Europe asked who would build its AI compute. Seventy-six answered, and a grid connection will decide which of them matters.

When the European Commission closed its call for expressions of interest in AI gigafactories on 20 June, it got what it called an overwhelming response: 76 submissions, spread across 60 sites in 16 member states, from a field that ran from data-centre operators and telcos to power suppliers and financiers. The programme they were chasing, funded through the roughly 20 billion euro InvestAI envelope, will back at most five facilities, each meant to house more than 100,000 advanced chips and draw on the order of a gigawatt. The formal call for proposals, slipped twice, is only now landing, with selection promised by year-end and the first sites meant to run in 2028. On the demand side, in other words, Europe has no problem at all.

The Aether consortium shows what a serious bid looks like, and where it gets stuck. Unveiled on 8 July, it pulls together the chip designer SiPearl, the server maker 2CRSi, the property developer Nhood and Dassault Systemes, and it has named its site: two industrial parcels in the Strasbourg region, FR-SXB1 and FR-SXB2, with acquisitions due to close by the end of October and December. The first hall is meant to switch on in 2027 and the group talks about eventually reaching more than 400 megawatts. The phrase that matters sits in the fine print of that ambition, which is available subject to grid availability and the proposals of RTE, France's transmission operator. The chips can be bought. The connection has to be granted.

That caveat is the whole industry in miniature. Across Western Europe's busiest data-centre markets, the queue for a grid connection now runs seven to ten years, and stretches past a decade in the most congested corners of Frankfurt, Amsterdam and Dublin. Ireland has parked new Dublin connections behind a moratorium into 2028; parts of the Netherlands and the Frankfurt area are effectively closed to fresh load until around 2030. In April the grid body ENTSO-E moved the worry along a step, warning that the risk was no longer that data centres could not get power but that they might absorb so much of the continent's spare generation that operators would have to hold back renewables to keep the system stable. A gigafactory that needs a gigawatt is not a tenant a stressed grid welcomes.

So the real contest is not for capital, which is plainly abundant, but for sites where the power is already spoken for. France has understood this more clearly than most and pre-allocated 18 gigawatts of grid capacity to data centres, handing it out on a first-ready, first-served basis through 2026. Read plainly, that is an incentive, but a new kind: not a tax holiday or a cash grant, but a reserved megawatt with a date attached. It rewards the developer who has done the unglamorous work of securing an interconnection over the one with the glossiest proposal, and it tells every rival jurisdiction what the currency of this cycle actually is.

For a practitioner the lesson travels well beyond the five European winners, because the same logic now governs any site pitched at a hyperscaler or a sovereign compute fund from Ohio to Johor. Competitiveness in this cycle is measured in the date a transmission operator will put in writing, not in acreage or headline incentives. The agencies that win will be the ones that pre-negotiate interconnection, publish honest energisation dates, and steer investors toward parcels near substations and firm generation rather than toward whichever industrial park has land to fill. The bravest and most useful thing an agency can tell an investor in 2026 is a true connection date, even when it is years out, because the alternative is a signed deal that stalls in a queue nobody disclosed.

Watch the awards at year-end, and in particular whether the Commission can fund more than two of its five gigafactories before 2028, the split earlier reporting flagged as the realistic near-term ceiling. If the binding constraint is power rather than money, the programme's real test is not how much it mobilises but how quickly member states can clear a connection. For agencies thinking about their own version of this, the Doyen Report on power-ready sites is the companion piece: the checklist there, on megawatts, water and substation distance, is now the first page of a site pitch, not the last.