OMNI Compute

When your cloud runs out of GPUs, your jobs don’t stop

OMNI Compute extends the Kubernetes cluster you already run onto GPU capacity anywhere: reservations, other regions and clouds, on-prem machines. No second cluster, no new control plane.

Talk to an Engineer
  1. Get answers to your technical questions instantly
GPU capacity
Capacity from other clouds and regions, or your own machines
  • Oracle Cloud
  • AWS
  • Google Cloud
Your own VM or bare metal machineConnected to your cluster
OMNI Compute
Your clusterControl plane stays here
H200 in your region
0 available
Remote H200 node added
Ready
Eligible workloads scheduled on it
Running
Managed through your existing cluster

Trusted by 2,100+ companies

  • Akamai
  • Yotpo
  • Iterable
  • NielsenIQ
  • Swisscom
  • Kakao
  • FICO
  • BMW Group
  • Hugging Face
  • ShareChat
  • WeWork
  • Cisco
  • PlayPlay
  • Roku

The GPUs you need arenā€™t available where you run

Your cluster lives in one region on one cloud. When that region runs out of the GPUs you need, jobs wait, launches slip, and the usual fix is a second cluster, new tooling, and a migration project.

H200 GPUs

Rezolve AI sourced H200 GPUs from Oracle Cloud when Azure ran short, and managed them through its Azure clusters.

Read the story

Capacity from anywhere, managed from your cluster

Add regions from other clouds, machines you already run, or reserved capacity from Cast AI. All of it joins your existing cluster as edge nodes.

Add other regions and clouds to your cluster

Cast AI connects AWS, GCP or OCI regions to your EKS, GKE or AKS cluster as edge locations.

  • Provisions GPU nodes in regions and clouds where hardware is available.
  • Edge nodes join your cluster, show up in kubectl and the console, and are labeled by location.
  • Each edge location is set up with one script, in about 2 to 3 minutes.

Bring GPUs you already have

Custom edge locations add any Linux VM or bare metal machine, on any provider, to your cluster as an edge node.

  • Put GPUs you’ve already paid for to work. Reserved, spare, and on-prem capacity becomes schedulable for the jobs sitting in your queue.
  • Setup is one script per machine. Once connected, the hardware answers to the same scheduler, node templates, and namespace rules as your other nodes.

Get reserved capacity from Cast AI

When your own machines and the other clouds don’t cover it, Cast AI can offer reserved capacity from GPU providers.

  • It joins your cluster as edge nodes: labeled by location, opt-in by namespace, and managed like the rest of your OMNI Compute capacity.
  • Tell us the GPU required and the region. We confirm availability and terms before you sign.

Capacity chosen for every cloud node

When your node template includes cloud edge locations, the autoscaler picks the cheapest matching option, and moves to the next one if that capacity is unavailable.

  • GPU, CPU, memory and other node template constraints still apply.
  • Evictor can evict workloads from edge nodes so they reschedule in your main cluster when capacity frees up.

You decide what leaves your cluster

Nothing runs on an edge node unless you allow it.

  • Label a namespace to opt it in, or pin a workload to a location or provider with a nodeSelector.
  • System namespaces such as kube-system stay in your main cluster.
ā€œI had an aha moment ā€“ an iPhone moment ā€“ with Cast. Literally two minutes into the integration, we saw the cost analytics, and I had an insight into something I had never had before and had tried to get for a very long time.ā€
Dekel Shavit
Dekel ShavitSr. Director of Engineering

Connect, choose, run

  1. Connect

    Enable OMNI Compute on your cluster, then create an edge location with one script. The cluster onboarding takes about 2 minutes, and the edge location setup about 3 minutes.

  2. Choose

    Add edge locations to a node template, then label one namespace to start. Add more when you are ready.

  3. Run

    The autoscaler places nodes on cloud edge locations where it is acceptable, available and cheapest. Custom edge machines join when you add them.

Talk to an Engineer

Works with your existing cluster

Works with EKS, GKE and AKS clusters. Add capacity from AWS, GCP and OCI regions, or connect any Linux machine, including bare metal, from any provider.

01
See where capacity is idle

Six workloads sit across four partially used nodes.

Resources usedResources requested
6 workloads4 active nodes

More built in

Time-slicing

Share a GPU across multiple workloads through node templates, with no extra hardware.

GPU partitioning on edge nodes

Configure MIG on edge nodes where your GPUs support it.

Dynamic Resource Allocation

Fine-tune GPU claims with Kubernetes-native filtering and sharing controls.

Console view by location

Edge nodes carry an ā€œExternal regionā€ label in the node list.

Kubernetes Optimization

Cast AI sets CPU and memory to real usage, then consolidates pods and fits node capacity to match. Keep Cluster Autoscaler or Karpenter, or use Cast AIā€™s own autoscaler.

Cast Radar

See what each cluster costs and how its workloads perform, down to error rate, latency, and OOM kills. No code changes.

Storage Autoscaler

Watch how full each volume gets and expand it in time. Tune IOPS and throughput up with demand, on GKE and EKS.

Take the next step

Read how it works, see what other teams got, or talk to an expert about your setup.

See how it works

Architecture, edge locations, custom machines, and supported environments.

Read the OMNI Compute documentation

See customer results

Rezolve AI sourced H200 GPUs from Oracle Cloud when Azure ran short.

That story and more

Talk to an expert

Walk through your setup with an engineer and see where OMNI Compute fits in your setup.

Book a discovery call

Your questions,
answered

Canā€™t find what youā€™re looking for?

Contact us
Which clouds and regions can I use?

Edge locations can be any AWS, GCP, OCI, Nebius or on-prem region. Your main cluster must be EKS, GKE or AKS. Custom edge locations add any Linux machine, so other providers are covered too.

Which workloads fit?

GPU-intensive, batch and stateless workloads that can tolerate some extra latency, such as batch inference and LLM serving. Training that spans clouds or regions, databases, real-time and stateful apps are not a fit today.

How do I control what runs outside my cluster?

You opt in by namespace or by workload. Workloads you haven’t opted in are never placed on edge nodes.

Can I use GPUs I already reserve or own?

Yes, through custom edge locations. Each machine, in a cloud, a rack or your data center, runs one install script and joins as an edge node. Machines need Linux kernel 5.10 or later, and Ubuntu is recommended.

How do edge nodes connect, and what does it cost me?

Edge workloads run on the edge machine, and traffic between edge and standard nodes goes through a WireGuard VPN. Traffic between clouds and regions is billed by your providers, so egress costs belong in your decision.

Does my data stay in my region?

No. Workloads that run on an edge node run in that region or cloud. Keep anything with residency requirements out of offloaded namespaces.

How do I get started?

Onboard your cluster with the Cast AI agent in two minutes. Then enable OMNI Compute and create your first edge location. Get hands-on onboarding guidance with our OMNI Compute team (24/7 on-call support included), or follow the docs on your own.

#1 on G2262 reviews

Stop waiting for GPUs in your region

Talk to our team and have your first edge location live in minutes.

Contact sales