Europe opened a ten billion euro tender for AI gigafactories on 30 July 2026. The call is Europe’s attempt to buy its way back into a race it has been narrating from the sidelines.
The European Commission is offering €10 billion ($11.4 billion) in public funding for up to seven AI gigafactories, and wants the money to attract a further €20 billion in private investment, according to Reuters via Yahoo Finance. The tender closes on 12 November, successful bidders are expected in early 2027, and facilities are to be operational within eighteen months of contract signing.
What is actually on offer
The structure splits into two categories. The first supports up to four projects. The second supports up to three larger ones.
Each gigafactory combines advanced AI processors, cloud and software systems, high-speed connectivity and energy-efficient data centres. Access is aimed at start-ups, scale-ups, small and medium enterprises, industrial companies, universities and public authorities — organisations that cannot currently rent frontier-scale compute on commercial terms, per Broadband Breakfast.
The Washington Post framed the announcement plainly: the European Union laying out money to catch up with the United States and China.
The number in context
Ten billion euros of public money sounds substantial until placed beside the private commitments in play elsewhere.
Nvidia has been in talks over a roughly $250 billion financial backstop for a single OpenAI data-centre lease in Ohio, with separate discussions on up to $350 billion in chip financing. One American company negotiating one facility exceeds the entire European public programme by a factor of twenty-five.
Even at the €30 billion blended figure, Europe is proposing across seven sites and several years what a single hyperscaler commits in a quarter.
Europe is bringing a public procurement process to a capital expenditure race. The process may be better designed. The cheque is not the same size.
Why the design might matter more than the money
There is a serious argument for the European approach, and dismissing it on headline numbers alone would be lazy.
American AI compute is overwhelmingly private, concentrated, and rented on the owner’s terms. A European researcher, a Portuguese manufacturer, or a Polish start-up gets access to frontier compute by satisfying a commercial vendor’s credit and contract requirements — or does not get access at all.
A publicly funded gigafactory with a mandate to serve universities, small enterprises and public authorities is a different instrument. Access is allocated by policy rather than by purchasing power. If the facilities are built and the access rules are honest, the model produces something the American stack does not: compute as public infrastructure.
The risk is equally clear. Public infrastructure programmes are slow, and eighteen months from contract signing to operation is optimistic for facilities of this scale. Hardware procured to a 2027 specification will meet a 2029 frontier.
Three blocs, three theories
Put the month’s policy news side by side and a pattern appears.
💡The United States restricts the export of advanced chips, including to Chinese firms operating outside China. The theory: control the compute, control the frontier.
💡China is weighing controls on downloads of its own model weights, training data, and foreign fabrication of its chip designs. The theory: control the intelligence, and stop giving the advantage away.
💡Europe is funding shared compute and allocating access by policy. The theory: if you cannot own the frontier, own the on-ramp.
One more design question decides whether the programme matters. Allocation rules are not yet published, and allocation is where public compute either reaches a Latvian university and a Portuguese manufacturer or quietly reverts to the largest national champions in each member state. Europe has run industrial programmes both ways before. The tender documents will show which instinct won.
The view from Johannesburg
I write from South Africa, and every one of these three theories has the same blind spot.
No bloc is building compute for the global South. The European gigafactories serve European institutions. American capacity serves whoever can pay American prices. Chinese open weights were the one genuinely accessible route for a team in Lusaka, Lagos or Nairobi — and Beijing is now weighing whether to close that route.
I wrote about that closure earlier this week. Set the two stories together and the shape is unmistakable. Three powers are each building or withholding compute according to their own strategic interest, and the countries with the youngest populations and the fastest-growing developer bases are in none of the rooms.
The argument for what I call Emergent Intelligence (EI) — a dignity-first frame for what the world calls artificial intelligence — includes an argument about access. A technology that reasons about human affairs, and whose infrastructure is allocated by three governments to their own citizens, will encode whose affairs count. Africa does not need a seat at Europe’s tender. Africa needs its own gigafactory programme, and the case for one gets harder to make every month the others build.
Frequently Asked Questions
These are the questions readers have been asking since the European Commission opened the call. Short answers follow, drawn from the tender documents and published reporting.
What is an AI gigafactory?
In short, an AI gigafactory is a large publicly co-funded facility combining AI processors, cloud systems, connectivity and energy-efficient data centres for training and fine-tuning advanced models. The answer is that Europe plans up to seven. The key is that access is designed for universities, small enterprises and public authorities rather than sold purely commercially.
How does the European commitment compare?
Data from the Commission shows €10 billion ($11.4 billion) in public funding, intended to attract a further €20 billion privately for roughly €30 billion total. Research into comparable private commitments reveals a single American data-centre financing discussion of around $250 billion.
Why is Europe funding compute rather than models?
European institutions cannot reliably rent frontier-scale compute on commercial terms. According to the Commission, the facilities target start-ups, scale-ups, industry, universities and public authorities. The answer is that Europe is buying the on-ramp rather than competing directly on frontier models.
Who is eligible to bid, and when?
The tender closes on 12 November 2026, with successful bidders expected in early 2027 and facilities operational within eighteen months of signing. In other words, the earliest realistic capacity arrives in 2028.
Which risks could derail the programme?
Analysis of the programme demonstrates three: a funding gap against private American capital, procurement timelines that risk delivering 2027 hardware to a 2029 frontier, and allocation rules that will decide whether the compute genuinely reaches smaller organisations. Evidence from earlier European technology programmes shows execution, not ambition, is the binding constraint.
Sources
