Where the World Is Actually Building Its AI Data Centres

Artificial intelligence is a physical industry. Every model runs in a building that consumes electricity, water and land, and where those buildings get built is being decided by a short list of factors that have little to do with technology.

What determines location:

Electricity availability comes first, ahead of price. A site with cheap power and a seven-year wait for a grid connection loses to a site with moderate power available next year. Grid interconnection queues have become the industry's binding constraint in most developed markets.

Land, cooling conditions, network connectivity, political stability and permitting speed follow. Water availability is increasingly decisive as cooling demands grow and communities push back.

The established clusters:

Northern Virginia remains the world's densest concentration, built on decades of internet routing infrastructure. Its constraint is now power, with data centre demand dominating regional utility planning and provoking genuine political conflict over transmission lines and electricity bills.

Texas has become the fastest-growing American market, chosen for its independent grid, fast interconnection, abundant generation and permissive development policy. Georgia, Ohio, Arizona and the Pacific Northwest follow.

Ireland and the Netherlands, both early European hubs, imposed effective restrictions after data centre electricity consumption reached politically unsustainable shares of national supply.

The rising locations:

The Nordics, Sweden, Norway, Finland, Iceland and Denmark, offer cold climates, clean dispatchable power and political stability. Iceland and Norway have the cleanest electricity; Finland pioneered waste heat recovery into district heating.

Spain and Portugal have become serious contenders on cheap solar and wind, plus Atlantic cable landings, with water scarcity as the main constraint.

France offers nuclear-backed stable low-carbon supply, its most durable structural advantage.

In Asia, Malaysia's Johor absorbed demand that Singapore could not host, while Indonesia, India and Japan are all expanding rapidly.

The Gulf states are building aggressively with sovereign capital and cheap energy, constrained by extreme cooling loads and water scarcity.

The constraints tightening everywhere:

Electricity is the story. Data centre demand growth is forcing utilities to plan generation and transmission expansion on timelines that regulatory processes were not designed for. In several markets, residential consumers face rising bills attributable partly to grid investment driven by industrial load, which is politically combustible.

Water is the second constraint. Evaporative cooling is efficient and consumes water, and in Spain, Chile, Arizona and parts of Asia this has generated organised local opposition. The industry response, closed-loop and air cooling, uses more electricity, trading one constraint for another.

Local politics is now a genuine siting factor. Communities increasingly weigh modest permanent employment against visible resource consumption, and projects are being modified, delayed or blocked.

What comes next:

Two shifts are visible. Compute is moving toward electricity rather than toward users, which is viable for training and less so for latency-sensitive inference. And developers are increasingly building or contracting dedicated generation, including renewables, gas and, in several announced projects, nuclear.

The countries that will host the next wave are those that can offer power quickly. That list is shorter than the list of countries that want the investment, which is why energy policy has quietly become AI policy.