When Nebius announced in February 2026 that it would build a 240 MW AI data center on the former Bridgestone tyre plant in Béthune, northern France, the plan was clear: around 120 MW by the end of 2026 and full capacity in 2027, on a site described as already connected to the grid. Seven months later, the picture reported locally is more nuanced, and far more instructive. Some figures are still debated, so we separate what is reported from what remains unknown.
WHAT HAS CHANGED
| Announced (February 2026) | Reported (September 2026) | |
|---|---|---|
| First capacity | Late summer 2026 | Commercial go-live reported around mid-October 2026 |
| Grid power | Site described as already connected | About 45 MW available from the grid at first |
| Full 240 MW | 2027 | Permanent connection reportedly around three years away |
| Bridging solution | Not mentioned | Up to 110 MW of gas generation |
Sources: Reuters and French trade press (February 2026); La Voix du Nord, as relayed by market commentators (September 2026). Figures reported in the press, not confirmed by the companies.
THE BROWNFIELD DELIVERED THE BUILDING
The speed of the building is real. According to Azur Datacenter, the site went from initial construction to live operations in about ten months, unusually fast for AI infrastructure at this scale. The existing site, its buildings and its legacy connection made that possible. This is the brownfield advantage we have described for months.
THE GRID DECIDES THE REST
But an old factory connection is sized for an old factory. Around 45 MW can start a first phase; it cannot power 240 MW of AI compute. The full capacity depends on a new permanent connection, reportedly around three years away. On the reported figures, grid plus gas would cap the site at roughly 155 MW until then, unless additional grid capacity arrives in intermediate steps, which the reports neither confirm nor rule out.
This is the clearest French example of the gap between compute ordered and compute installed. The building and the chips can move in months. The grid moves in years.
WHAT WE DON'T KNOW
The exact timeline. Some market observers point to about 60 MW handed over earlier in the summer. The mid-October date may refer to GPU deployment and revenue rather than power delivery: Nebius has indicated that several weeks can separate the two. It would be wrong to read the October date as a delay without more information.
The grid ramp. The reports give a date for the full connection, not the path to it. Grid capacity could increase in phases before then.
The customer contracts. Whether the capacity sold to customers is sized for the interim level or the full 240 MW is not public.
THE GAS BRIDGE IS NOT A DETAIL
Up to 110 MW of onsite gas generation is a significant project in its own right. In France it raises questions of environmental authorisation, classified-installation rules for combustion plant, emissions and local acceptance, in a country that markets its low-carbon power. It will also shape how the site's electricity origin appears under the new EU data centre rating. And an open question matters for the economics: will this generation serve only the site, or also participate in the electricity market?
LESSONS FOR DEVELOPERS AND LENDERS
Treat the permanent connection date as the real full-capacity date
Price the bridging solution: gas, permits, emissions and acceptance
Model revenue on the grid ramp, not on the building ramp
Check whether announced phases match the connection schedule
None of this makes Béthune a failure. It remains one of the fastest AI deployments in Europe. It shows that a brownfield wins the first phase, and that the grid still decides the full one. For a lender, that difference is the whole revenue curve. See our due diligence for lenders and funds and our earlier analysis of why Nebius chose Béthune.
FREQUENTLY ASKED QUESTIONS
How much grid power is available at Nebius's Béthune data center?
According to La Voix du Nord, as relayed in September 2026, about 45 MW is available from the grid at first, with the permanent connection needed for the full 240 MW reportedly around three years away.
Why is Béthune using gas generation?
To bridge the gap between the power available from the existing connection and the site's target capacity until a permanent grid connection is in place. Up to 110 MW of gas generation has been reported.
Was the Béthune brownfield a good choice?
It allowed an unusually fast building, about ten months, and a quick first phase. But the existing connection only covers part of the target capacity, so the grid still sets the timeline for full capacity.