The question is not whether the investment was needed. The question is whether it is moving fast enough to matter.

What Are EU AI Gigafactories Europe Is Building

EU AI gigafactories are not ordinary data centres. They are national supercomputing clusters being upgraded to train and run frontier AI models: the kind of large-scale infrastructure that has, until now, been the exclusive domain of US hyperscalers and Chinese state-backed compute programmes.

In January 2026, the EU Council amended the regulation governing the European High-Performance Computing Joint Undertaking (EuroHPC JU), extending its mandate to include the development and operation of AI gigafactories across the continent. This was the second amendment to the EuroHPC framework: the first, in 2024, introduced AI Factories as dynamic compute clusters for AI experimentation. The 2026 amendment goes further, enabling facilities equipped with roughly 100,000 advanced processors each, approximately four times the capacity of the current generation of data centres.

Each gigafactory sits within the AI Continent Action Plan, the Commission's broader roadmap for European AI leadership through infrastructure investment, data access, skills, and regulation. The InvestAI initiative, which underpins the procurement process, received 76 proposals from 16 EU member states in a June 2025 call for interest, far exceeding initial expectations.

The Numbers That Define the Race

Europe's case for public AI compute investment starts with a stark gap. According to the Stanford AI Index, the United States produced more than 50% of all significant AI models in 2024; Europe produced just 6%. American and Chinese companies control more than 90% of the world's AI-specialised data centres, according to Oxford research cited by the Centre for European Policy Studies (CEPS). On private venture capital, the imbalance is equally sharp: between 2023 and mid-2025, 66% of global VC investment in AI start-ups went to the United States, compared with 12% to Europe, according to the CEPS AI World Index.

Building its way into relevance is the EU's response. Commissioner Henna Virkkunen has stated that majority owners of the gigafactories "should come from Europe," a political signal that this is a sovereignty project, not merely an infrastructure spend. Seven member states will co-finance the programme: France, Germany, Italy, Spain, the Netherlands, Finland, and Portugal.

The amended EuroHPC regulation explicitly safeguards the interests of start-ups and scale-ups, making compute access a structured part of the mandate rather than an afterthought. We have previously covered European AI infrastructure startups, including Verda, which submitted a proposal to build one of Europe's first gigafactories with up to 100,000 accelerators: a direct signal of private-sector ambition aligning with the public investment framework.

Seven Countries, One Strategic Bet

Participating member states represent a coalition of Europe's largest economies alongside some of its most strategically positioned compute locations. The programme's geography is shaped by practical constraints: electricity costs, existing grid infrastructure, cooling capacity, and proximity to AI talent clusters. France, Germany, and the Netherlands bring established digital infrastructure hubs; Finland and Sweden offer some of the most competitive electricity prices on the continent, running closer to US levels than the European average.

That price gap matters. European industry pays roughly double the electricity rate of US counterparts, according to ACER data published in 2026. Electricity accounts for a single-digit percentage of total model training costs, but with training runs costing hundreds of millions of euros, every percentage point compounds.

Private capital is reading the same map. Nebius, the Netherlands-based AI cloud provider, announced plans to build a 310-megawatt AI facility in Finland and a 240-megawatt site near Lille, France, both at the same order of magnitude as the EU's planned gigafactories. Goldman Sachs Research forecasts that global electricity demand from data centres will rise 50% by 2027 and 165% by end of decade, compared with 2023. Europe's high-performance computing infrastructure market is entering a period of capital intensity that will define the competitive landscape for a decade.

Training Models vs Controlling the Stack

This is where the sovereignty argument gets complicated. CEPS analysis published in 2025 identified a structural dependency at the heart of the EU's gigafactory ambitions: all planned facilities will run on Nvidia chips. Training AI models on Nvidia GPUs requires the CUDA programming model, owned by Nvidia, which controls which chips are compatible and limits alternatives. Building gigafactories at European scale while remaining dependent on a single US semiconductor company's software stack is, as CEPS put it, the sovereignty blind spot in the programme.

Chip architecture is the deepest layer of AI sovereignty, not compute capacity alone. European photonics and semiconductor alternatives are beginning to emerge: from graphene-based transceivers (CamGraPhIC) to photonic processors already deployed at Germany's LRZ Garching supercomputing centre (Q.ANT). We covered the leading edge of this cohort in our earlier piece on European photonics and semiconductor alternatives.

Mistral AI, Europe's most prominent frontier model company, announced plans to build 200 megawatts of its own data centre capacity in Europe by end of 2027, funded independently from the EU programme. Mistral's decision to move outside the gigafactory framework is evidence that European AI champions are fast enough to outrun public procurement timelines, not a failure of the programme itself. Both tracks are necessary: the EU programme provides the foundation; companies like Mistral provide the proof that European compute can produce world-competitive AI outputs.

---

What the Race Means for Europe's Builders

Safeguarding the interests of start-ups and scale-ups is written into the EuroHPC JU amendment explicitly. Compute access for European AI builders is no longer dependent on the goodwill of US hyperscalers or the budget of a well-funded research institute. The gigafactory programme is designed to make frontier compute a shared European resource.

Three concrete implications follow for founders and investors. Structured compute access for training and inference will become available at a scale and price point previously inaccessible to European AI startups. The public-private partnership model embedded in the EuroHPC JU framework creates procurement opportunities for companies building across the stack: networking, cooling, security, and the chip layer itself.

Most significantly, the programme creates a sovereignty mandate that will consistently favour European-built alternatives at every layer where they can credibly compete. The €176 billion in cumulative data centre investment expected across Europe between 2026 and 2031 is not a one-cycle trend. For venture capital, this is the structural shift that changes the risk profile of European AI infrastructure investment.

Inside the Compute Debate at DTM26

Leaders building and stress-testing European compute sovereignty will be on stage at Deep Tech Momentum 2026. Among them: Daniel Schall, Co-Founder and CEO of Black Semiconductor, who takes the Titans Stage on Day 2 at Wilhelm Studios Berlin. Black Semiconductor works at the intersection of photonics and next-generation semiconductor architectures, directly relevant to the CUDA dependency question at the centre of the EU gigafactory sovereignty debate.

Explore the HPC and AI Gigafactories programme at DTM26 to see the full line-up of builders, investors, and corporate leaders turning this policy framework into operational reality. Deep Tech Momentum 2026 runs 20 and 21 May 2026 at Wilhelm Studios Berlin, with more than 3,000 attendees from Europe's leading VC firms, deep tech founders, and corporate innovation teams.

The gigafactory race will be won or lost not in Brussels, but in the decisions made by the people building and funding European AI infrastructure in the next 18 months.

Secure your place at DTM26 to be part of it.