Next-Generation AI-Native Datacenters for your GPU clusters

Yes. DC2SCALE has 40 MW of IT power available immediately in France, on more than 50 MVA of connected power: 36 MVA in the Paris region at our PAR9 site, 10 MVA in Castres and 5 MVA in Lille. The engineer who calls you back within 48 business hours tells you the site and the power you can reserve for your project.

In the Paris region, at our PAR9 site, connected at 36 MVA, and on our Edge sites, chosen for available power: Castres, with 10 MVA, and Lille, with 5 MVA. For workloads that must stay close to their users, such as inference, our Hub sites in Paris and Marseille focus on interconnection.

Yes. Our PAR9 site in the Paris region is connected at 36 MVA and hosts multi-megawatt GPU clusters. It complements our Hub sites in Paris-Vélizy and Paris-Orly, chosen for interconnection. The engineer who calls you back within 48 business hours tells you the power you can reserve there.

This offer covers our sites in France. For a project outside France, contact us: we will let you know whether we can meet the need.

Because the long steps of a room project are done before your order. The land of our sites is secured and the power is already connected. Chillers, generators, UPS systems, transformers and room infrastructure are in stock, so nothing waits on manufacturing. Our works team of about 50 employees is ready to start the fit-out.

From 400 kW to 1 MW, the first tranche is delivered in 4 to 8 weeks. Beyond 1 MW, allow 3 months. The exact lead time depends on the site and the cooling you choose: the engineer who calls you back within 48 business hours gives you the handover date.

We keep in stock chillers, gas and diesel generators, UPS systems and battery racks, transformers and room infrastructure: raised floor, access control and building management systems. This stock is held in our logistics and reuse centres in western and northern France, in addition to our regular stock in the Paris region. The fit-out of your room therefore starts without waiting for manufacturers' production lead times.

Our works team of about 50 employees, on permanent and apprenticeship contracts, carries out the fit-out with our design office: power distribution, cooling and room infrastructure. Our electrical switchboards are built and refurbished in our own workshop. The engineer who calls you back within 48 business hours gives you the handover date for your room.

150 kW per rack is a common density on our sites, and higher densities are possible after a study. Our PAR9 site in the Paris region accepts racks well above 150 kW, including NVIDIA Vera Rubin NVL72 racks, AMD Helios racks and Kyber racks of about 600 kW. Tell us the hardware you plan to deploy and we will validate the required configuration.

Yes. Our PAR9 site in the Paris region can host Vera Rubin NVL72 racks, which draw about 190 to 230 kW each and are liquid-cooled with a 45 °C water inlet. It can also host the Kyber racks of Vera Rubin Ultra, announced at about 600 kW and powered by 800 VDC.

Both racks are in the configurator catalogue.

Yes. An AMD Helios rack holds 72 liquid-cooled Instinct MI455X GPUs and draws about 225 to 245 kW. It uses the Open Rack Wide format, 1.2 m wide and 1.3 m deep. Our PAR9 site in the Paris region accepts this format and racks well above 150 kW.

The Helios rack is in the configurator catalogue.

Yes. The architecture is defined with you during the technical study: rear-door, direct-to-chip, immersion, or air cooling if your hardware allows it.

At our PAR9 site, the liquid cooling loop accepts a 45 °C water inlet, the temperature NVIDIA specifies for Vera Rubin.

Yes. DC2SCALE is a Dell partner, and our rooms meet the hosting requirements of the Dell AI Factory with NVIDIA and of Lenovo HPC. They are HPC-ready: power distribution and cooling are sized for racks of 150 kW and more.

At the rate we actually pay our utility, with supporting documents, plus a markup set in the contract. The PUE is contractually capped at 1.35, and electricity is billed separately from the power rent.

The price depends on the site, the power, the cooling method and the adaptations required. The configurator gives an order of magnitude; the firm price is set after the feasibility study carried out by the engineer who calls you back.

No. The configurator is freely accessible. Your details are only requested to obtain a costed answer from an engineer.

It is set case by case, depending on the power and the adaptations required. We discuss it at the first call.

A broker passes your request to several operators and relays their answers. At DC2SCALE you deal directly with the operator running the capacity and the teams: the estimate and the engineer's answer come straight from us.

No. DC2SCALE hosts your hardware; we do not operate a GPU fleet and we do not resell compute. An AI cloud provider can therefore host its infrastructure with us without competing with us.