AI Data Centres Are Becoming Energy Infrastructure Projects

Google’s latest European data-centre investment comes with something that would have looked unusual in a technology announcement a few years ago: a 22-year nuclear power agreement. The company announced a €13 billion investment in AI infrastructure in Finland in September, including three new data centres in the north of the country. Google also agreed to buy as much as half of the output from one Fortum nuclear plant under a long-term contract.

Data centres once entered local planning debates mainly through land, fibre connections and jobs. AI has pushed electricity to the centre of the decision. A hyperscale facility now arrives with power demand large enough to affect grid investment, generation contracts, storage and the political argument over who receives scarce electrical capacity.

Cities and regions are no longer hosting buildings full of computers. They are negotiating with industrial energy consumers.

Northern Finland offers more than cold weather

The climate helps. Colder outside temperatures reduce part of the energy required for cooling servers, one reason Nordic countries have attracted large computing facilities for years. Electricity supply matters more as AI infrastructure expands. Training and operating large AI systems requires dense computing hardware that runs for long periods. The resulting load is less flexible than many commercial users. A data-centre operator does not want a region to promise abundant electricity today and discover after construction that grid congestion prevents the site from expanding.

Google’s nuclear agreement addresses supply over decades rather than buying electricity only through short-term markets. Fortum gains a large customer. Google gains greater certainty over a major operating input. Finland gains an investment that the government expects to add €3.6 billion to GDP during construction in 2027 and 2028. The infrastructure around the servers now forms part of the investment case.

The grid connection is becoming a planning constraint

Europe already has cities where data-centre demand competes with housing, electrified transport and industry for grid capacity. Adding a large computing site therefore affects more than its own electricity bill. Utilities need substations, transmission capacity and sometimes new generation. Local authorities need to decide where the site belongs in land-use planning. Governments have to judge whether attracting computing infrastructure justifies committing power that another industry also wants.

The calculation gets harder when several developers arrive at once. A region with cheap low-carbon electricity attracts one data centre partly because power is available. The first investment encourages others to look at the same location. Connection queues grow. Electricity prices and network requirements change. An advantage based on spare capacity begins disappearing as investors use it. Northern Europe has started to expose that cycle earlier than many regions because data centres arrived there before the latest AI build-out.

Long-term power contracts change who develops generation

Large technology groups now have enough electricity demand and enough balance-sheet strength to sign contracts measured in decades. That gives them a role once associated more closely with utilities and heavy industry. A nuclear plant offers steady production that suits round-the-clock computing. Wind and solar provide cheaper generation during favourable conditions but need balancing, storage or complementary supply. Batteries cover shorter fluctuations. Grid connections determine whether the electricity reaches the site at all.

An AI campus therefore pulls several infrastructure decisions into one project. Technology companies have also started investing directly in nuclear technologies, renewable projects and storage rather than waiting for utilities to expand generation independently. Cities courting AI investment need to examine the complete energy package rather than the building permit.

Waste heat gives the city another piece of the equation

Servers turn electricity into computing output and heat. Nordic cities have spent years experimenting with feeding that heat into district heating systems. Larger AI facilities increase the amount potentially available, although temperature, distance and seasonal demand determine whether recovery makes economic sense.

A well-located facility therefore interacts with urban infrastructure in both directions. Electricity enters the site. Recoverable heat returns to nearby buildings or industrial users. Poor location makes the same idea much harder.

A data centre placed far from heat networks because land is cheap needs additional pipes and pumping infrastructure. A city that approves sites before considering heat use loses part of the potential benefit. Planning authorities increasingly need energy maps alongside zoning maps.

Local politics will follow the electricity

Data centres produce fewer permanent jobs per megawatt than many traditional industrial plants. Residents therefore judge them differently when local electricity prices rise or utilities announce major network investments.

Promises around tax revenue and construction employment compete with questions about power consumption, water use, land and grid access. Finland currently has enough electricity capacity for the government to welcome another large programme. Other European regions face tighter conditions. AI demand is making those differences commercially important.

Companies selecting sites now compare not just tax rates, connectivity and property costs but generation mix, permitting speed, transmission constraints and the likelihood that enough power will still exist when the second phase of a campus opens.

The next generation of data centres will still contain servers. Their economics increasingly start several kilometres away, at the power station and the substation.