Europe’s most ambitious AI infrastructure program formally opened its bidding process on Thursday \u2014 and the fine print tells a different story than the headline.<\/p>\n
The European Commission launched the official call for tenders to build up to seven AI Gigafactories across the EU on July 30, with a headline investment figure of more than \u20ac30 billion (approximately $34.6 billion USD). Behind that number: only approximately \u20ac1 billion (approximately $1.2 billion USD) of Brussels’ public share is actually locked in. The rest depends on a long-term EU budget that has not been agreed, as The Next Web reported<\/a>. And all the chips that would fill these facilities \u2014 the entire compute layer of Europe’s “sovereignty” play \u2014 will come from Nvidia, AMD, and Qualcomm, three American companies with which the Commission has signed letters of intent, Euronews confirmed<\/a>.<\/p>\n A senior Commission official acknowledged the gap directly on Thursday. “We cannot pre-empt the decisions about the next MFF,” the official told reporters, describing the \u20ac30 billion figure as a “best estimate” of what might be available rather than money in hand, The Next Web reported<\/a>.<\/p>\n The initiative calls for up to seven AI Gigafactories \u2014 large-scale compute facilities each equipped with at least 100,000 cutting-edge AI chips \u2014 spread across EU member states, according to the European Commission announcement<\/a>. At that chip count, a single gigafactory is roughly four times more powerful than the largest AI data centers currently operating in the EU. Together, seven of them would more than double the bloc’s total AI compute capacity.<\/p>\n The math on power consumption alone illustrates why the energy question is as consequential as the funding question. A 100,000-chip facility drawing roughly 700 watts per GPU-class accelerator would require approximately 70 megawatts of raw compute power \u2014 and 120 to 150 megawatts of total facility load once cooling, networking, and support systems are included, according to a CEPS analysis<\/a>. European industrial electricity costs are roughly two to three times those in the United States or China, meaning a facility that might cost \u20ac100 million to \u20ac200 million per year to power in Europe could be operated for half or a third of that cost elsewhere, Euronews energy analysis<\/a> found.<\/p>\n The tender is split into two funding lots, as the IBTimes UK report<\/a> detailed. Projects in the smaller lot are eligible for up to \u20ac100 million (approximately $115 million USD) in early-stage EU funding, rising to \u20ac400 million (approximately $461 million USD) in a second phase. The larger lot offers up to \u20ac200 million (approximately $231 million USD) initially and as much as \u20ac800 million (approximately $922 million USD) subsequently. Eighteen EU member states \u2014 Croatia, Czechia, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Ireland, Italy, Latvia, Lithuania, Poland, Portugal, Slovakia, Spain, and Sweden \u2014 have signed a joint procurement agreement with the EuroHPC JU to participate, according to the Commission<\/a>.<\/p>\n Consortia or special purpose vehicles comprising technology providers, cloud companies, public entities, and investors can apply. Projects may be hosted in a single member state \u2014 at one or multiple sites \u2014 or span borders through distributed infrastructure, IBTimes UK reported<\/a>.<\/p>\n What “\u20ac30 Billion” Actually Means<\/p>\n The headline figure requires unpacking. Of the roughly \u20ac10 billion (approximately $11.5 billion USD) in public funding that the Commission describes, Brussels itself is targeting approximately \u20ac5 billion (approximately $5.8 billion USD), matched by another \u20ac5 billion from the member states hosting facilities. The remaining \u20ac20 billion (approximately $23.1 billion USD) is expected from private investors, Euronews reported<\/a>.<\/p>\n Under the current EU budget, however, the Commission can commit only roughly \u20ac1 billion (approximately $1.2 billion USD) \u2014 the amount available from existing programs including Horizon Europe, Digital Europe, and the Connecting Europe Facility. Council Regulation 2026\/150, which entered into force in January 2026 and formalized the EU’s contribution to the gigafactories through the EuroHPC Joint Undertaking, caps the Union’s share at 17 percent of the capital expenditure for each facility’s computing infrastructure, the Open Future analysis<\/a> documented. The remaining approximately \u20ac4 billion (approximately $4.6 billion USD) of Brussels’ stated target depends on the next Multiannual Financial Framework \u2014 the EU’s long-term budget, negotiations for which are currently ongoing and unresolved, Euronews reported<\/a>.<\/p>\n That 17 percent cap is the structural fact that the headline obscures: European law limits the public subsidy to a minority share of each project’s cost. The other 83 percent must come from private investors and member states. Whether private capital commits at the necessary scale \u2014 against a backdrop of energy cost disadvantage, delayed timelines, and competing global alternatives \u2014 is the central unanswered question of the program.<\/p>\n Sovereignty Built on American Chips<\/p>\n At the heart of the gigafactory program is a tension the Commission has addressed rather than resolved. The initiative is framed as reducing Europe’s dependence on foreign AI infrastructure \u2014 but the infrastructure itself will run on chips from Nvidia, AMD, and Qualcomm, all headquartered in the United States, Euronews confirmed<\/a>.<\/p>\n A group of 18 European Parliament lawmakers flagged this directly in a warning to the Commission before the tender launch, questioning how the gigafactory initiative would reduce Europe’s strategic dependencies when the compute layer remained entirely in American hands, Techzine analysis<\/a> reported.<\/p>\n The Commission has signed three letters of intent with Nvidia, AMD, and Qualcomm \u2014 a move framed as ensuring “seamless access to the necessary hardware” following the EU-US trade deal, Euronews reported<\/a>. Henna Virkkunen, the Commission’s Executive Vice-President for Technological Sovereignty, Security and Democracy, described access to gigafactory-scale compute as “a strategic necessity for Europe as AI development accelerates,” the Commission stated<\/a>. The specification that “part of the procurement may be dedicated to European start-ups and scale-ups in order to strengthen the European digital supply chain” signals awareness of the dependency \u2014 but not a structural solution to it, as IBTimes UK noted<\/a>.<\/p>\n The chip architecture dependency is not merely a procurement concern. Nvidia’s NVLink\/NVSwitch interconnect architecture and AMD’s Infinity Fabric are proprietary \u2014 facilities built around one vendor’s interconnect cannot easily absorb workloads from the other. Individual gigafactories that choose different chip vendors face real workload-portability constraints, limiting how interoperable the seven facilities will actually be, the CEPS analysis<\/a> warned.<\/p>\n Does Europe Actually Have Room for Seven Training Supercenters?<\/p>\n A subtler technical critique has emerged from European policy research: gigafactories are designed for AI training workloads, but the center of gravity in enterprise AI spending is already shifting toward inference.<\/p>\n Training workloads \u2014 the computationally intensive process of building an AI model from scratch \u2014 are centralized, latency-tolerant, and power-hungry. They are well-suited to large, remote facilities. Inference workloads \u2014 the real-time serving of an AI model to users \u2014 require proximity to end-users to meet latency requirements typically under 100 milliseconds, per CEPS analysis<\/a>. As Microsoft, Google Cloud, AWS, and others move toward disaggregated AI infrastructure with dedicated inference chips and distributed deployment, the training-cluster model that gigafactories represent may be optimizing for a workload mix that enterprise AI is moving away from, according to Techzine’s analysis<\/a> of the plan.<\/p>\n A study led by Maria Nowicka at Interface, a European energy and digital policy think tank, warned that without careful siting and grid integration, the facilities risk becoming costly stranded assets: “Constructing multi-hundred-megawatt facilities that fail to use their contracted capacity effectively would be unsustainable not only economically but also from an energy- and climate-system perspective,” Euronews energy analysis<\/a> reported.<\/p>\n Nowicka, speaking specifically about the gigafactory tender as it stands, added that the specifications remain intentionally open in several areas: “The tender is quite packed, and while it contains references to energy efficiency, networking, and supply-chain resilience, some of those requirements are intentionally left open and vague. That leaves a real risk that individual AI gigafactories end up pursuing their own purposes, rather than fitting into a single coherent EU-level strategy.”<\/p>\n A History of Delays: Narrowing Bids and a Slipping Schedule<\/p>\n The formal tender launch on Thursday arrives more than a year after the initiative was first announced. Commission President Ursula von der Leyen launched InvestAI at the AI Action Summit in Paris in February 2025, announcing plans to mobilize \u20ac200 billion (approximately $230.5 billion USD) for AI investment, including a \u20ac20 billion (approximately $23.1 billion USD) fund specifically for AI Gigafactories, per the Commission InvestAI announcement<\/a>. The original call for tenders was targeted for late 2025. It slipped to early 2026, then again to summer 2026, The Next Web delay report<\/a> documented.<\/p>\n The delays have taken a toll on private-sector appetite. An earlier expression-of-interest round drew 76 responses from consortia in 16 member states across 60 sites, EuroHPC JU data<\/a> confirmed. But Bloomberg reported \u2014 and Droids Substack citing Bloomberg<\/a> subsequently documented \u2014 that the field of expected actual bidders had narrowed to approximately 10, with at least two consortia reconsidering participation as the project was scaled back. Among confirmed bidders: Deutsche Telekom and Brookfield, and France’s Scaleway-led AION consortium.<\/p>\n The CEPA think tank framed the pattern in broader historical terms in February 2026: AI Factories “fail to address Europe’s main AI weaknesses: high energy costs, stringent copyright rules, and tough AI regulation. They increase potential supply but fail to stimulate what’s needed most: demand.” The organization drew comparisons to prior large-scale European industrial initiatives \u2014 Galileo, the satellite navigation system, and Gaia-X, the European cloud initiative \u2014 that came in late, over cost, or underperforming against their stated objectives, the CEPA analysis<\/a> found.<\/p>\n Europe’s most prominent AI developer, Mistral, did not wait for the gigafactories: the French startup invested \u20ac1.2 billion (approximately $1.4 billion USD) in a Sweden data center in February 2026 and raised hundreds of millions of euros for a facility near Paris, raising the question of who the gigafactories are actually designed to serve, Techzine analysis<\/a> noted.<\/p>\n What the Commission Says This Is For<\/p>\n Commission spokesperson Thomas Regnier pushed back on demand-side concerns, arguing that the initiative is not primarily about raw compute volume: “This is not just about raw compute power, it’s about sovereign compute” \u2014 environments where data is “fully protected under European law, without the possibility of third country interference,” Techzine analysis<\/a> reported.<\/p>\n The argument has merit. European enterprises building AI systems on Amazon Web Services, Microsoft Azure, or Google Cloud face structural exposure under the U.S. CLOUD Act, which allows U.S. law enforcement to compel disclosure of data held by U.S. companies regardless of where their servers sit. A June 2026 incident in which Anthropic was ordered to suspend its Fable 5 and Mythos 5 models under U.S. export controls \u2014 disabling access for users globally \u2014 made the dependency concrete in a way years of policy debate had not, as TechTimes previously reported<\/a>.<\/p>\n The Commission’s answer to the chip-dependency critique is proportionality: getting the facilities built now, on available hardware, is preferable to waiting for European chip production that does not yet exist. Criteria in the tender call specifically address “measures to avoid potential lock-in effects from suppliers,” and the letters of intent with AMD, Nvidia, and Qualcomm explicitly allow procurement from “hardware providers in Europe or like-minded countries,” with a carve-out encouraging hardware from European startups and scale-ups, Euronews reported<\/a>.<\/p>\n Is This How to Answer That Question?<\/p>\n Whether the gigafactory initiative succeeds depends on factors the Commission controls only partially. The deadline for submitting bids is November 12, 2026. Following evaluation by the EuroHPC JU, award decisions are expected in early 2027. Construction is then set to begin in 2027, with the first facilities expected to be operational within 18 months of contract signature \u2014 pointing to a mid-2028 earliest operating date for the first gigafactories, per the European Commission announcement<\/a>.<\/p>\n That timeline depends on: a Multiannual Financial Framework that has not been agreed; private investors willing to commit the 83% of project capital that the EU Regulation does not cover; energy grid capacity that European network operators are already warning is under strain; and a chip supply chain that remains entirely in American hands. The Commission’s position is that building imperfect infrastructure is preferable to building nothing \u2014 and that compute is a precondition for everything else Europe wants to accomplish in AI.<\/p>\n