FAQ
Is the hardware real?
Yes. C1 is a rack we own, in a colocation facility in Hillsboro, Oregon. It has 8 NVIDIA H100 SXM 80 GB GPUs on one NVLink fabric. The spec is on the clusters page. The telemetry on the site comes from that rack once the preview period ends. During preview, the site shows simulated telemetry, marked with the word preview wherever it appears.
What does preview mean?
Preview means the console and API are open by invitation only. Public endpoints answer with HTTP 503 and the message Console access is invite only during preview. Pricing, regions and cluster specs on this site are final for C1 unless a capacity report says otherwise.
How do I get access?
Fill in the request access form. Tell us your workload, your team size and how many hours a month you expect to run. We reply within two business days. Invitations go out in the order that fits available capacity, not strictly first come first served.
What can I run?
Anything that runs on Linux with CUDA 12. Training runs, fine tuning, inference servers, batch evaluation, data processing that wants a GPU. You bring a container image. We run it on the GPUs you asked for and stream telemetry back.
Can I get the whole rack?
Yes, with a reserved plan of 1 month or more. See reserved capacity. On demand and batch plans allocate individual GPUs.
What happens to my data?
Local NVMe is wiped when your job ends. Shared storage persists for the length of your reservation plus 7 days. We do not read your data and we do not train on it. Details are in the SLA.
Do you have a second region?
Not yet. eu-central in Frankfurt is next, listed once C1 is sold out and the new rack has completed soak testing. See regions.
How do I pay?
Monthly invoice, card or bank transfer, in USD. Reserved capacity is invoiced at the start of the term. On demand and batch are invoiced at the end of each month against metered minutes.
Is there an SLA?
Yes. 99.5 % monthly availability for reserved capacity, credits on breach. The SLA page has the terms.
Who is behind this?
A small infrastructure team that has run GPU fleets before and got tired of not knowing what an hour cost. Use the request access form and a person answers.