Neocloud software for GPU cloud operators

RunOS is infrastructure software for hosting companies and GPU cloud operators. Use your own servers to offer bare metal, GPU virtual machines, and inference endpoints. RunOS manages the infrastructure and exports GPU usage. You set prices and handle invoicing.

What can you offer?

Offer	What RunOS does
Dedicated bare metal	Allocate a physical server from your provider account to a tenant account. The tenant can use the published golden images, power controls, and serial console on its allocated server.
On-demand and spot GPU VMs	Offer on-demand GPU VMs or interruptible spot capacity. VM groups have quotas and default-deny firewalls.
Model inference	Serve a model from a pool of GPU VMs behind an API key. Set the minimum and maximum number of replicas.

RunOS uses standard Kubernetes and KubeVirt underneath. For the physical server workflow, see data center operations. For allocation rules and spot notices, see GPU capacity.

How does billing work?

RunOS records the time that each metered workload holds GPU capacity. The usage export gives your billing system usage records rather than charges. Apply your own prices, discounts, and billing rules to those records.

Import the usage export by sequence. Update the existing record when the same interval appears again with a higher sequence. A workload that starts and ends between meter passes can be absent from the export. Test that behavior against your billing requirements during the pilot.

Should you build the control plane yourself?

Use RunOS if you own or control the fleet and already have a billing system you want to keep. Compare it with assembling provisioning, virtualization, and usage collection yourself. That gives you more control over each layer, but your team owns the integrations and upkeep. If you need a bundled store, payments, and invoicing, evaluate platforms that provide those pieces together.

What should you test first?

1. Allocate one physical server to a pilot tenant. Verify which controls the tenant can use from its account.
2. Run an on-demand GPU VM and an inference endpoint. Add a spot VM if interruptible capacity is part of your offer.
3. Import the usage export into your billing system. Reconcile the records against the capacity each workload held.

RunOS does not issue invoices. Confirm how you will handle sign-up, payment, and support before you launch. For infrastructure requirements and access controls, see the infrastructure overview and release notes.