The short answer
On 30 July 2026 the European Commission opened a call for expression of interest for up to seven AI gigafactories — very large data-centre complexes meant to train and run next-generation AI inside the bloc. It pledged €10 billion in public funding and aims to pull in at least €20 billion in private investment, for a total near €30 billion. Each facility is expected to hold 100,000-plus cutting-edge AI chips and be roughly four times more powerful than today's EU data centres.
The plan grew from five sites to seven after strong demand: 76 consortia expressed preliminary interest and ten countries offered to host. But nothing goes live soon — construction starts in early 2027 and the first sites target mid-2028. The practical takeaway for engineering teams is not to wait for that capacity, but to build so you can adopt it when it lands.
What the Commission actually announced
The Commission framed the gigafactories as the flagship of its InvestAI initiative and a core piece of Europe's drive for technological sovereignty. The idea is blunt: Europe trains and serves too much of its AI on infrastructure owned and operated outside the bloc, and it wants home-grown capacity at a scale that can compete with the largest US and Chinese build-outs. The call for expression of interest, opened on 30 July, invites consortia of chipmakers, data-centre operators, cloud providers and investors to bid to build and run the sites.
The numbers are deliberately large. Each gigafactory is expected to combine advanced AI processors, software, cloud technology, high-speed connectivity and data centres, house at least 100,000 cutting-edge AI chips, and be around four times more powerful than the data centres the EU runs today. EU tech chief Henna Virkkunen put the rationale plainly, calling access to raw computing power "a strategic necessity for Europe as AI development accelerates." That is the same logic driving demand for scarce GPUs everywhere, from a hyperscaler region down to a single AI, ML & data team trying to reserve capacity for a training run.
How the €10 billion is meant to work
The €10 billion of public funding is a lever, not the whole cost. The Commission said it wants that money to attract at least €20 billion in private investment, putting the total project value near €30 billion — roughly one-third public and two-thirds private. The public share itself is a blend of EU-level money and member-state contributions, with further EU funding expected to depend on the bloc's next multi-year budget, which is still being negotiated. In other words, the headline figure is a commitment of intent backed by a real but partly future funding stack.
Demand has already outrun the original plan. The number of facilities rose from five to seven after strong interest from member states, and the Commission reported that 76 consortia expressed preliminary interest in submitting a project. Ten countries put their hands up to host: Germany, Italy, France, Poland, Czechia, Denmark, Finland, Greece, Portugal and Spain. That spread matters for data-residency planning — it hints at where EU-region AI capacity is most likely to concentrate over the next few years.
| Detail | What the Commission said |
|---|---|
| Public funding | €10 billion (about $11.4 billion), EU plus member states |
| Target private investment | At least €20 billion, for ~€30 billion total |
| Number of sites | Up to seven (raised from five) |
| Compute per site | 100,000-plus AI chips; ~4× current EU data centres |
| Timeline | Call opened 30 Jul 2026; build from early 2027; first sites ~mid-2028 |
Why Europe is building its own compute
The gigafactories are the physical counterpart to the regulation Europe has already been writing. Over the past year the bloc has moved to set sovereignty criteria for how the public sector buys cloud and AI — the direction we covered when the EU Cloud and AI Development Act was proposed. That act writes the rules; the gigafactories are meant to build the supply those rules point to. It is hard to demand EU-hosted, EU-operated compute for sensitive workloads if the capacity to provide it does not exist at scale. This is the Commission trying to close that gap.
The competitive framing is explicit: the EU is chasing the US and China, where private hyperscalers and state-backed programmes have poured far more into AI data centres. For regulated buyers in FinTech, healthcare and the public sector — the ones who care most about where data is processed and under whose law — a credible EU compute option changes the calculus. It is one thing to prefer a European region on principle; it is another to have gigawatt-scale capacity actually available to run a large model on.
Why the timeline matters more than the headline
The easy mistake is to read "€30 billion" and "seven gigafactories" as capacity you can plan on. You cannot — not yet. Construction is not expected to start until early 2027, and the first sites are targeted to come online around mid-2028. Large data centres also slip; power connections, chip supply and permitting are all constraints that have pushed similar projects. Treat 2028 as the optimistic case, not a delivery date.
That gap is the whole story for engineering teams. Every AI training run and inference endpoint you stand up between now and then will run on today's supply — the same constrained pool of hyperscaler GPUs everyone else is competing for. The gigafactories do not relieve that pressure this year. What they do is make a bet about the destination: by the late 2020s, meaningful EU-based AI compute should exist. The teams who benefit are the ones whose architecture can point at it when it arrives, without a rewrite.
What it means for US & EU software teams
Strip away the geopolitics and this is a forcing function on one design decision: how tightly is your AI stack bound to a single region or provider? For a lot of teams, the honest answer today is "very." Models are fine-tuned in one cloud, weights and vector stores live in one region, and inference is wired to one vendor's endpoints. That is fine until a residency requirement, a price shock, or new EU capacity makes you want to move — and then it is a project.
The teams that will get value from the gigafactory build-out are the ones treating compute location as a variable, not a constant. Concretely, that means keeping model training and inference behind abstractions you can repoint, keeping weights and training data portable, and knowing what a move to an EU region would actually cost you. Standing up or shifting AI workloads across clouds and regions is exactly the kind of work our cloud & DevOps team scopes for US/EU products, and it is far cheaper to build that optionality in now than to retrofit it under deadline.
There is an upside worth naming for European-facing products. If you build the residency and portability story once — where data lives, where models run, and how quickly you could relocate — you turn a future EU compute option into a selling point rather than a scramble. Early movers get to say "yes, we can run this in the EU" before a buyer even finishes asking.
A portability checklist you can run now
Here is the shippable version. Run it as a scoping exercise while the gigafactories are still a call for interest, not a live region.
- Inventory where AI runs. For each model and pipeline, record where it trains, where it serves, where weights and embeddings live, and which provider APIs it calls. Most portability questions fall out of this map.
- Abstract the inference layer. Route model calls through an interface you control, not vendor SDKs scattered through the code, so swapping a provider or region is a config change, not a refactor.
- Keep weights and data movable. Store fine-tuned weights, datasets and vector indexes in formats and locations you can export without a vendor-specific unlock.
- Map data residency now. Know which datasets must stay in the EU and which can roam; this determines which workloads a future EU gigafactory could actually host.
- Model the cost of a region move. Estimate egress, re-training, and re-validation for shifting one real workload to an EU region — the number tells you how locked-in you are.
- Watch the host-country shortlist. The ten candidate countries hint at where EU capacity will land; factor that into latency and residency planning.
- Assign an owner. Name someone to track the gigafactory timeline and revisit the plan as sites are awarded and dates firm up.
None of this is investment or legal advice, and the final shape of the gigafactory programme will depend on which consortia win, how the next EU budget lands, and whether the timelines hold. But the strategic signal is already clear: Europe is committing serious money to home-grown AI compute, and the part that touches ordinary software teams — keeping AI workloads portable across regions — is buildable today. Build it.
Frequently asked questions
What are the EU AI gigafactories?
AI gigafactories are very large, publicly backed data-centre complexes the EU wants built across the bloc to train and run next-generation AI. On 30 July 2026 the Commission opened a call for up to seven of them, each expected to house at least 100,000 cutting-edge AI chips and be roughly four times more powerful than today's EU data centres. They are the flagship of the InvestAI initiative and Europe's push for technological sovereignty.
How much funding is behind the AI gigafactories?
The Commission pledged €10 billion (about $11.4 billion) in public funding and aims to attract at least €20 billion in private investment, for a total near €30 billion — roughly a third public, two-thirds private. The number of planned sites rose from five to seven after strong demand: 76 consortia expressed preliminary interest and ten countries offered to host.
When will the AI gigafactories be operational?
Not soon. The call opened on 30 July 2026, construction is expected to begin in early 2027, and the first facilities are targeted for around mid-2028. EU sovereign compute at scale is a 2028-and-beyond reality, so cloud-region and residency decisions you make in 2026 and 2027 still run on today's hyperscaler capacity.
How does this affect US and EU software teams?
Indirectly but strategically. More EU-based AI compute should ease GPU scarcity and give teams a credible EU-region option, strengthening data-residency stories for European buyers. But it is years out, so the near-term move is architectural: keep AI workloads portable across providers and regions so you can shift to EU capacity when it lands without a rewrite.
How is this different from the EU Cloud and AI Development Act?
They are two halves of one strategy. The Cloud and AI Development Act is regulation and procurement policy that sets sovereignty criteria for how the public sector buys cloud and AI. The gigafactories are the physical supply side — the chips, data centres and power meant to give Europe home-grown capacity to meet those criteria.
Should teams wait for EU gigafactories before choosing a cloud strategy?
No. The facilities will not be operational until roughly 2028, so waiting means stalling. Design for optionality now: run AI training and inference on abstractions you can repoint, keep weights and data movable, and document your residency posture. When EU capacity arrives, teams that built portability in can adopt it as a switch rather than a migration.
Sources
European Commission — AI Gigafactories (Competitiveness)
Euronews — EU opens call for seven gigafactories to train next-generation AI, 30 July 2026
Associated Press (via The Washington Post) — EU lays out $11.4 billion for 7 AI gigafactories, 30 July 2026