Power
Sized to your subscribed load
Power feed, redundancy, generators and fuel autonomy. You plug in, it holds.
AI clusters
Your AI cluster in France, from a few racks to several megawatts
You bring the GPUs. We bring the power, the cooling, the security and the teams who stay on site. First tranche delivered in 4 to 8 weeks.
How much does your cluster draw?
NVIDIA
AMD
Ancillary racks: network, storage, CPU nodes
- Mellanox Quantum-X800 InfiniBand Counted per rack, 15 kW
- Mellanox Spectrum-X Ethernet Counted per rack, 12 kW
- Baie stockage Counted per rack, 8 kW
- Baie réseau et management Counted per rack, 6 kW
- Baie nœuds CPU Counted per rack, 12 kW
Other hardware
Your configuration
Pick a hardware type to start.
IT power
—
—
per rack at most
—
racks
In four figures
Density
128 kW per rack: a routine case for us, not a ceiling
Lead time
4 to 8 weeks for a first tranche up to 1 MW
Energy
1.35 PUE contractually capped
Answer
48 h working hours for a costed answer from an engineer
What we cover
You run the cluster. We run everything around it.
Cooling
The architecture follows your hardware
Chilled water and free cooling across our sites. Rear-door, direct-to-chip or immersion depending on what you deploy.
Operations
Our own staff, on site, 24/7
Teams on site, on-call rota, badge and biometric access. They are our own staff, not a passing contractor.
Density
One GB300 rack draws as much as ten classic colocation racks
:quality(65))
The problem
This is no longer about floor space
A colocation room fills up when the space runs out. A cluster room fills up when the power runs out, often while half empty.
Our answer
Rooms built for power density
128 kW per rack is routine for us. Power distribution and cooling are sized for it from the start, not retrofitted.
Liquid
A loop that follows your hardware
Rear-door, direct-to-chip, immersion: the architecture is decided with you before signing.
Location
Two ways to site a cluster in France
Greater Paris
For what needs to sit close to the network
Peering, carrier interconnection, proximity to your teams and your customers. Inference and production.
Regional sites
For what needs to be big
More available power, more favourable energy costs, room for the next tranche. Training and long-run capacity.
Energy
You pay for your consumption, not for our efficiency
Spot
Indexed on the market
Your bill follows the market, the following month
Electricity is rebilled at the market spot price plus market. No fixed three-year rate that keeps you paying for a peak that is long gone.
1.35
PUE cap
Beyond that, it is on us
Datacenter overhead is contractually capped at 1.35. An operator that cools badly does not bill you for it.
MW
What you reserve
Power, not a flat rate
The contract covers megawatts made available at all times. Energy consumed is rebilled separately, on meter readings.
Lead times
From first description to power-on
1.
You describe your hardware
The configurator turns your fleet into power, density and rack count. Five minutes, no appointment.
2.
An engineer calls you back
Within 48 working hours. We tell you what is available, where, and by when.
3.
You deploy
400 kW to 1 MW in 4 to 8 weeks. 1 to 3 MW in 3 to 4 months. Your teams get access from delivery.
High density
What it looks like when a customer really pushes the power
:quality(80)){kind=logo}
:quality(80))
No middleman
You talk to the people who hold the breakers
We operate what we sell: the capacity is ours, so are the teams. An estimate without an NDA, a file treated as a file, an engineer who answers within 48 working hours.
Your cluster
Tell us what you are deploying
Five minutes in the configurator is enough to know what it means in power, racks and budget. An engineer then calls you back with a date.
:quality(50))