchers, start-ups, and smaller businesses should receive meaningful open access to publicly supported computing capacity. Governments should seek regional benefits, along with mechanisms through which the public can share substantially in the economic returns its investment helps generate.

Most importantly, Europe should distinguish among competing demands for scarce infrastructure. Not every data centre becomes socially valuable simply because it belongs to the digital economy. Where electricity, water, or grid capacity is constrained, governments must decide which uses generate the greatest public benefit.

That is not hostility to AI. It is what a serious, social-justice-oriented industrial policy requires.

The AI debate has spent too long pretending that the technology lives in the cloud. It does not. AI occupies land, runs through cables, and consumes electricity and water. Its supposedly immaterial economy rests on intensely material — and frequently public — foundations.

Europe should build AI capacity. But if citizens finance the supercomputers, expand the grids, and accommodate the resource demands, they should meaningfully share in the benefits as well.

Artificial intelligence may be digital. Its political economy is anything but.