Skip to main content

GPU capacity on startup terms

Sign for six to twelve months instead of three years. We pool startups into one book that lenders finance, and our router keeps every GPU busy in between.

What you get

Compute without the three-year contract

Contracts that fit a startup

Six to twelve months, with room to grow or shrink at renewal. The lender underwrites the pool, not you, so there is no three-year commitment to sign and no year of capacity to prepay.

Open models on tuned GPUs

DeepSeek, Qwen, GLM, and other open-weight models, served on GPUs we tune for them. Older cards cost less and run them well.

A named operator behind it

The GPUs sit in financed clusters with Valnce named as standby operator. If an operator fails, we run the fleet and your contract keeps running.

Your usage, on the record

Every GPU-minute you use is logged per card. Your bill and the lender's record come from the same source.

How it works

One book, financed once

A lender finances the whole pool. Anchor tenants and startups share the capacity, and our router sells whatever is left.

One lender finances the whole pool
Anchor tenants12 to 36 month contracts
Startups6 to 12 month contracts. That is you.
Our routerEvery unsigned hour, sold as inference
  1. 01

    A lender finances the pool.

    One diversified book of GPUs instead of a loan to each startup. Lenders will not underwrite startups one at a time, so we pool them.

  2. 02

    Anchors and startups sign.

    Anchor tenants take 30 to 50% of the capacity on longer contracts. Startups take the rest on six to twelve month terms.

  3. 03

    The router sells the gaps.

    Hours nobody has signed for are sold as open-model inference, so the pool earns every month and the loan stays covered.

Need GPUs on terms that fit?

Tell us how many you need, which ones, and for how long. We come back with what the pool can offer.

Join the pool

GPUs financed on what they earn.

We keep financed GPUs earning, and show lenders what they make.