Custom-fit modular data centers for AI and edge compute

A complete data centre, or a critical subsystem of one, engineered and tested in a factory before it is delivered. Racks, power distribution, cooling, fire suppression, monitoring and access control arrive as one integrated thing instead of as a construction project.

The point is not that it sits in a container. The point is that the difficult, expensive integration work happens somewhere it can be tested properly, and site work shrinks to foundations, utilities and connection.

Because fabrication and site works run in parallel, and a conventional build cannot do that. Six to nine months from order, against two to three years for a building.

Worth being straight about this: that is a property of the method, not something unique to us. Any competent modular supplier can make the same argument honestly. What differs between suppliers is what the module is built to, and whether it survives your tenant's technical review.

Yes, and the useful answer depends on which end of the range you are at.

Low-power inference silicon runs comfortably on air at 10 to 20 kW a rack. A GB300 NVL72 rack draws up to 142 kW maximum IT load, split 128 kW to direct liquid cooling and 14 kW to residual air, and there is no air-only version of that. We build for both ends. Part of the first conversation is working out which one your workload actually needs, rather than which one sounds more impressive.

Six to nine months from order to a working data centre, with factory build and site works running in parallel.

Some modular suppliers publish faster than that. If schedule is the single variable that decides your project, say so at the start and we will tell you honestly whether we are the right fit. Better that conversation now than at contract.

Cooling is selected against the site, not picked from a catalogue. Air handles the lower densities through in-row, precision, indirect and direct free cooling. Rear-door heat exchangers on a closed chilled-water loop handle the middle of the range. Warm-water direct liquid cooling at 41/51 degrees C handles GB300 density.

Dry operation is the normal mode. Adiabatic assistance runs from on-site storage, only when ambient conditions require it, so no permanent utility water connection is needed. Annual energy and water performance is established by site-specific climatic simulation, which is why we do not publish one figure and pretend it holds everywhere.

Dust, sand, humidity and vibration hardening are standard options, and EMP or IEMI shielding is available where the site calls for it. We have built shielded modules for exactly that requirement.

We do not publish a single operating temperature range. The honest limit depends on the configuration inside the module and the site around it, so it is confirmed per project at design review rather than printed as one number that would be wrong for half the people reading it.

Yes, and it is worth designing for at step one rather than discovering the constraint later. Modules are sized for standard road transport, so relocating one does not turn into a permitting exercise with escort vehicles.

Redeployability is one of the genuine structural advantages of modular over a building. If there is any chance the site changes, say so early. It costs almost nothing to plan for and a great deal to retrofit.

Both, depending on the project. Larger and more complex builds we usually take direct. Where a system integrator already owns the client relationship, we work behind them.

Tell us which situation you are in and we will be straightforward about how we would run it, including the cases where we think the other route serves you better.