Compute

Rent capacity by the hour.

GPU and accelerator capacity from operators across the network — hobbyist home labs through full-scale data centers. Tell us what the workload needs and we route it to hardware that fits. Operators list with us because they want the demand; buyers and operators are equally free to find each other on the network directly.

What you can rent

Consumer GPUs

Single consumer-class cards and Apple Silicon. Enough to serve a small open model, run embeddings, or fine-tune with adapters.

Prototyping, small-model serving, batch jobs

Data-center GPUs

Server-class accelerators with the memory bandwidth that mid-size and mixture-of-experts models need to decode quickly.

Production inference, concurrent users, training runs

Multi-GPU nodes

Several accelerators in one machine with fast interconnect, for models that will not fit on a single card.

Large models, sharded serving, distributed training

Confidential compute

Capacity inside a hardware enclave, so the operator of the machine cannot read your data or your weights. Every result carries an attestation.

Regulated data, private weights, sensitive workloads

How it works

Say what you need

Accelerator class, memory, how long, and whether it has to run in an enclave.

We route it

Your requirement is matched against operators with capacity, by price, latency, region, and track record.

Run and pay

A ticket is issued for the term and the balance reserved. Extend it, let it expire, or cancel — you are billed for what it consumed. Pay by card, bank, stablecoin, or credits; the operator earns in TNZO.

Have capacity instead?

If you are running the hardware, the same marketplace works in the other direction.

List idle capacity and get paid for it. We handle the demand, the routing, the billing, and the settlement — you keep the hardware and the margin.

Operator services →

Request capacity Pricing