Spain bids for 100,000-chip AI gigafactory

Europe can write AI regulations, but if the computing power sits on another continent, someone else holds the off switch. Spain is now making one of Europe's most concrete bets to change that, not with new rules, but with concrete, cables, and electricity.

A factory for calculations

Think of a regular data centre as a shopping mall: thousands of independent tasks running side by side, from company databases to streaming video. An AI gigafactory is more like a car assembly line. Tens of thousands of specialised chips are wired together so they can collaborate on a single enormous calculation, like training an AI system that understands language or recognises medical images. Every component, from storage to cooling to networking, must be engineered so the chips never sit idle waiting for data (Hammerspace, Schneider Electric).

The EU plans to fund up to five of these gigafactories across Europe, each housing around 100,000 advanced AI chips (Data Center Dynamics). The call for bids opens this month, with a decision expected by late 2026 (elDiario.es). Spain is competing with a proposal centred on Móra la Nova, a small town in Tarragona province, with a secondary site near Madrid (Europa Press).

Why the Spanish state wants to co-own it

The ownership structure tells you what kind of project this is. A state investment vehicle would hold roughly 48%, with Telefónica, construction group ACS, and bank Santander each at about 16%, and quantum-computing firm Multiverse Computing taking 4% (La Vanguardia, 20minutos). Madrid wants this to behave like strategic infrastructure, not just another private cloud campus. The construction and banking partners signal that this is partly a financing-and-building problem, not only a technology one.

The money picture needs honest framing. Identifiable public commitments so far total about €1 billion, combining Spanish state funds and a contribution to the EU's joint supercomputing programme (esmartcity). The widely reported €5 billion figure is the total hoped-for investment if the bid succeeds, not money already on the table (elDiario.es).

The real bottleneck is electricity

An AI gigafactory is an industrial plant that needs land, power, and water. At the scale of 100,000 chips, power demand alone reaches the level of a small city's electricity supply. Móra la Nova's local substation currently delivers 50MW, with plans to expand toward 125MW by 2030 (20minutos).

Ireland shows what happens when grid planning falls behind. Data centres grew into such a large share of national electricity that regulators now assess each new connection against local capacity and system stress (CRU). Spain needs transmission upgrades, substation capacity, and agreements to reduce or shift electricity demand when the grid is stressed, all ready before the chips arrive.

Who actually gets to use it

The clearest answer to "why bother?" comes from Portugal's parallel bid: gigafactory capacity would be open to small businesses, startups, researchers, and public services (Government of Portugal). A hospital fine-tuning a medical AI model or a university training a language system could use European computing power without routing sensitive patient data through a non-European cloud.

France shows both the promise and the limit. The country attracted €93 billion in foreign AI-related investment pledges at its Choose France summit, partly by offering nuclear-powered electricity (Le Monde). Mistral AI's new data centre near Paris runs on 13,800 Nvidia chips, all American-designed (Le Nouvel Economiste). Europe can control where AI runs, who gets access, and what rules govern the data. Full independence from US-designed chips remains far off (EU AI Policy Monitor).

Spain has not won anything yet. If selected, operations could begin around late 2028 (La Vanguardia). Much of Europe's AI money still exists as programmes and financing vehicles, not machines that researchers can book time on (Thorsten Meyer AI). The decisive test is whether Móra la Nova's substation, grid connection, and cooling systems can be built before the AI race moves on. AI sovereignty, it turns out, depends less on who writes the rules than on who pours the concrete.