Revenue share
We sell open-model inference on GPU hours that would sit idle and share what they earn with the operator and the lender.
Turn idle GPU time into inference revenue
Financed GPUs sit idle between jobs. We run open-model inference on those hours, log what each GPU earns, and share the revenue with the operator and the lender.
What you get
Start with your idle time
A record you can see
Every GPU-minute is logged per card, so owners and lenders see what idle hours produced. Over time, it is the earnings history a loan can be underwritten on.
Operators stay in control
The operator can take GPUs back whenever they need them. Their own contracts always come first.
Older GPUs keep earning
We tune models and serving software so cards like the H100 keep earning, which helps protect the value of the collateral.
One steady endpoint
When GPUs return to their owner, requests move without dropping. Customers see one endpoint.
Capacity for startups
Startups and smaller clouds sign with us for six to twelve months instead of prepaying a year for capacity. The same router keeps their endpoint steady.
How it works
One fleet from many operators’ idle hours
- 01
Pool
Idle hours from many operators combine into one fleet large enough for production traffic.
- 02
Route
Our proprietary router moves requests across the pool and keeps one steady endpoint when an owner takes GPUs back.
- 03
Price
Each job runs on the best available GPU configuration for its price.
Terms
Agree on the terms before we run anything
Every operator sets their own limits. We write them down together, then our engine works inside them.