EU AI Gigafactories: €30B Planned, €1B Committed

What happened

EuroHPC, the EU body that runs its public supercomputers, opened its AI Gigafactories tender on 30 Jul 2026. It seeks up to seven sites in at least seven member states: four medium sites that must reach 120 MW of IT load in Phase 2 and three large ones at 150 MW. On 18 Oct 2026, DC Market Insights published a detailed read of the tender specifications. If all seven are awarded, the floor is 930 MW of new AI capacity.

Bids are due on 12 Nov 2026, according to the Commission and EuroHPC. One detail is unsettled: when DC Market Insights checked on 18 Oct 2026, the EuroHPC call page still listed 3 Dec 2026, and no correction had been published.

The call aims to mobilise more than €30 billion: up to €10 billion of EU and national money and at least €20 billion of private capital. The EU's own share is up to €5 billion, and only €1 billion of it, the Phase 1 money, is legally committed today. The other €4 billion depends on the 2028-2035 EU budget, which member states have not agreed.

Private money is not waiting. On 12 Oct 2026, Nebius signed as the first tenant of VOLT's €7 billion AI campus in Rotterdam, which is entirely privately financed and sits outside the EU scheme. PitchBook reported on 16 Oct 2026 that high-yield bond issuance for data center expansion has reached $50.1 billion this year, against $13.6 billion in all of 2025.

What the chart shows

The squares split the planned €30 billion by how firm each slice is. The EU's committed Phase 1 money is €1 billion, or 3.3% of the plan. The indicative Phase 2 money adds €4 billion. Member states must at least match the EU's share, and private capital is the largest slice: at least €20 billion, or 67% of the plan.

The per-site numbers explain why. The EU ceiling is €1 billion for a large site and €500 million for a medium one, and never more than 17% of computing-infrastructure capex. To collect the full €1 billion, a large site must spend about €5.9 billion on computing alone. DC Market Insights' build-cost model, at US$44.6 million per MW, puts a 150 MW site at about US$6.7 billion all in. With the host state matching, DC Market Insights estimates that public money can fund up to about a third of computing capex.

The second chart shows where the rest is coming from. PitchBook counts $66.6 billion of high-yield bonds and institutional leveraged loans for data centers so far this year. Bond issuance is already 3.7x its 2025 total, and loans are 8.3x. These are dollar markets and the EU figures are in euros, so we compare direction, not size.

What it means for founders

The tender is less a subsidy than a sales contract. In DC Market Insights' reading, the award's value is five years of guaranteed compute purchases from the EU and member states, and that anchor contract is what lets a bid team borrow the rest. The scoring says the same thing: financial feasibility counts for 40% of the first-stage score, as much as technical quality.

We read that as a distribution point. In AI infrastructure the scarce asset is not the GPU, which any well-funded team can order. It is the committed buyer, the grid connection and a lender who trusts both. In Rotterdam, VOLT's campus took no EU money; what it has is an anchor tenant, Nebius, which has leased up to 200 MW. The tender offers bidders the same asset from the public side: a guaranteed purchase of compute time for five years from the EU and member states.

For founders, the useful move is to sell into the bid teams, not around them. Each bid needs a co-funding letter from its host state, Phase 1 capacity ready within 18 months of signing, and it may reserve part of its procurement for European start-ups. Companies that sell grid connections, cooling, power and scheduling software will have a named list of buyers once awards land in early 2027.

What would change our mind

We read the gigafactories as an anchor-contract market whose public money is mostly a promise. Two things would change that. If the 2028-2035 budget commits the €4 billion for Phase 2 before the awards, the EU becomes a real co-investor and the subsidy starts to matter as much as the contract. If fewer than seven sites are awarded because bids cannot show financing, the anchor contract was not enough to unlock private debt.

On the private side, CoreWeave bond yields rose to 15% in mid-October, PitchBook reported, on concern about its ability to service its debt. If that caution spreads to the loan market, this year's $66.6 billion will look like a peak, and a signed public buyer becomes one of the few ways left to finance a site.

The gigafactory award is a signed buyer with a subsidy attached, not a subsidy with a buyer attached.

What we would back: Companies that sell power, cooling and scheduling software into EU gigafactory bid teams before the awards in early 2027.