// SAIG COMPUTE · EXPLAINED
How it works
GPU compute for data scientists, AI engineers and public-sector teams — built around one simple idea: compute and data are separate. The machine is disposable. Your data survives it.
Compute and data are different things
A lot of confusion comes from treating the GPU machine and the data as one unit. They're not. The compute module is ephemeral — it exists to run a workload and nothing more. Your datasets, checkpoints and outputs live in persistent object storage that outlives any machine.
Compute module
An ephemeral GPU machine in a European region. Billed per hour while it's running. Tear it down at any time — billing stops the moment you do.
disposable · €/hourData layer
S3-compatible object storage in the EU. Your bucket exists independently of any machine. Egress is free — your data is never held hostage.
persistent · €/GB/month · egress freeThe workspace lifecycle
A workspace goes through four stages. You control each transition.
1 · Provision
Pick a GPU class, a European region, a base image and how you'll connect (Jupyter, SSH, API endpoint, WireGuard private network). The wizard walks you through cost before you pay anything.
2 · Run
The machine is live. You connect via whichever access method you chose. Credit burns from your prepaid balance at the published GPU-hour rate.
3 · Pause
Pause a machine when you're not using it. Billing stops. Your S3 data is unaffected. Resume any time.
4 · Tear down
Destroy the compute module completely when you're done. Your S3 bucket and any attached persistent volumes remain intact. Nothing you care about is lost.
How you connect
Choose one or more access methods at workspace creation. You can enable all four at once.
Jupyter notebook
JupyterA browser-based GPU-backed notebook. No local install required. Ideal for exploration, data analysis and one-off experiments.
SSH
SSHFull terminal access. Run any tool, script or framework. Forward ports to reach other services running on the machine.
Inference API
APIAn OpenAI-compatible HTTP endpoint (served by vLLM or TGI). Point any existing client or pipeline at it. Bearer token auth.
WireGuard private network
WireGuardA private encrypted tunnel between the GPU machine and your other servers or CI runners. No public port exposure required.
Prepaid credit and the hard stop
There are no surprise bills on SAIG Compute. You top up a prepaid GPU credit balance before you launch. The hourly GPU rate burns from that balance. When it reaches €0, the machine pauses automatically — it does not continue running and bill you retroactively.
Monthly plan
A flat monthly fee that covers your workspace entitlement and support tier. You pay this regardless of whether the machine is running.
GPU credit
A prepaid balance. Consumed per GPU-hour while the machine runs. Top up any time. Unused balance carries forward.
Hard stop at €0
No runaway meter. No invoice surprise. The machine pauses; your data is safe. Add more credit to resume.
EU-only data residency
Sovereign AI Grid is operated by Nixpay B.V., a 100% Dutch-owned company. All workloads run on European hardware under EU jurisdiction. Your data never leaves the European Union — not for processing, not for backups, not for anything. We are GDPR-compliant by design, not by policy checkbox.