_.webp?width=115&height=43&name=Aurora_Logo_Brand(DARK%20THEME)_.webp){kind=logo}
Private cloud compute,
run by the engineers who build it
VMs, managed Kubernetes, and dedicated instances, one control plane.
In-region, your hardware or ours, fully managed.
What is Aurora Compute?
Aurora Compute is private, sovereign cloud compute.
You get three ways to run workloads on one management layer: virtual machines, managed Kubernetes, and dedicated single-tenant instances. You provision, network, and monitor all of them from the same console, with the same IAM, the same private networking, and the same encryption.
It is the raw compute layer of the Aurora AI Cloud. When you need to serve models behind an endpoint, that lives in Aurora Inference. When you need GPU hardware, that is GPU Clusters. When you need data to sit next to your compute, that is Storage. Compute is where your VMs, clusters, and instances run.
Three deploy modes, one management layer, one console.
Runs in-region on your hardware, Aurora's hardware, or a new private build.
Private networking, IAM, and encryption built in.
GPU-aware, integrated with Aurora GPU Clusters
Aurora engineers operate the ops layer under a 99.9% uptime SLA.
How can you deploy compute on Aurora?
Pick per workload, or mix them in the same environment.
Virtual Machines
Flexible vCPU and RAM for general-purpose workloads.
Spin up a VM, size the vCPU and RAM to the job, and reach it over console, SSH, or API. Ubuntu 24.04 by default, other operating systems on request. Public and private IP. IAM is built in from the first instance.
- Flexible vCPU + RAM sizing
- Ubuntu 24.04 default (other OS on request)
- Console, SSH, and API access
- Public and private IP
- Built-in IAM
VM Specifications & Access
- Sizing: choose vCPU and RAM per instance. Custom shapes available on request.
- OS: Ubuntu 24.04 LTS ships by default. Other operating systems available on request. [CONFIRM] supported OS list for early access.
- Access: web console, SSH, and a programmatic API for provisioning and lifecycle.
- Networking: each VM can carry a public and a private IP, and attaches to your VPC.
- Identity: IAM is applied at the instance level, so access follows your roles and policies.
- Storage: attach volumes and pair with Aurora Storage for datasets and checkpoints.
Managed Kubernetes
Managed provisioning and lifecycle, with GPU-aware scheduling.
Aurora provisions and manages the control plane and node lifecycle, so your team ships workloads and skips the cluster upkeep. Pods schedule with GPU awareness. Clusters scale horizontally and vertically. Platform teams that want the raw Kubernetes API get a handoff for it.
- Managed provisioning and lifecycle
- GPU-aware pod scheduling
- Horizontal + vertical autoscaling
- Kubernetes API handoff available
Managed Kubernetes Details
- Managed lifecycle: provisioning, upgrades, and node management are operated by Aurora.
- GPU-aware scheduling: pods requesting GPUs land on GPU-capable nodes, integrated with Aurora GPU Clusters (B200, B300, GB300).
- Autoscaling: scale out with more nodes (horizontal) and scale up node sizing (vertical).
- API handoff: teams that want direct control get access to the Kubernetes API. [CONFIRM] scope of the handoff (namespaced vs cluster-admin) for early access.
- Networking: clusters run inside your VPC with firewall and security groups applied.
- To serve models on Kubernetes: run the workload here, and point endpoints at Aurora Inference for managed serving.
Dedicated Instances
Single-tenant hardware for isolation, compliance, and dedicated GPUs.
When shared infrastructure will not do, run on single-tenant hardware. You get performance isolation, a clean compliance separation, and dedicated GPU access. Sizing is custom to the workload.
- Single-tenant hardware
- Performance isolation
- Compliance separation
- Dedicated GPU access
- Custom sizing on request
Dedicated Instance Details
- Single-tenant: hardware is not shared with other tenants, so there is no noisy-neighbor contention.
- Performance isolation: consistent performance for latency- or throughput-sensitive workloads.
- Compliance separation: physical separation supports stricter compliance and data-handling requirements.
- Dedicated GPU: direct access to dedicated GPUs, integrated with GPU Clusters (B200, B300, GB300).
- Sizing: custom configurations on request. Tell us the workload and we size the hardware.
Where does Aurora Compute run?
Run the Aurora platform on hardware you already own or procure. You keep physical control of the machines; Aurora provides the management layer and operates the ops layer on top. Good for teams with existing data-center footprint or hardware mandates.
Run on Aurora's hardware. No procurement, no rack-and-stack. You provision VMs, Kubernetes, and dedicated instances through the console, and Aurora runs the underlying infrastructure. Good for teams that want private and/or sovereign compute without owning the metal.
A new private build stood up for you. Aurora designs and deploys dedicated infrastructure to your requirements, then operates it. Good for teams that need a purpose-built, in-region private cloud from the ground up.
What comes with Aurora Compute?
Private networking
Every environment gets a VPC with firewall rules and security groups. Public and private IP where you need them, private-only where you do not.
Encryption
Data is encrypted in transit and at rest. Bring your own KMS to hold your own keys.
A real console
The dashboard is not a mockup. Manage VMs, networks, and volumes, handle credentials, and read the event log from one place.
Operated for you
Aurora runs the ops layer under a 99.9% uptime SLA, deployed in-region. You run workloads; Aurora keeps the layer beneath them healthy.
- VPC: isolated virtual private cloud per environment.
- Firewall + security groups: control ingress and egress at the network and instance level.
- Encryption in transit: traffic between and into services is encrypted.
- Encryption at rest: stored data is encrypted; bring-your-own-KMS supported for customer-held keys.
- IAM: identity and access management built into every deploy mode.
- Console surfaces: dashboard, VMs, networks, volumes, credentials, event log.
Where Aurora Compute Fits
- Need to serve a model behind an endpoint? That is Aurora Inference, serverless and dedicated inference endpoints, and agents.
- Need GPU hardware? That is GPU Clusters, B200, B300, and GB300, integrated with Compute.
- Need data next to your compute? That is Aurora Storage.
- Want the whole platform? Start at the Aurora AI Cloud hub.
Frequently-Asked Questions
Compute runs your VMs, Kubernetes clusters, and dedicated instances. Inference serves models behind endpoints and runs agents. Use Compute to run workloads; use Inference to serve models.
Yes. Aurora manages provisioning and lifecycle, schedules pods with GPU awareness, autoscales horizontally and vertically, and offers a Kubernetes API handoff.
Yes. Data is encrypted in transit and at rest, and bring-your-own-KMS is supported.
