| GPU · instance | VRAM | vCPU | RAM | Fabric | Rate / GPU·hr | vs. hyperscaler |
|---|
* Hyperscaler deltas compare on-demand rates as published by AWS, GCP, Azure as of Q2 2026. Rates include NVLink, local NVMe, and fabric.
Spin up, run, tear down. Billed per second after the first minute. Perfect for experiments, CI, and spiky inference.
Lock capacity for a training run. As short as 24 hours. 18% off on-demand. No quarterly contract.
Deep discount for teams with sustained demand. Dedicated cluster, private fabric, named SRE.
We run our own datacenters, buy GPUs directly, and don't carry the overhead of selling 300 other services. We also don't charge for egress or inter-node traffic, which is how hyperscaler bills balloon.
No. On-demand clusters bill per second after a 60-second minimum. Reserved starts at 24 hours.
Our availability page is live. If a region shows green, it's yours. We overprovision rather than oversubscribe — if we can't fulfill a reservation, it's free.
Yes — 40% off on-demand for verified .edu teams, plus shared Slurm clusters with fair-share scheduling. Email research@neoscale.ai.
Yes. Install our operator; it provisions managed node pools on Neoscale GPUs and streams them into your existing control plane.
Dallas (DFW-01), Reno (RNO-01), Oakland (OAK-01), and Amsterdam (AMS-01). Phoenix and Singapore come online Q3.