The numbers: €30 billion, 700,000 chips, zero EU hyperscalers
On Thursday, 30 July 2026, the European Commission published its call for tenders for AI Gigafactories — the latest installment in the bloc's drive toward what it calls "technological sovereignty." The €10 billion in direct public funding ($11.4 billion) is meant to be matched and then some: the Commission expects to unlock at least €20 billion ($22.8 billion) in private co-investment, bringing the total envelope to north of €30 billion.
"Access to the raw scale of computing power within AI gigafactories is a strategic necessity for Europe as AI development accelerates," said Henna Virkkunen, the Commission's executive vice president overseeing tech sovereignty, in the announcement.
Let's put those chip numbers in perspective. The EU currently runs a network of 19 AI data centers stretching from Finland to Spain. Each of the seven new gigafactories must pack at least 100,000 state-of-the-art AI accelerators — likely a mix of NVIDIA H200-equivalent and next-gen chips. That is roughly 700,000 chips total. For reference, Meta's two clusters for Llama 4 training each used about 100,000 H100s. The EU is essentially proposing to build seven Llama-4-training-scale facilities, with public money.
| Metric | Number |
|---|---|
| Public funding | €10 billion ($11.4B) |
| Expected private investment | €20+ billion ($22.8+B) |
| Total investment target | €30+ billion |
| Gigafactories planned | Up to 7 |
| Chips per facility (minimum) | 100,000 |
| Power vs current EU data centers | ~4× per facility |
| Bid deadline | 12 November 2026 |
| Cost per gigafactory (avg, public share) | ~€1.43 billion |
Why this matters: Europe's compute dependency problem
The political framing — sovereignty, strategic autonomy — sounds lofty, but the underlying numbers are stark. A June 2026 Commission report presented to the European Parliament laid it out: the EU's top five cloud service providers are all American. "European businesses and public authorities will continue to rely on US AI providers to the detriment of European service providers struggling to work at the frontier," the report stated bluntly.
This is not just about corporate market share. The Commission's concern — articulated with increasing urgency since Trump's return to the White House — is that critical AI infrastructure in foreign hands can be "weaponized" against European interests. Whether through access restrictions, pricing leverage, or data jurisdiction issues, depending on AWS, Azure, and Google Cloud for frontier AI training puts Europe's industrial and research base at the mercy of decisions made in Seattle and Mountain View.
Europe also has structural disadvantages. Electricity in the EU costs two to three times what it does in the US and China, according to the same Commission report. The bloc doesn't manufacture most of the millions of components data centers require. And the US Federal Reserve's 2025 assessment placed Europe firmly behind both the US and China in the sectors critical for AI development.
What €30 billion buys — and what it doesn't
At a rough average of €4.3 billion per gigafactory (public plus private), these are serious facilities. But building the hardware is the easy part. The harder question is who will train on it.
France's Mistral currently runs the EU's largest AI data center at its Paris campus, and its Le Chat chatbot remains the continent's most visible answer to ChatGPT and DeepSeek. But Mistral hasn't kept pace at the frontier. The gap between European AI companies and their American and Chinese counterparts is not primarily a compute gap — it's a talent, capital, and ecosystem gap. More GPUs won't fix that by themselves.
The Commission is at least aware of the talent dimension. The gigafactory program includes provisions for cross-border collaboration and connections to existing supercomputing networks. The tender also mandates that AI products developed on this infrastructure "follow EU standards on data protection, safety, security and ethics" — which means GDPR and the AI Act are baked into the procurement from day one, not layered on after the fact.
Interested consortia or special-purpose vehicles must submit their bids by 12 November 2026, giving bidders just over three months. That is an aggressive timeline for projects of this scale, which suggests the Commission already has candidates in mind — or wants to filter for consortiums that already have sites, power agreements, and chip supply chains lined up.
The European angle: AI Act compliance from the silicon up
Here is what distinguishes these gigafactories from the US hyperscalers: the Commission is requiring that the compute infrastructure and the models trained on it comply with EU regulatory frameworks from the start. That means GDPR-grade data handling, AI Act transparency requirements, and Digital Services Act obligations for any services built on top. For European companies and public authorities that handle sensitive data — healthcare, defense, justice — this is the pitch. Compute that does not come with the legal ambiguity of US cloud jurisdiction.
The practical consequence: if you are a European startup training a medical AI model on patient data, you currently face a dilemma. Use US cloud and navigate complex data transfer mechanisms, or use limited on-premise hardware. With a gigafactory physically located in the EU, operated under EU law, that dilemma shrinks considerably.
The catch — and there is always a catch — is electricity. At 100,000 accelerators per facility, each gigafactory will draw power on the scale of a small city. In June, 40 mayors worldwide signed a pact to limit the impact of AI data center construction on natural resources, energy prices, and climate targets. The EU is simultaneously pushing aggressive climate goals and massive power-hungry infrastructure. That tension is unresolved.
When will these gigafactories actually be operational?
The bid deadline is November 2026. Assuming selection by early 2027 and a construction timeline of 18–24 months, the first gigafactories could come online in 2028–2029. Chip supply chains and power infrastructure permitting, which are not trivial assumptions.
Who is likely to bid?
Expect consortia led by large European telecoms, energy companies, and existing data center operators — possibly including companies like OVHcloud (France), Deutsche Telekom, or Equinix. EU-based AI companies like Mistral are natural partners but unlikely to lead bids given the capital requirements. US hyperscalers may participate through European subsidiaries, though the sovereignty framing makes that politically delicate.
What does this mean for AI prices in Europe?
In the short term, nothing. These gigafactories are years away. In the medium term, increased domestic compute supply should lower inference and training costs for EU-based customers — but only if the facilities achieve high utilization rates and competitive electricity pricing. If power costs remain 2–3× US levels, the per-token economics will still favor American cloud providers.
