OFFER Limited time only: -10% site-wide with code WELCOME10

GPU colocation in Paris: host your AI servers

GPU server colocation at the Equinix PA6 data center. You keep your hardware and your models; we provide the rack, the power and the network.

  • From 800 W to 3 kVA
  • Free 10 Gbps uplink
  • Redundant A/B power
  • Data hosted in France
See the plans, from 399 €/month

Our GPU colocation plans

Fixed-price plans sized for the power draw of a GPU server. Request a quote in one click and we get back to you within 24 business hours.

GPU Starter

A 1 to 2 GPU machine for inference, fine-tuning or image generation.

399 €/month excl. VAT
  • Reserved power800 W
  • Rack space2U
  • Guaranteed bandwidth100 Mbps
  • IPv4 addresses2
  • 10 Gbps uplink portFree

GPU Cluster

Several GPU servers side by side, connected in the same rack space.

1,498 €/month excl. VAT
  • Reserved power3 kVA
  • Rack space8U
  • Guaranteed bandwidth500 Mbps
  • IPv4 addresses8
  • 10 Gbps uplink portFree

Custom

An 8-GPU server, more than 3 kVA or half a rack: build your own space.

On quote
  • Rack space1 to 24U
  • Power0.5 to 6 kVA
  • Guaranteed bandwidth50 to 1,000 Mbps
  • 10 Gbps uplink portFree
  • Tier III+ infrastructure (N+1 redundancy)
  • IPMI / KVM monitoring on request
  • Multi-carrier transit
  • IPv6 /64 included on every plan
  • Setup fee: €49.99 (waived with a 12-month commitment)

Why host your GPU server in a data center

A multi-GPU machine runs hot, makes noise and draws as much power as a heater. It belongs in a rack, not under a desk.

Reserved electrical power

Each plan reserves a power envelope, from 800 W to 3 kVA, on redundant A/B power: if one feed fails, the other takes over without interruption for your running training jobs.

Network to serve your models

Free 10 Gbps uplink port, multi-carrier transit, guaranteed bandwidth of 100 to 500 Mbps depending on the plan, dedicated IPv4 and IPv6 /64. Enough to expose an inference API or pull in your datasets.

Your data stays yours

The server belongs to you and stays in Aubervilliers, at the Equinix PA6 data center. Training data, model weights, prompts: nothing goes through a third-party cloud platform.

Remote control of your machine

IPMI/KVM access on request to reboot, reinstall an NVIDIA driver or switch kernels without travelling. Our team receives, racks and cables your hardware.

How much power does your GPU server need?

In GPU colocation, power determines the plan, well before rack space. To estimate the power to reserve: add up the maximum draw of each GPU and of the processor (CPU), then about 150 W for the motherboard, RAM, NVMe SSDs and fans. Then add a 20% margin to absorb load peaks.

Graphics cardVideo memoryMax. power drawTypical uses
GeForce RTX 409024 GB GDDR6X450 WLLM inference up to ~30B quantised parameters, Stable Diffusion, LoRA fine-tuning
GeForce RTX 509032 GB GDDR7575 WSame use as the 4090 with more video memory
RTX 6000 Ada48 GB GDDR6300 WAI and 3D rendering, dual-slot blower card suited to rack chassis
L40S48 GB GDDR6350 WInference and image generation in production, multi-GPU servers
A100 PCIe80 GB HBM2e300 WTraining and fine-tuning of mid-sized models
H100 PCIe80 GB HBM2e350 WTraining and inference of large language models

Example: a server with 2 RTX 4090s (2 × 450 W) and a 170 W processor draws at most 900 + 170 + 150 = 1,220 W. With a 20% margin, that comes to about 1.45 kVA: the GPU Pro plan fits. A single card of this type fits in GPU Starter. NVIDIA reference values; check the datasheet of your exact model, as some custom cards draw more.

GPU colocation, cloud, dedicated server or office?

Four ways to run your AI models, with very different costs, levels of control and flexibility.

GPU colocationCloud GPU rentalRented dedicated server or VPSServer in the office
HardwareYours, paid off over timeRented, never ownedRented from the host, rarely fitted with GPUsYours
VirtualisationNone: physical machine, direct GPU accessOften a virtual machineVPS: shared virtual server; dedicated: noneNone
CostFixed monthly feeBilled by the hour, adds up with continuous useMonthly server rentElectricity and air-conditioning bill
AdministrationFull root access, you manage OS, firewall and backupsThrough the provider's consoleRoot access, managed service depending on the planEverything is on you, on site
PowerRedundant A/B in the data centerHandled by the providerHandled by the hostOne socket, no backup
Network10 Gbps uplink, fixed IP addressesDepends on the planBandwidth depends on the planBusiness internet connection
DataOn your machine, in FranceAt the providerAt the hostOn your premises
Best for24/7 use: inference, APIs, long training runsOne-off needs of a few hoursWorkloads without GPUs, or no hardware to buyPrototyping, a single card

No hardware yet? Dedicated hosting remains an option: a dedicated server saves you the purchase. For a standard server without GPUs, our colocation plans start at 1U and 100 W. Still unsure? Our guide colocation or managed dedicated server compares both approaches, while data center colocation: how it works explains rack units, racks and connection.

Managing your colocated GPU server

ElypseCloud is your physical host, not your managed service provider: the machine stays fully under your control.

System and drivers

You have full root access. Install the distribution of your choice, for example Ubuntu Server or Debian, then the NVIDIA drivers, CUDA and, if you work with containers, Docker with the NVIDIA Container Toolkit.

SSH access and firewall

Key-based SSH login, password disabled, and a firewall (UFW or nftables) that only leaves SSH and your inference API port open. Our first security settings apply as they are.

Storage and backups

Model weights and datasets are best kept on NVMe SSDs. Backups remain your responsibility in colocation: keep a backup copy off the machine for anything you cannot download again.

What the data center provides

High availability of the infrastructure (A/B power, N+1 redundancy), guaranteed bandwidth, fixed IP addresses and IPMI/KVM access on request to take back control remotely.

What you can run in GPU colocation

LLM inference

Serve an open-source model (Llama, Mistral, Qwen…) with vLLM, Ollama or llama.cpp behind your own API, with no per-token billing.

Training and fine-tuning

Adapt a model to your data (LoRA, QLoRA) for days with PyTorch and CUDA, without watching an hourly meter.

Image and video generation

Stable Diffusion, Flux or ComfyUI in production for an application or a creative studio.

3D rendering and compute

Blender or Unreal render farm, simulation, scientific computing: anything that benefits from thousands of CUDA cores.

FAQ - GPU Colocation

Everything about colocating GPU and AI servers at the Equinix PA6 data center.

What is GPU colocation?

GPU colocation (or GPU housing) means installing your own server fitted with graphics cards in a data center rack. You remain the owner of the hardware, the system and the AI models; we provide the rack space, redundant power and the network connection at the Equinix PA6 data center in Paris.

How is it different from standard colocation?

The principle is the same: you bring the machine, we host it. What changes is the sizing. A GPU server draws far more than a standard server (a single RTX 4090 is rated at 450 W), so GPU colocation plans start at 800 W and go up to 3 kVA, whereas standard colocation starts at 100 W.

Which graphics cards can I install?

Any of them, as long as the machine fits in a 19-inch rack chassis and its power draw stays within the plan: GeForce RTX 4090 or 5090, RTX 6000 Ada, L40S, A100, H100 PCIe… For 8-GPU HGX/SXM servers, whose power draw exceeds our standard plans, use the custom plan.

How do I know how much power to choose?

Add up the maximum draw of your GPUs and of the processor, then about 150 W for the motherboard, RAM, disks and fans. Add a 20% margin: that is the power to reserve. The table on this page gives the manufacturer values for the most common GPUs.

What happens if my machine exceeds the plan's power?

Each plan's power is an envelope reserved for you. If your server needs more, we move you to the next plan or extend the power on quote. Just tell us in the form which machine you plan to install, and we check it with you before racking.

Is GPU colocation cheaper than renting a GPU in the cloud?

For continuous use, often yes: cloud rental is billed by the hour, whereas in colocation you pay off your hardware and only pay a fixed monthly fee for the rack, power and network. For a one-off need of a few hours, renting remains simpler. Do the maths on your real volume of GPU hours.

Do my data and models stay in France?

Yes. Your server is installed at the Equinix PA6 data center in Aubervilliers and remains your property: your datasets, model weights and logs do not leave your machine unless you send them elsewhere yourself. ElypseCloud is a brand of ElypseGroup SAS, a French company.

How long does setup take?

Allow 24 business hours to prepare your space once the quote is approved (rack space, network patching, IP addresses), then the reception and connection of your hardware, delivered directly to the data center or dropped off by appointment.

Is there a setup fee?

Yes, €49.99 excl. VAT, waived with a 12-month commitment, as for standard colocation.

What happens next?

Three steps between your request and your GPU server running in our rack.

1

You describe your machine

Pick a plan and list your GPUs in the form. We check the power with you and reply within 24 business hours with a firm quote.

2

We prepare your space

Rack units assigned, A/B power connected, network patching, transit opened and IPs assigned.

3

You deliver, we install

Direct delivery to the data center or drop-off by appointment. Our team racks and cables it, and you take over remotely.

Not sure how much power your setup needs?

Send us your component list and we will tell you which plan to choose before you sign anything.