Our Mission
About Ratel
We rent dedicated GPUs inside the EU, with the model already on them. This page is what that means and where its limits are.
The sovereignty thesis
European enterprises are navigating a perfect storm: the EU AI Act's rules arrive in stages through 2028, GDPR enforcement is intensifying, and CLOUD Act concerns about US providers are mounting. Yet the AI infrastructure landscape forces a false choice between raw EU clouds (complex, engineer-only) and US-hosted AI services (convenient but jurisdictionally risky).
Ratel exists at this intersection. We rent dedicated GPUs in EU regions, with open-weight models whose serving arguments are already chosen. The GPU is pinned to an EU region, and the control plane is not in the inference path — so your prompts reach your machine and nothing else.
The problem we solve
The compliance trap
Using US-hosted AI services creates jurisdictional risk. Your data may be subject to CLOUD Act requests, and GDPR compliance becomes complex when data crosses borders. European enterprises need EU-only data paths.
The complexity wall
Most European cloud providers sell raw infrastructure — powerful, but you must be an engineer to configure models, networking, and security. Most teams don't have that expertise.
The budget uncertainty
Per-token pricing charges most exactly when a system is working hardest. A model that answers all day costs more every month it succeeds, which is a strange thing to budget against.
The deployment delay
Enterprise AI vendors require sales cycles, contracts, and custom deployments. When you need a model running today, a three-month procurement process is a non-starter.
Our approach
Defaults over options
Quantization, context length and tensor parallelism are chosen before the build reaches the catalog. Every one of them can be overridden, and a deployment that overrides none of them still serves.
Sovereign by default
The GPU is pinned to an EU region, the data path stays inside it, and every subprocessor is listed by name with what it processes.
Flat-rate, transparent
No per-token fees and no rate limits — you pay for the hours the card is powered on. Scheduled stop times and a monthly budget cap are there to stop it running when you are not using it.
No procurement cycle
No contract to negotiate and no deployment to schedule. Once you have an invite, a GPU is a form and a few minutes, not a quarter.
Private by design
Your prompts go to your GPU and stop there. The control plane is not in the inference path, so there is no point at which we could read one — and you hold root on the box, so you can check.
Honest positioning
A dedicated GPU is the wrong shape for bursty, low-volume work — a per-token API is cheaper below roughly a few hours of daily use. We say so on the pricing page rather than only here.
Why this, why now?
The EU AI Act's rules arrive in stages through 2028, GDPR enforcement is tightening, and the CLOUD Act still reaches US-owned providers wherever the servers sit. Those are the reasons a European team looks for a European GPU, and none of them is a trend that reverses.
The gap we build in is between raw infrastructure and a contract. European clouds rent you a bare machine and leave the model serving to you. Enterprise AI vendors sell a procurement cycle. We sell an hour of a GPU with a model already on it, at a price published before you deploy.
We are small, and we are not trying to outspend anybody. What we can do is keep the product narrow: one GPU, one model, one endpoint, in a region you choose — for a team that needs the data to stay in Europe and does not want to run the infrastructure.
Our commitments
What EU residency means here
We're not "EU-aligned" while hosting in Virginia. EU-region pinned infrastructure, EU-only data path. We describe our posture honestly — we're not SOC 2 or ISO 27001 certified, and we never imply it.
What reaches our logs
This is a hard guardrail, not a bolted-on feature. We architect around it. Your prompts, responses, and model outputs never touch our logs or observability systems.
What a price includes
No hidden fees. No per-token surprises. We show you the GPU-hour cost, the projected daily/monthly cost, and what you're actually paying for. Indicative pricing until our capacity deal lands — never firm final promises we can't honor.
What we are not building
We don't try to be everything. We would rather run a short catalog that works than a long one that mostly does.
Ready to deploy?
Anyone can sign up. Our team turns on GPU access for each account, so tell us what you need to run.