GPU pool

Connect your GPU pool

Got GPUs in your rack, datacenter or colocation? Connect them to HostYourAI. Run compliant AI on them with our full management layer, or supply capacity to the European pool and earn from hardware that sits idle today.

Early access: we onboard the first pools personally.

For teams with their own hardware

Run your own AI on it

You already have GPUs, on-prem or in colocation. We put the HostYourAI layer on top: vLLM deployments, the model garden, and an OpenAI- and Anthropic-compatible API on your own hardware.

  • Data stays in your own rack, prompts never leave your network
  • API keys per team or client, with a usage dashboard and cost allocation
  • Verified open models from the model garden, warm within minutes
  • You keep the hardware, we handle deploys, monitoring and updates
Discuss your setup
For datacenters and GPU owners

Supply capacity to the pool

Do your GPUs sit idle part of the day? Connect them to the shared Router and earn per GPU-hour consumed. We bring the workloads, the billing and the customers. You bring power, uptime and an EU address.

  • Compensation per GPU-hour actually used, terms agreed per partner
  • EU-only: your hardware is physically in the EU, that is the whole point
  • We vet and benchmark every pool before it goes live
  • No sales channel needed on your side, demand comes from our customers
Register your pool
How it works

From rack to live in four steps

The same path for both tracks. Vetting and benchmarking take the most time, the technical part is an afternoon.

01Intake

Tell us what you run

GPU types, VRAM, network and location. A short email is enough for a first answer.

02Vetting

Location, ownership and benchmark

We verify the hardware is in the EU and yours, and run a benchmark on your machines.

03Connect

Our runtime on your machines

vLLM plus a secured agent, rolled out via Docker. All traffic is encrypted end-to-end.

04Live

Dashboard, usage and settlement

Your pool shows up in your own dashboard or in the Router. From then on you see usage, uptime and the monthly settlement.

Requirements

What your hardware needs

  • NVIDIA GPUs with 24 GB VRAM or more per card (L40S, A100, H100, RTX 4090/5090 or similar)
  • Physically located in the EU, verifiably
  • A stable connection: 1 Gbit/s or more and a fixed IP address
  • Docker and current NVIDIA drivers, or give us access and we will set it up
  • For capacity partners: uptime terms we agree, measure and report together

Not sure your setup qualifies? Email your specs and you will get an honest answer.

FAQ

Frequently asked questions

What does it cost to connect my own pool?

We are in early access and scope this per project: the price of the management layer depends on the number of GPUs and what you want to run on them. There is deliberately no fixed price list yet. Book a call and you will get a concrete proposal.

What do I earn as a capacity partner?

You are compensated per GPU-hour actually consumed by customers. The rate depends on the GPU class and the uptime you can guarantee, and is agreed per partner. Settlement is transparent and monthly.

Does my data stay on my own hardware?

With your own pool: yes. Prompts and responses are processed on your GPUs and never leave your network. We do not log prompt or response content by default, only usage metadata for the dashboard. See our security page for details.

What hardware qualifies?

NVIDIA cards with 24 GB VRAM or more, located in the EU, with a stable connection. Older or smaller cards are not rejected automatically, but the benchmark decides whether they fit the pool. Email your specs if in doubt.

Can I combine both: run my own workloads and rent out spare capacity?

That is exactly what we are building towards: your own workloads first, and idle capacity earning in the pool. During early access we assess per partner what is already possible.

How fast can my pool go live?

The technical part is not the bottleneck: with Docker access the runtime is up in an afternoon. Vetting and benchmarking take the most time. Think days to weeks, not months.

Tell us what is in your rack

One email with your GPU types and location is enough for a first answer, usually within the week.