Managed Kubernetes.
A conformant, managed control plane with GPU scheduling, autoscaling node pools and zero-downtime upgrades.
A fully managed, upstream-conformant Kubernetes control plane. You keep the portability and ecosystem of Kubernetes; the platform runs the control plane, the upgrades and the GPU scheduling.
- Category
- Compute
- Deployment
- Managed → air-gapped
- Governance
- IAM · encryption · audit
Kubernetes won. Running the control plane and scheduling GPUs didn't get easier.
Kubernetes is the portable substrate everyone standardised on — and that's exactly why you don't want to hand-run its control plane, chase CVEs in it, or hand-roll GPU scheduling for training jobs. Aether runs a managed, upstream-conformant control plane: patched, highly-available and upgraded without downtime, so your team keeps the ecosystem and the portability without owning the parts that page you at night.
GPU scheduling that understands the hardware.
Generic Kubernetes treats a GPU as an opaque resource. Aether's scheduler does fractional and whole-GPU allocation with topology awareness, so a multi-GPU training pod lands on accelerators that share a fast interconnect instead of being scattered across the rack. Autoscaling spans spot and on-demand node pools, so a training burst gets capacity and the cluster shrinks when it's done.
Ecosystem-native, no proprietary lock-in.
Bring your Helm charts, operators and CNI — it's conformant Kubernetes, so your tooling works unchanged and you can leave without a rewrite. The difference from running it yourself is that the model, the data platform and the GPU fabric are on the same cloud, so an ML workload on the cluster reaches the model and the lake without crossing a boundary. It deploys managed or fully air-gapped, like everything else.
A training pod that asks for whole GPUs on a shared interconnect — the scheduler places it on the right hardware, not just any free node.
apiVersion: v1
kind: Pod
metadata: { name: train }
spec:
containers:
- name: trainer
image: registry.aether/trainer:latest
resources:
limits:
aether.dev/gpu: "8"
aether.dev/gpu-topology: "nvlink" # co-located, not scatteredOn generic Kubernetes that last line doesn't exist — you get eight GPUs wherever they're free, and your all-reduce pays for it. Here the scheduler understands the fabric, so the pod lands on accelerators that actually talk to each other fast.
Three steps to running.
A managed, upstream-conformant control plane comes up patched and HA.
Mix spot and on-demand, CPU and GPU pools with autoscaling and topology awareness.
Bring your Helm charts and operators; upgrades are zero-downtime and managed.
Managed Kubernetes, in full.
HA, patched and upgraded for you, with conformance to upstream Kubernetes.
Fractional and whole-GPU scheduling with topology awareness for training jobs.
Cluster and pod autoscaling across spot and on-demand node pools.
Bring your Helm charts, operators and CNI — no proprietary lock-in.
Provision it in a few lines.
Every service is reachable from the same SDK, CLI and infrastructure-as-code — one identity, one bill, one audit trail across the whole catalog.
# Provision managed kubernetes on Aether Cloud
aether compute create \
--service managed-kubernetes \
--name app \
--region us-1 \
--deploy managed # or vpc | on-prem | air-gappedAt a glance.
- Conformance
- Upstream Kubernetes
- Control plane
- Managed, HA, auto-patched
- GPU
- Fractional + whole-GPU scheduling
- Autoscaling
- Cluster + pod, spot pools
- Ecosystem
- Helm · operators · CNI, no lock-in
Built for real work.
Platform teams standardizing on k8s
ML training and serving on GPUs
Portable multi-cloud workloads
On one model, not stitched together.
The usual stack runs managed kubernetes in one product, the model in another and the data in a third — and the seams between them are the cost. Aether Cloud runs it on the same platform that serves the model, governs your identity and deploys into your boundary, with the rest of the catalog one hop away.
No stitching a vector DB to one place, a warehouse to another and a model to a third — managed kubernetes sits next to the rest of the catalog, one identity, one bill.
The provider that runs Aether runs your managed kubernetes — so the data and the model never leave the same governed boundary to talk to each other.
Need a capability that isn’t here yet? The model writes and deploys it into the same boundary — the catalog is a starting point, not a ceiling.
Good to know.
Conformant managed Kubernetes like them — but with first-class GPU scheduling and the model and data platform on the same cloud.
Yes — conformant Kubernetes, so your charts, operators and tooling work unchanged.
Yes, with fractional and whole-GPU scheduling and topology awareness for training jobs.
Your boundary, your choice.
Pairs well with.
On-demand and reserved VMs across CPU and GPU shapes, with per-second billing and live resize.
Latest-generation accelerators by the hour or the cluster, with fast interconnect for training and high-throughput inference.
Event-driven functions that scale to zero — run code and agents without managing a server.
Managed container runtime with autoscaling, health checks and rolling deploys — bring an image, get a URL.
Massively parallel batch and simulation jobs with scheduling, checkpointing and fault tolerance built in.
Dedicated single-tenant hardware for the workloads that need full control of the silicon.
Run Managed Kubernetes on Aether Cloud.
A conformant, managed control plane with GPU scheduling, autoscaling node pools and zero-downtime upgrades. Deployable managed, in your VPC, on-prem or fully air-gapped — talk to us about the configuration your workloads and your boundary require.