Nebius anchors Volt's €7bn, 200 MW Rotterdam AI data center
Nebius is the first tenant for Volt's planned Rotterdam AI data center, FD reported on 12 October. The letter of intent lets Volt try to raise the €7bn, 200 MW first phase, but first capacity is not due until late 2029 or early 2030 and Nebius can reportedly walk away if it slips.
Nebius has signed a letter of intent to be the first tenant of Volt's planned AI data center on the Rotterdam Maasvlakte, and the second-order effect lands on the host, not the GPU renter: a named tenant is what lets Volt try to raise the €7 billion first phase of about 200 MW, with first capacity not due until late 2029 or early 2030.
What happened?
- The deal: On 12 October 2026 the Dutch financial daily FD reported that Nebius, the Amsterdam-based AI cloud company, is the first customer for the AI factory Volt wants to build in the Port of Rotterdam. NL Times, citing FD, says Nebius signed a letter of intent to take "a significant portion" of the planned capacity. AD reported the same morning that Nebius has stated its intention to take capacity and that the details still have to be worked out.
- The size: The first phase is two building sections totalling about 200 MW, and FD puts its cost at €7 billion. Techzine reports that about €2.5 billion of that goes to buildings and infrastructure and €4.5 billion to chips and other equipment, including 100,000 advanced AI chips for Nebius alone. Citing De Telegraaf's sources, Techzine says Nebius will take up to 200 MW.
- The timeline: First capacity is expected in late 2029 or early 2030, according to Techzine, and FD says the campus is meant to open in 2030. Volt's founder, Han de Groot, told FD he hopes to have the permits this year and then start building.
- The exit clause: NL Times, citing FD, reports that Nebius can end the agreement if the project is delayed.
- What Nebius said: In AD, Nebius chief executive Arkady Volozh said the company supports the Rotterdam AI Gigafactory and is working with Dutch entrepreneurs to strengthen the Netherlands' own AI capacity. Neither company has published the terms.
Why does it matter?
The direct signal. A data center developer with no tenant has a site plan, not a business. Volt's own page describes a campus of up to 800 MW with Eneco as its energy partner. Emerce reports that Volt's first capacity is 14 MW at Switch in Amsterdam. A named tenant is the first thing banks and funds ask for, and De Groot told FD as much: building only becomes possible once there is a customer. Techzine describes the structure the same way. Nebius leases the computing capacity for the long term, and funds and banks lend against that lease to pay for both the building and the hardware.
Our arithmetic on the first phase.
- €7 billion for about 200 MW is roughly €35 million per MW, all-in. That covers the building, the power and cooling, and the chips.
- The building-and-infrastructure share of €2.5 billion is about €12.5 million per MW. The chips-and-equipment share of €4.5 billion over 100,000 chips is about €45,000 per chip, and that figure also covers networking and other gear.
- 200 MW over 100,000 chips is about 2 kW per chip. The reports do not say whether 200 MW is the load of the computers alone or the whole site, so treat that as a rough ceiling.
- The first phase is a third of the 600 MW campus that FD and NL Times describe, or a quarter of the 800 MW on Volt's own page. The two sources do not agree on the full size, so we treat 200 MW as the only firm number.
The second-order signal: the host carries the risk, the tenant keeps the option. This is the same pattern we covered on 11 October, when AIB said Nebius prepayments fund its 50 MW build. Nebius signs early to lock in powered halls in Europe. The developer uses the signature to raise money. In Rotterdam, though, the commitment is softer. It is a letter of intent, not a lease, and the reported exit clause means a delay at Volt costs Volt the tenant. Nebius loses only time. For the AI Data Centers layer, the deal shows that powered capacity in Europe is scarce enough for a GPU cloud to reserve halls about three years before they exist.
What does it change?
Near term (to the end of 2026)
- The permit clock starts. De Groot hopes to have permits this year. With a tenant now named and an exit clause tied to delay, every month of slippage puts the anchor tenant at risk.
- Financing becomes testable. The project is privately financed. The Dutch government said on 31 March 2026 that it supports AI gigafactories that the market pays for in full, and Techzine notes that the Netherlands did not apply for the EU's €20 billion InvestAI gigafactory fund. So the €7 billion has to come from lenders and investors. The next real test is a named bank or fund and a signed lease.
- Power is the selling point. Techzine says the campus connects to the high-voltage grid where North Sea wind comes ashore, not to the local grid that serves homes and businesses. De Groot describes the Dutch constraint as transmission, not generation. The plan rests on that existing connection.
Longer term (2027 to 2030)
- Europe's neoclouds keep pre-buying halls. If Rotterdam closes financing, it gives other European developers a template: sign a GPU cloud first, then raise debt against the lease. If it stalls, the lesson runs the other way, and letters of intent will count for less with lenders.
- The chips arrive last. €4.5 billion of the €7 billion is for hardware that will not ship until the halls are ready around 2029 or 2030, so the chip generation, and its power per rack, is still open. A longer build makes it more likely that the hall design has to change before it is finished.
- How long the build takes matters. A late 2029 or early 2030 start fits the multi-year timelines in our explainer on how long it takes to build an AI data center.
| Metric | Value | Source name | Date |
|---|---|---|---|
| Tenant commitment | Letter of intent, "a significant portion" of capacity | NL Times, citing FD | 12 Oct 2026 |
| Nebius share | Up to 200 MW | Techzine, citing De Telegraaf | 12 Oct 2026 |
| First phase | About 200 MW in two building sections, €7bn | FD; NL Times | 12 Oct 2026 |
| Cost split | €2.5bn buildings and infrastructure; €4.5bn chips and equipment | Techzine | 12 Oct 2026 |
| Chips for Nebius | 100,000 advanced AI chips | Techzine | 12 Oct 2026 |
| First capacity | Late 2029 or early 2030 | Techzine | 12 Oct 2026 |
| Full campus | 600 MW (FD) vs up to 800 MW (Volt) | NL Times; Volt | 12 Oct 2026 |
| Volt's first capacity | 14 MW at Switch, Amsterdam | Emerce | 12 Oct 2026 |
| Cost per MW, first phase | About €35m all-in (our arithmetic) | From FD and Techzine figures | 12 Oct 2026 |
What should we watch next?
- Permits before 31 December 2026. De Groot's own target. A slip into 2027 would test the exit clause.
- A binding lease. The letter of intent has to become a signed contract with a term, a price and a volume before lenders treat it as collateral.
- A named lender or fund for the €7 billion first phase, and how much of it is debt.
- A Nebius statement or filing. Nebius has not published the terms. Its next quarterly report would be the natural place to describe the commitment; the date has not been announced.
- What would weaken the catalyst: no permit by year-end, a second tenant that never appears, or Nebius using the reported exit clause.
Also on the tape
- 8 Oct: Marvell said it will show its 102.4 Tbps Teralynx T100 data center switch at the OCP Global Summit in San Jose, 12 to 15 October (Marvell press release).
- 9 Oct: Micron said the base dies for its NVHBM and HBM4E memory will be made by an outside foundry, and NVIDIA says NVHBM gives up to 30% more bandwidth and 15% lower HBM power than standard HBM4E (TrendForce, citing The Elec).
- 1 Oct: JERA, Dell Technologies and RHAELM signed a memorandum for an AI data center of up to 400 MW at JERA's Chiba thermal power station, targeting operations around 2028 (JERA press release).
- 15 Oct: TSMC reports third-quarter 2026 results at 14:00 Taiwan time (TSMC investor relations).
FAQ
Is the Nebius deal a binding contract?
Not yet, as reported. NL Times, citing FD, calls it a letter of intent, and AD says the details still have to be worked out. Emerce describes a multi-year contract. Until the companies publish terms, treat it as a reported commitment.
Why would a GPU cloud sign for halls that open in 2030?
Because powered, high-voltage-connected space in Europe is scarce, and signing early secures it. The reported exit clause means Nebius gives up little if the project slips.
Is this one of the EU's AI gigafactories?
No. The Dutch government did not apply for the EU's InvestAI gigafactory fund, so the project is privately financed. Volt still uses the name "AI Gigafactory" for the campus.
How big is the full campus?
The sources differ. FD and NL Times describe 600 MW, and Volt's own page says up to 800 MW. Only the first phase of about 200 MW has a reported cost and a tenant.
Takeaway: Nebius gives Volt the tenant it needs to raise €7 billion, but the reported exit clause leaves the delay risk with the host, so permits and a signed lease are the next real tests.
Figures come from the dated sources listed under References below. Our own arithmetic is shown and labelled as ours. Where reports disagree, we say so.
Last reviewed: 12 Oct 2026 · Next review: when Volt reports permits or financing, or Nebius publishes the terms
Spot an error? Tell us and we will correct it and note the change here.
Our editorial standards →
References
- FD (Het Financieele Dagblad), "Eerste klant voor groot AI-datacenter bij Rotterdam – nu gaat de klok tikken", 12 Oct 2026
- NL Times, "Cloud company Nebius signs on as first client at new major data center", 12 Oct 2026
- Techzine Global, "Nebius is the first customer of a Rotterdam-based AI gigafactory", 12 Oct 2026
- AD, "Veelbelovend cloudbedrijf verbindt zich als eerste afnemer aan plan voor AI-gigafabriek in Rotterdamse haven", 12 Oct 2026
- Emerce, "Nebius eerste klant Rotterdams AI-datacenter", 12 Oct 2026
- Volt, "The Rotterdam AI Gigafactory" (company page), checked 12 Oct 2026
- Marvell, "Marvell to Showcase Open AI Infrastructure Solutions at OCP Global Summit 2026", 8 Oct 2026
- TrendForce, "NVIDIA's NVHBM Reshapes Base Die Strategies", 9 Oct 2026
- JERA, "JERA, Dell Technologies and RHAELM Sign MoU to Accelerate National-Scale AI Infrastructure Development in Japan", 1 Oct 2026
- TSMC Investor Relations, "2026 Q3 Quarterly Results" (conference notice), checked 12 Oct 2026