Retrofit is our specialty.
Reusing what stands beats starting over: faster, far lower in carbon, and considerably more technical.
A data center retrofit converts an existing industrial building or data center to host a new computing load without rebuilding it. It usually covers upgrading the electrical capacity, replacing the cooling plant and bringing physical security up to standard. Its difficulty is contextual rather than technical: you work inside a building that already stands, often still in operation, whose constraints are inherited rather than chosen.
Why retrofit has become the default
Demand for computing power has outgrown both available land and grid connections. A new-build data center needs a site, a substation, planning permission and a grid connection lead time measured in years. A disused industrial building already connected at high voltage, on the other hand, exists today - and its connection was often sized for heavy manufacturing that has since stopped.
That mismatch is where retrofit creates value: where a greenfield project waits for its connection, a retrofit starts from power that is already there. The carbon argument reinforces it, and is no longer cosmetic. A building's structure accounts for a major share of its footprint; keeping it avoids emissions that no later efficiency gain will offset.
That said, retrofit is not a given. A building is not convertible because it is empty; it is convertible if its clear height, floor loading, column grid, grid connection and heavy-goods access allow it. The first job is therefore triage - and it often means saying no.
What changes compared with new build
On a greenfield site, design precedes the building. In a retrofit it is the reverse: the building drives the design. Room layouts follow the existing columns, cable routes follow available penetrations, chiller positions follow what the structure can carry. Every decision is a negotiation with something you did not draw.
The second difference is uncertainty. The drawings of a thirty-year-old building do not describe that building: they describe it as handed over, before the successive modifications nobody recorded. A serious retrofit therefore starts with site surveys and intrusive checks, not with reading as-built files.
The third, when the site is live, is working alongside operations. You intervene next to a load that must not drop, with isolations to obtain, transfers to rehearse and work windows to negotiate. This is where a contractor and a data center contractor part ways.
A retrofit at 240 MW scale: Béthune
We are a contractor on the Béthune programme in northern France: the conversion of the former Bridgestone plant into a data center dedicated to artificial intelligence, developed by Azur Datacenter for Nebius. The site is planned for 120 MW by the end of 2026 and 240 MW in 2027, across 26,000 m² of rehabilitated building - among the most powerful in Europe.
It is a textbook case of what retrofit allows and what it demands. What it allows: an industrial brownfield site, already connected and sized for heavy manufacturing, becomes computing infrastructure without consuming a square metre of farmland. What it demands: structural steelwork to requalify, slabs to verify against loads bearing no relation to the original design, and an infrastructure built for direct liquid cooling - a concept absent from the original building's vocabulary.
This programme is also why we are established in the north. Our Mons-en-Barœul and Cambrai branches are not commercial addresses: they are the bases that serve the site.
Where this expertise applies
The data center packages this expertise covers are shown in full, the others are dimmed. Click a package for detail.
Where energy enters the site, at the interface with the public grid. It caps the available capacity: on a retrofit, what it can carry decides what the building can host.
Medium-voltage switchgear distributes and protects; transformers step the voltage down to usable low voltage. These are very long-life assets, where reuse saves several quarters of lead time.
The main low-voltage switchboard distributes power to the halls and the utilities. This is where discrimination is decided: a fault must be cleared as close as possible without tripping anything upstream.
Chillers and free cooling, producing the chilled water. In the French climate a large share of the year can be covered without compression - provided the loop accepts warm enough water.
The usable space: rack layout, aisle containment, cable routes and raised floors. At AI densities, room geometry matters as much as installed cooling capacity.
What a retrofit engagement covers
Scope varies with the condition of the site, but these seven areas come up almost every time. We take them on as a whole or as separate packages, depending on how your project is organised.
- Convertibility assessment
- Site surveys, intrusive checks, verification of floor loading and clear height, condition of the grid connection and existing substations. Ends in a straight answer on feasibility, including a negative one.
- Electrical capacity upgrade
- Requalifying the incoming substation, replacing or reinforcing medium-voltage switchgear and transformers, rebuilding low-voltage distribution to the target capacity.
- Backup and continuity
- Sizing and installing UPS units, batteries and standby generators, with the associated transfers and load testing.
- Cooling plant and distribution
- Replacing chillers, reworking hydraulic networks, moving to free cooling where the local climate allows, and building infrastructure ready for direct liquid cooling.
- White space fit-out
- Rack layout, aisle containment, cable management, raised floor rework and fire compartmentation.
- Security and detection
- Access control, video surveillance, very early smoke detection, and bringing physical security zones in line with the certification frameworks in scope.
- Supervision and commissioning
- Reworking or creating the building management system, level-by-level testing, staged commissioning and documented handover to operations.
How a retrofit runs
Six stages, the first being by far the most profitable: it is the one where changing your mind is still cheap.
Assessment and feasibility opinion
Surveys, intrusive checks, analysis of the grid connection and the structure. We issue a reasoned opinion on what the building can host - and what it cannot.
Capacity scenarios
Two or three costed ramp-up scenarios with their implications for the grid connection, the structure and cooling. Most of the final cost is decided here.
Detailed design
Our Vélizy-Villacoublay design office produces the electrical, thermal and layout studies, based on the surveyed building rather than archive drawings.
Procurement
New versus refurbished equipment, decided item by item. On long-lead equipment, reuse is often the only way to hold the schedule.
Construction
Phased execution, alongside operations where the site is live, with isolations and work windows agreed in advance.
Testing and commissioning
Level-by-level testing through to load tests and full transfers, then handover with as-built documentation.
Frequently asked questions
Which buildings are genuinely convertible into data centers?
In practice, recent industrial and logistics buildings work best: adequate clear height, wide column grids, slabs designed for heavy loads, heavy-goods access and an electrical connection already sized up. Offices are rarely convertible beyond very small capacities - their floor loading and heights do not follow. The decisive criterion remains the grid connection: without available or obtainable capacity nearby, no other quality of the building compensates.
Is retrofit really faster than new build?
On the part that usually causes delay, yes: the building exists and the connection is often already there, which removes the two longest items of a greenfield project. The design phase, however, takes longer, because an existing building has to be surveyed and understood. The net gain is real but it is won upfront, not on site.
Can a retrofit be done on a live site?
Yes, and it is the most common case. It means working alongside operations: planned isolations, transfers prepared and rehearsed, night or weekend work windows, and a fallback plan for every operation that could affect the load. The methodological overhead is real; shutting down operations costs far more.
Can a retrofit achieve Tier or ISO certification?
Yes, provided the requirements are built into the capacity scenarios. What blocks projects late on is almost always a design decision: genuinely physically separated redundancy, compartmented technical rooms, physical security zones, energy metering. Addressed late, these become expensive rework.
How much existing equipment can be kept?
It depends on its age and on the target capacity. Enclosures, cable containment, substations and some transformers can often be reused. Switching devices, batteries and chillers are more readily replaced, for spare-parts and efficiency reasons. We assess this item by item, against one requirement: whatever is kept must be requalified, not merely put back into service.
Who are you, exactly?
Three very different people land on these pages. Take the door that fits you.
- You areA data center operator or builderYou have works to get done, on a greenfield site or one in service: electrical, cooling, retrofit, decommissioning.Talk about works
- You areThe owner of land or a buildingAn industrial brownfield, a warehouse, a dormant grid connection: we tell you what your site can host, before you commit spend.Have a site qualified
- You areLooking for capacity in FranceAI operator, hyperscaler or enterprise: you need megawatts, a convertible site and a contractor able to deliver them. We are a contractor on 240 MW at Béthune.Talk capacity and sites