Brussels is putting €10 billion into seven AI gigafactories, but you should read the number correctly. This is Europe buying a foothold in the AI infrastructure race, not catching the US in one tender.
The numbers are real. The ambition is genuine. And the timing tells you how far behind Europe already is. The Associated Press reported Thursday that the European Union is offering €10 billion in public funding for seven AI gigafactories, with the European Commission trying to pull in another €20 billion from private investors. Each site is meant to run at least 100,000 advanced AI chips, roughly four times the power of the AI data centers now operating in the EU.
The bid deadline is November 12, 2026. Companies, public bodies, investors and consortia can apply, and the sites can be built in one location or spread across member states. That matters for Europe because the easiest place to build a large AI cluster is rarely the same place with the most researchers, startups or government demand.
The interest is not theoretical. The Commission's earlier sounding round brought in 77 proposals across 16 member states and 60 possible sites, according to its AI gigafactories page. France has moved most visibly. AION, the consortium led by Scaleway and backed by partners including Ardian, EDF, Orange, Capgemini and the iliad Group, has proposed infrastructure equivalent to more than 288,000 Nvidia H100 GPUs. Romania has put forward the Black Sea AI Gigafactory, with proposed sites at Cernavodă and Doicești and more than 100,000 AI accelerators, according to Romanian government filings and AGERPRES. Spain's bid includes sites in Móra la Nova in Tarragona and San Fernando de Henares near Madrid, with Telefónica, ACS, Santander and Multiverse Computing involved.
That is a serious field.
It is also where Europe's structural problem shows up most clearly: the bloc is trying to coordinate by committee while the American hyperscalers spend like infrastructure ministries. The EU is not short of smart people. It is short of compute, cheap power and fast permitting - and companies willing to spend tens of billions before the revenue is obvious.
The Spending Gap Is Brutal #
Stanford's 2026 AI Index put US private AI investment in 2025 at $285.9 billion, more than 23 times China's $12.4 billion. Europe is not in the same capital conversation. Amazon said in Andy Jassy's 2025 shareholder letter that it expected about $200 billion in capital expenditure in 2026, much of it tied to AWS and AI infrastructure, after spending $131.8 billion in 2025. Microsoft reported $41 billion of capital spending in its latest quarter and, according to AP, kept a 2026 capex forecast of $175 billion after accounting adjustments. Meta has pushed its 2026 capital expenditure forecast as high as $145 billion.
Against that, €30 billion spread across seven European sites is not parity. It is a floor.
Frankly, that is still better than pretending the market will fix this by itself. European startups and universities cannot train frontier systems if the practical answer to every serious compute need is AWS, Azure or Google Cloud. You can talk about sovereignty all day, but if the chips, software stack, data center and billing relationship sit outside Europe, the sovereignty is mostly a slogan.
The Commission knows this. AP cited a Commission report warning that Europe's five largest cloud providers are all American and that dependence on hyperscale cloud and AI providers creates risks for highly critical use cases. That is the part founders should pay attention to. The issue is not whether a French or Romanian data center sounds patriotic. It is whether a European AI company can train and deploy serious models at scale without building its whole business on someone else's infrastructure terms.
Power May Decide More Than Policy #
The harder constraint is electricity. The Interface report published in May put the problem plainly: large AI clusters behave less like normal server rooms and more like energy-intensive industrial plants. In the most congested European data center markets, including Frankfurt, London, Amsterdam, Paris and Dublin, grid connection waits can run seven to 10 years, and even longer in the worst cases. You cannot run 100,000 accelerators on a press release.
This is why the country-level bids are starting to look like energy bids. Romania is pointing to nuclear power at Cernavodă and a future small modular reactor site at Doicești. Finland is building around the LUMI AI Factory environment in Kajaani, where CSC says a new data center is already under construction in the Renforsin Ranta area. Spain is leaning on public-private money and specific sites rather than a vague national promise.
Good. Specificity is the only thing that will save this project from becoming another Brussels acronym.
The EU already has 19 AI factories across 16 member states, plus 13 antennas, according to the Commission's AI factories policy page. Those sites are meant to give startups, SMEs, researchers and public bodies access to AI-optimised supercomputers. That is the base layer. The gigafactories are the next step up: fewer, larger facilities built for frontier-scale training and deployment.
Building them is not the finish line. It is the admission fee. If the tooling is worse than Azure, if the power contracts are too expensive - and access gets tangled in public procurement rituals on top of that - founders will do what founders always do. They will go where the work gets done fastest.
Europe does not need seven monuments to technological sovereignty. It needs seven places where a serious AI team can get compute, data, support and predictable pricing without leaving the bloc. That is the test. The tender opened on July 30, 2026, but the real answer will come years later, when you can see whether European startups are actually training at home or still renting the future from someone else.
Also read: Microsoft posted record fiscal 2026 revenue as Azure crossed $100 billion and proved its AI bet is paying off • Encore AI raises $30M after its bank and insurer customers decided to become investors • Zuckerberg bets billions of personal AI agents on Meta's most expensive quarter ever