Spain chases AI chip gigafactory

Europe has spent years proving it can regulate artificial intelligence. The harder question is whether it can run the machinery. If the computing power needed to train and operate AI systems sits elsewhere, sovereignty becomes a legal phrase with a very practical weakness: someone else controls access.

Spain is trying to answer that weakness with a bid built around Móra la Nova, a small town in Tarragona province. The proposal is less about another Brussels rulebook than about land, substations, cooling systems and the very large electricity bills that come with industrial-scale AI.

A factory for calculations

A normal data centre is built to handle many separate jobs at once: company records, cloud software, streaming video, payments, backups. An AI gigafactory is different. It is designed so tens of thousands of specialised chips can work together on one vast calculation, such as training a language model or an AI system for medical imaging.

That makes the engineering unforgiving. Storage, cooling and networking all have to be tuned so the chips are not left waiting for data. In this business, an idle chip is not a small inefficiency. It is expensive capacity doing nothing (Hammerspace, Schneider Electric).

The EU plans to support up to five such gigafactories across the bloc, each with about 100,000 advanced AI chips (Data Center Dynamics). The bidding process opens this month, with a decision expected by late 2026 (elDiario.es). Spain’s proposal is centred on Móra la Nova, with a second site near Madrid (Europa Press).

Why the Spanish state wants to co-own it

The ownership plan shows how Madrid sees the project. A state investment vehicle would take about 48%, while Telefónica, construction group ACS and Santander would each hold roughly 16%. Multiverse Computing, a quantum-computing company, would take the remaining 4% (La Vanguardia, 20minutos).

That is not the shape of an ordinary private cloud campus. It is the shape of strategic infrastructure. The presence of ACS and Santander is a reminder that the challenge is not only technical. Someone has to finance it, build it, connect it and keep it supplied with power.

The money needs careful reading. Public commitments identified so far amount to about €1 billion, combining Spanish state funding and a contribution to the EU’s joint supercomputing programme (esmartcity). The €5 billion figure widely attached to the project is the hoped-for total investment if Spain wins, not money already committed (elDiario.es).

The real bottleneck is electricity

For all the talk about AI, a gigafactory is an industrial plant. It needs land, water and, above all, electricity. At 100,000 chips, demand moves into the territory of a small city. Móra la Nova’s local substation currently provides 50MW, with plans to expand towards 125MW by 2030 (20minutos).

Ireland knows this problem well. Data centres here grew into such a large part of national electricity demand that regulators now judge new connections against local grid capacity and wider system pressure (CRU). It is a useful warning for Spain. The chips can be bought, but they are useless without transmission upgrades, substation capacity and agreements to cut or shift demand when the grid is under strain.

Who actually gets to use it

The strongest argument for building these facilities comes from Portugal’s parallel bid. Its government says gigafactory capacity would be available to small businesses, startups, researchers and public services (Government of Portugal). That is the public-interest case: a hospital fine-tuning a medical AI model, or a university training a language system, could use European computing power without sending sensitive data through a non-European cloud.

France shows both the opportunity and the limit. At its Choose France summit, it attracted €93 billion in foreign AI-related investment pledges, helped by the promise of nuclear-powered electricity (Le Monde). Mistral AI’s new data centre near Paris runs on 13,800 Nvidia chips, designed in the United States (Le Nouvel Economiste). Europe can decide where AI runs, who gets access and what rules apply to the data. Independence from US-designed chips is still a distant prospect (EU AI Policy Monitor).

Spain has not won the bid. If it does, operations could begin around late 2028 (La Vanguardia). Much of Europe’s AI funding still exists as programmes and financing vehicles rather than machines that researchers can actually book time on (Thorsten Meyer AI). The test for Móra la Nova will be physical before it is political: whether the substation, grid connection and cooling systems are built before the AI race has moved somewhere else. AI sovereignty depends on rules, but it also depends on who gets the concrete poured.