Dedicated GPUs from the people who own the racks
GPU capacity by the hour in data centers we build and operate. You rent the machine we racked, with nothing resold and nothing virtualized away. Serverless inference already runs in production on the same infrastructure.
Built for serious compute
Training and fine-tuning
Take an open model and make it yours. Dedicated GPUs hold your training run for as long as you book them, with no preemption and no noisy neighbors.
Self-managed inference
Serve your own stack on hardware you control. Bring your own runtime and weights, and keep the tokens between you and your users.
Batch and research compute
Embeddings, evaluations, renders, and experiments. Book capacity by the hour, run the job, and stop paying the moment it ends.
Why rent from the operator
The operator, not a marketplace
You rent from the company that owns the building, the racks, and the machines. One party is accountable for the entire stack.
One balance for everything
GPU hours bill from the same prepaid credits as inference. One ledger shows every hour and every token.
Capacity we can vouch for
We list SKUs the day the machines are racked and burned in. What you can book is what physically exists.
Indicative pricing
| Class | Typical spec | Price | Good for |
|---|---|---|---|
| Workstation class | 24 GB VRAM per GPU | from $0.40 / GPU-hr | Fine-tuning small models, embeddings, image workloads |
| Datacenter class | 80 GB VRAM per GPU | from $1.50 / GPU-hr | Training and serving large models, multi-GPU jobs |
| Reserved capacity | Whole machines, monthly | Custom | Sustained workloads at committed pricing |
Indicative until launch. Final per-GPU-hour prices are published with the SKUs, and billing uses the same prepaid credits as inference.
How the waitlist works
Tell us your workload
GPU count, VRAM, and how long you need it. One email is enough.
We scope it against the fleet
We match your requirements to machines in commissioning and confirm timing and pricing with you directly.
You get first access
Waitlist position is first come, first served. Your account and credits already work when the hardware goes live.
Get in line
Tell us your workload: GPU count, VRAM, and duration. We contact you as capacity comes online, and waitlist position is first come, first served.