Summary
- The EU is seeking industry-led proposals for up to seven large-scale AI computing facilities.
- Public funding of as much as €10 billion is intended to unlock at least €20 billion of private investment.
- Power, cooling, connectivity, security, planning, and delivery readiness will determine which proposed sites can proceed.
The European Commission has opened a tender for up to seven AI gigafactories, beginning a contest to place some of Europe’s largest planned computing facilities in locations capable of supplying the necessary power, land, cooling, and connectivity.
As much as €10 billion of EU and national funding will be made available, with Brussels expecting the public contribution to attract at least €20 billion from industry. Individual projects will be led principally by private-sector consortia and are intended to combine advanced AI processors, storage, networking, software, and the data centre infrastructure required to operate them at scale.
The new facilities will sit above the EU’s existing network of 19 AI Factories, which provide computing access and support to businesses, researchers, and public bodies. Gigafactories are intended for the development and operation of substantially larger models, placing them in a different class of electrical and mechanical demand from most existing European research-computing sites.
Site selection moves from policy to engineering
The procurement process will test whether proposed locations have resolved the physical constraints behind their headline capacity. Large AI clusters need firm grid connections, reinforced substations, diverse fibre routes, substantial cooling plant, and enough land to accommodate phased expansion without compromising security or maintainability.
Electrical demand is likely to arrive in large blocks rather than through the gradual occupancy associated with conventional colocation. Training clusters can bring thousands of accelerators online within a relatively short commissioning window, and their synchronised workloads can produce fast changes in power demand. The utility connection, internal distribution system, UPS architecture, and generation controls must all be designed around those operating characteristics.
Cooling requirements will also shape the competition. Direct-to-chip liquid cooling is becoming the default assumption for many high-density deployments, but its adoption extends the engineering brief beyond the server rack. Facilities need cooling distribution units, redundant pumping, water treatment, leak detection, secondary loops, and heat-rejection plant capable of operating through summer design conditions.
Water availability will influence the choice of cooling architecture and the planning case. Some locations may favour closed-loop dry or hybrid systems to limit consumption, accepting higher electricity use during hot weather, while others may pursue evaporative designs or industrial-water supplies. Any heat-reuse proposal will need an identified customer, a viable temperature profile, and infrastructure to transport heat beyond the campus boundary.
Connection dates are likely to carry as much weight as nominal grid capacity. A site with a large allocation that depends on reinforcement late in the next decade may be less useful than a smaller location with an established route to staged energisation. Developers will also need to show that transformers, switchgear, generators, cooling equipment, and specialist construction capacity can be secured within the programme.
Public capital brings delivery conditions
With industry expected to provide at least twice the public contribution, each consortium will need credible customer demand and a financing structure that can absorb long construction and hardware-procurement cycles. Public funding may reduce the initial capital burden, but it will not protect an underused facility from rapid changes in processor technology or customer requirements.
Flexibility will have to be designed into the buildings. Electrical rooms, busway, cooling distribution, slab loading, and network spaces must accommodate hardware generations that may not be finalised when construction begins. Building every hall to the most extreme forecast density would lock up capital, while a conventional design could require disruptive reconstruction before the first expansion phase is occupied.
Security and sovereignty requirements will add another layer to the engineering programme. The Commission expects projects to comply with EU standards covering data protection, safety, security, and ethics, which will translate into controls around physical access, network separation, firmware, supplier assurance, and workload governance.
The tender also exposes the differences between European power markets. Regions with available industrial land, shorter connection queues, established renewable generation, and planning systems accustomed to large infrastructure will enter the process with an advantage. Markets where developers are already waiting years for grid capacity may struggle to reconcile the programme’s delivery ambitions with the physical pace of network construction.
Local authorities will examine more than the investment value. Campuses of this scale can require new transmission assets, extensive generator compounds, cooling plant, water infrastructure, and prolonged construction activity, while permanent employment is often modest relative to the land and energy consumed. Successful proposals will need a clear account of regional supply-chain work, network investment, heat or water arrangements, and the benefits retained in the host area.
The tender creates a funded route towards a limited number of strategic European computing sites. It does not shorten transformer lead times, create grid capacity, or resolve a planning application by itself. The projects that progress will be those whose sponsors have already aligned capital, customers, power, engineering, and public approvals into a deliverable campus programme.
