The European Union threw open the bidding on Thursday for up to seven AI gigafactories, a roughly €30 billion effort to expand the bloc's computing muscle and trim its reliance on foreign cloud and AI infrastructure. The call for tenders, run by the European High Performance Computing Joint Undertaking, pairs up to €10 billion in public money from the EU and member states with at least €20 billion in expected private investment, the European Commission said.
Submissions close on November 12. Award decisions are expected by early 2027, with construction beginning the same year. Each gigafactory will combine advanced AI processors, software, cloud technology stacks, high-speed connectivity and energy-efficient data centers, a package designed to give companies, startups, universities and public authorities access to computing power for training, inference and fine-tuning frontier AI models.
"Access to the raw scale of computing power within AI gigafactories is a strategic necessity for Europe as AI development accelerates," Commission Executive Vice President Henna Virkkunen said in a statement.
Two lots, two sizes
The tender is split into two lots with different ceilings. The first supports up to four projects, each eligible for up to €100 million in EU funding in an initial phase and up to €400 million more in a second phase. The second supports up to three projects, each eligible for up to €200 million initially and up to €800 million more in phase two. Eighteen member states have signed a joint procurement agreement and will match EU funding.
The scale requirements are stiff. Every gigafactory must deploy at least as many top-end AI processors as are currently installed in Europe's most powerful AI facility. In the second phase, capacity in the first lot must reach at least three times that level; in the second lot, at least four times. Those multiples are effectively a demand that the winning consortia build for expansion from day one, with the halls, power and cooling loops for several generations of growth.
Hardware can be procured from providers in Europe or from what the Commission describes as likeminded countries. Brussels has already signed letters of intent with AMD, Nvidia and Qualcomm to make sure bidding consortia can actually get the chips, a hedge against supply bottlenecks at a time when every major cloud builder on the planet is chasing the same silicon. The letters also carry a political edge: they keep the door open to European and allied suppliers while leaving Chinese vendors out of the picture, a line that matches the bloc's broader push to screen sensitive technology purchases.
Who actually bids remains an open question. The natural candidates are the big European telecom and hosting groups, the national supercomputing centers that already run the 19 AI factories, and consortia of smaller firms that could not afford a gigafactory on their own. Private equity and infrastructure funds have circled European data centers for years, and a project with a state-backed revenue floor looks attractive to that crowd. The Commission has not named any shortlist; the November 12 deadline means the bidding consortia are still forming, and the member states that signed the joint procurement agreement will influence which projects get the strongest backing.
Building on the AI factory network
The gigafactories will sit alongside Europe's existing network of 19 smaller AI factories attached to supercomputers. When the new sites come online, the Commission says, they and the existing network would push the bloc's total AI computing capacity to more than twice its current level.
That is a meaningful step for a continent that has watched most frontier-model training happen on the other side of the Atlantic. The EU's approach is explicitly public: state money, shared infrastructure, and access for startups and researchers who cannot rent clusters from American hyperscalers at hyperscale prices. The program is also an answer to a running worry in Brussels that European developers depend on US clouds for the raw compute their models need, a dependence that looks more fragile every time export rules or geopolitical tensions shift.
The 19 AI factories already in place give the gigafactory program a running start. Those facilities, built around EuroHPC supercomputers, were the bloc's first attempt to hand researchers and small companies a slice of serious compute. The gigafactories are a bigger, bolder version of the same idea: instead of bolting AI capacity onto scientific machines, they are purpose-built for model training from the ground up. The existing factories also provide the operational know-how, the procurement teams and the site-selection experience that a new build-out will need.
The contrast with US hyperscalers
The timing matters. US clouds are spending at a pace that dwarfs the EU program. AWS raised its 2026 capital spending forecast to $220 billion, Microsoft's Azure just passed $100 billion in annual cloud revenue, and Google signed a $30 billion deal with SpaceX for AI compute capacity. Measured against those numbers, the gigafactory program's €30 billion total is roughly a week of AWS capital expenditure, which raises the obvious question of whether public money alone can close the gap.
On the defense side, the US is taking a more private route to sovereign compute. CoreWeave and Leidos announced a collaboration on July 30 to deliver secure, classified AI cloud services for US intelligence and defense missions, built inside SCIF-accredited data centers. The deal pairs an AI-native cloud provider with a federal systems integrator, and it shows how the American model leans on contractors and equity markets rather than state procurement.
Europe's bet is different: fewer euros per site, but public ownership of the infrastructure and a requirement that capacity serve the whole bloc, not just the biggest companies. The letters of intent with AMD, Nvidia and Qualcomm suggest the EU wants to avoid being locked into any single chip vendor even as it leans on non-European suppliers for now. The joint procurement agreement among 18 member states is itself a political signal, a rare instance of capitals pooling money for shared computing infrastructure.
What this means for edge computing
For cloud and edge watchers, the gigafactories are the centralized half of a two-sided story. Training frontier models needs big, concentrated clusters with dense interconnect and heavy cooling, the kind of facility the tender describes. But the same models, once trained, get served from locations much closer to users: city data centers, telecom exchanges, factory floors. The EU's 19 existing AI factories already reach into member states; the gigafactory program is meant to anchor the training tier, while edge sites handle the latency-sensitive inference work.
The split matters for how European networks get built. A gigafactory in one region does little for a hospital in another that needs sub-50-millisecond inference for diagnostics, and that is exactly the gap the edge tier fills. Brussels has been pushing edge infrastructure in parallel, funding cross-border cloud projects and working on standards for federated AI where models move to the data instead of the other way around. The gigafactories give those edge deployments a back-end to pull from, which is why the Commission describes the two layers as one system rather than competing bets.
There is also an energy angle. The Commission explicitly requires energy-efficient data centers, and European operators face some of the tightest grid constraints in the world, with several member states having paused or reworked data center approvals over power and water concerns. Designing for efficiency is not a marketing choice here; it is a precondition for getting a permit. The phase-two capacity multiples will only sharpen that pressure, since every doubling of AI compute typically arrives with a matching jump in power draw.
The energy question looms over every bid
Power is the constraint that could quietly reshape the whole tender. European data center developers already wait years for grid connections in places like Amsterdam, Frankfurt and Dublin, where utilities have capped new connections because substations are at capacity. A gigafactory running thousands of accelerators draws more power than a small town, which means site selection will hinge as much on substation capacity and renewable generation as on land prices.
Some member states are positioning for exactly this. The Nordics, with hydro and wind surpluses, and southern countries with solar potential have both signaled interest in hosting AI infrastructure. The joint procurement agreement among 18 member states suggests the political horse-trading over sites is already underway. The Commission's efficiency requirement pushes bidders toward liquid cooling and heat reuse, where the waste heat from servers warms district heating networks instead of escaping into the sky, a design that several European data centers already use.
None of this happens fast. Bids close November 12, awards land early next year, and construction starts in 2027, so the first gigafactories are realistically two or more years from running models. For now, the tender is the clearest signal yet that Brussels treats compute as strategic infrastructure, in the same category as roads and power grids. Whether the money is enough is a separate argument, and one the market will keep having as the program moves through procurement.
The full terms are in the Commission's call documentation, and the tender notice itself carries the fine print on lot sizes and phase-two multiples. Yahoo Finance's summary has the two-lot breakdown, and Politico's coverage frames the political stakes. Leidos and CoreWeave's announcement shows the US taking a different path to sovereign AI compute. For more on this site's coverage of infrastructure and AI, see the Cloud & Edge Computing section and the AI category.