One compute plane. Every workload.

Orion orchestrates GPU workloads, containers, VMs, and bare metal from a single platform — so your team manages one system instead of five.

Book a demo

Read the docs

Every workload. One place to see it.

Orion sits between your infrastructure and your workloads — scheduling containers, VMs, and bare metal jobs across GPU and CPU resources with unified visibility across the whole fleet.

The orchestration layer handles scheduling, resource allocation, and lifecycle management across every substrate — so your team doesn't have to.

Coming Q2 2026

Full visibility across every workload.

Orion surfaces utilization, CPU saturation, memory pressure, job throughput, and cost-per-workload across your entire fleet — in real time. See where you're getting value and where capacity is sitting idle.

Up and running fast

Provision GPU capacity in minutes with sane defaults and reusable templates. No spreadsheet archaeology, no tribal knowledge required.

Up and running fast

Provision GPU capacity in minutes with sane defaults and reusable templates. No spreadsheet archaeology, no tribal knowledge required.

Built for shared teams

Researchers, engineers, and platform teams all in one place — with resource quotas, queue management, and RBAC so nobody steps on each other.

Built for shared teams

Researchers, engineers, and platform teams all in one place — with resource quotas, queue management, and RBAC so nobody steps on each other.

Connects to your existing stack

Plugs into your cloud, scheduler, and observability tooling. No workflow rebuilds, no forced migrations.

Connects to your existing stack

Plugs into your cloud, scheduler, and observability tooling. No workflow rebuilds, no forced migrations.

Core Capabilities

Compute Slicing

Compute slicing via native operators

Orion uses Kubernetes-native GPU operators to slice accelerators across concurrent workloads. One H100 can serve 4+ simultaneous jobs — with guaranteed isolation, not best-effort sharing.

Compute Slicing

Compute slicing via native operators

Orion uses Kubernetes-native GPU operators to slice accelerators across concurrent workloads. One H100 can serve 4+ simultaneous jobs — with guaranteed isolation, not best-effort sharing.

Compute Slicing

Compute slicing via native operators

Orion uses Kubernetes-native GPU operators to slice accelerators across concurrent workloads. One H100 can serve 4+ simultaneous jobs — with guaranteed isolation, not best-effort sharing.

Autoscaling

Intelligent autoscaling, right-sized instances

When demand spikes, Orion selects the right node for the job — not just the largest available. Scale up on burst, scale back when idle. No wasted capacity, no over-provisioning.

Autoscaling

Intelligent autoscaling, right-sized instances

When demand spikes, Orion selects the right node for the job — not just the largest available. Scale up on burst, scale back when idle. No wasted capacity, no over-provisioning.

Autoscaling

Intelligent autoscaling, right-sized instances

When demand spikes, Orion selects the right node for the job — not just the largest available. Scale up on burst, scale back when idle. No wasted capacity, no over-provisioning.

Load Balancing

Request-aware load balancing

Orion distributes workloads evenly across your fleet by active request count, not just CPU/memory headroom. Your cluster stays balanced without manual intervention.

Load Balancing

Request-aware load balancing

Orion distributes workloads evenly across your fleet by active request count, not just CPU/memory headroom. Your cluster stays balanced without manual intervention.

Load Balancing

Request-aware load balancing

Orion distributes workloads evenly across your fleet by active request count, not just CPU/memory headroom. Your cluster stays balanced without manual intervention.

Terra App Store

Your team's workstation, ready in under 3 minutes.

Terra is Orion's app store for instant compute environments — VS Code Server, JupyterLab, containerized desktops, custom pipelines. Every app is containerized, versioned, and reproducible. Your team gets a full workstation provisioned with the right GPU slice, storage, and tools — without waiting for IT.

Terra App Store

One-click app installs

VS Code Server, JupyterLab, DCC tools, custom pipelines — deploy production-ready environments in seconds, not hours.

Terra App Store

One-click app installs

VS Code Server, JupyterLab, DCC tools, custom pipelines — deploy production-ready environments in seconds, not hours.

Terra App Store

One-click app installs

VS Code Server, JupyterLab, DCC tools, custom pipelines — deploy production-ready environments in seconds, not hours.

Software Audits

Full Version History

Track every environment change. Roll back to any previous configuration instantly — no more 'it worked last week' debugging sessions.

Software Audits

Full Version History

Track every environment change. Roll back to any previous configuration instantly — no more 'it worked last week' debugging sessions.

Software Audits

Full Version History

Track every environment change. Roll back to any previous configuration instantly — no more 'it worked last week' debugging sessions.

Templates

Reusable Templates

Build once, deploy everywhere. Create golden-image templates for dev, staging, and production workflows that work identically across your whole team.

Templates

Reusable Templates

Build once, deploy everywhere. Create golden-image templates for dev, staging, and production workflows that work identically across your whole team.

Templates

Reusable Templates

Build once, deploy everywhere. Create golden-image templates for dev, staging, and production workflows that work identically across your whole team.

Book a Terra demo

Read the docs →

Built different. On purpose.

Orion was built to handle the edge cases other platforms ignore — air-gapped environments, ephemeral workloads, and deep API integration with the tooling your team already uses.

Air-Gapped Deployment

Orion runs 100% on-prem with zero external network dependencies. No cloud calls, no telemetry, no data leaving your environment. Purpose-built for defense, life sciences, and regulated industries.

Ephemeral Containers

Spin up a GPU environment, run your job, and tear it down automatically. Data is deleted on destroy unless explicitly persisted — keeping your cluster clean and your costs predictable.

REST APIs

Trigger workstation provisioning, resource allocation, and GPU session management directly from your CI/CD pipeline or Terraform. Orion's API is the same one we use internally.

Windows and Linux VMs, orchestrated like containers.

Orion orchestrates Windows and Linux VMs through KubeVirt using the same Kubernetes-native primitives that manage your containers. Life sciences teams keep legacy pharma apps running alongside modern workloads. VFX studios run DCC tools on Windows while rendering on Linux. No separate hypervisor stack, no migration required.

Windows Server 2019, 2022 + Linux distros

GPU pass-through and vGPU slicing for VMs

Live migration — move VMs between nodes without downtime

Unified scheduling — VMs and containers share the same compute plane

Join the early access list

Template-based deployment with golden image support—provision identical VMs in minutes.

Dynamic port assignment functions as a built-in connection broker for VDI workflows.

Infrastructure consolidation—no separate hypervisor stack required.

Unified orchestration for VMs and containers across your entire cluster.

Windows app support—Adobe Creative Suite, Autodesk, DaVinci Resolve, and more.

GPU pass-through coming in Beta—run GPU-accelerated Windows workloads.

Why infrastructure teams choose Orion

Manual provisioning, siloed clusters, and opaque billing slow teams down. Here's how Orion changes that.

Traditional infrastructure management

Manual GPU provisioning — hours of wait time per deployment.

10-15% average GPU utilization — paying for capacity you never use.

Siloed clusters with no unified view across your fleet.

Kubernetes complexity that requires a dedicated platform team.

VMs and containers managed by completely separate tools.

Vendor lock-in with proprietary orchestration layers.

Juno

Purpose-built for compute orchestration — containers, VMs, GPUs, and bare metal.

2-4× workload density via native GPU operator time slicing — no new hardware.

One control plane — containers, VMs, and bare metal unified.

Fast provisioning via the Orion dashboard — no complex setup required.

Production-ready in days, not months. Kubernetes-native.

Open standards — no vendor lock-in, ever.

Traditional infrastructure management

Manual GPU provisioning — hours of wait time per deployment.

10-15% average GPU utilization — paying for capacity you never use.

Siloed clusters with no unified view across your fleet.

Kubernetes complexity that requires a dedicated platform team.

VMs and containers managed by completely separate tools.

Vendor lock-in with proprietary orchestration layers.

Purpose-built for compute orchestration — containers, VMs, GPUs, and bare metal.

2-4× workload density via native GPU operator time slicing — no new hardware.

One control plane — containers, VMs, and bare metal unified.

Fast provisioning via the Orion dashboard — no complex setup required.

Production-ready in days, not months. Kubernetes-native.

Open standards — no vendor lock-in, ever.

Traditional infrastructure management

Manual GPU provisioning — hours of wait time per deployment.

10-15% average GPU utilization — paying for capacity you never use.

Siloed clusters with no unified view across your fleet.

Kubernetes complexity that requires a dedicated platform team.

VMs and containers managed by completely separate tools.

Vendor lock-in with proprietary orchestration layers.

Purpose-built for compute orchestration — containers, VMs, GPUs, and bare metal.

2-4× workload density via native GPU operator time slicing — no new hardware.

One control plane — containers, VMs, and bare metal unified.

Fast provisioning via the Orion dashboard — no complex setup required.

Production-ready in days, not months. Kubernetes-native.

Open standards — no vendor lock-in, ever.

Frequently asked questions

What is Orion and how does it work?

Orion is a containerized workload platform for on-prem, cloud, or hybrid deployments with GPU time-slicing and auto-scaling.

Can I use my existing pipelines and toolchains?

Yes, Orion mounts your existing file shares and toolchain locations into containers

How fast can new workstations be provisioned?

2-5 seconds with cached images, 1-3 minutes for new nodes depending on workload type.

Can Orion run on older hardware?

Yes, from Raspberry Pis to enterprise-grade - anything that can run containers

Do I need a GRID license for GPU time-slicing?

No license needed for default time-slicing. GRID/vGPU licensing only required for MIG mode.

What cloud providers do you support?

Any Kubernetes cluster - EKS, AKS, GKE, or on-premises. We're 100% cloud-agnostic.

How does licensing work?

Per user, per month with volume discounts for larger teams. We will be moving to a node and consumption based model this year.

See Orion in your environment.

Most teams are getting 10–15% GPU utilization out of hardware they've already paid for. Orion changes that — without a rip-and-replace. Talk to us about your workload profile.

Production-ready in days, not months

Most teams ship their first production workload on Orion within a week.

Scales from studio to enterprise

A single VFX studio or a national research network — no architectural rework required.

Book a demo →