Kubernetes with GPUs that are actually there.
Managed Kubernetes and Slurm with CPU and GPU node pools in the same cluster. Your data pipelines, APIs and training jobs sit next to each other, on one network and one bill.
A conformant cluster, tuned for AI
Managed control plane
Upstream, CNCF-conformant Kubernetes with high-availability control plane, automatic patching and version upgrades on your schedule.
CPU and GPU node pools
Mix CPU pools with H200, B200, B300 or AMD Instinct pools. Autoscaling per pool, with scale-to-zero for GPU pools.
GPU-aware scheduling
NVIDIA GPU Operator and AMD GPU Operator pre-installed; gang scheduling and team queues with Kueue.
Managed Slurm
Slurm for teams that prefer sbatch, on the same nodes and storage as Kubernetes, so capacity moves between them.
Storage and networking
Block, shared file and S3-compatible object storage; InfiniBand or RoCE for multi-node training pools.
Attach what you already run
EKS, GKE, AKS and OpenShift clusters appear next to Nimbo clusters in Fleet, with the same policies and chargeback.
CPU capacity for everything around the model
Data preparation, vector databases, API gateways and agents need CPUs, not GPUs. Run them next to your GPU pools with no egress between them.
| Family | Size | Per hour | Per month (730 h) |
|---|---|---|---|
| General purpose | 4 vCPU · 16 GB | $0.192 | $140 |
| General purpose | 16 vCPU · 64 GB | $0.768 | $561 |
| Compute optimised | 32 vCPU · 64 GB | $1.280 | $934 |
| Memory optimised | 16 vCPU · 128 GB | $1.024 | $748 |
| Memory optimised | 64 vCPU · 512 GB | $4.096 | $2990 |
Priced at $0.032 per vCPU-hour plus $0.004 per GB-hour of memory, on current-generation AMD EPYC and Intel Xeon hosts. Block storage and public IPs are billed separately.
Pay for nodes, not for the control plane
Control plane
$0.10 per cluster-hour; your first cluster's control plane is free.
CPU nodes
CPU instance rates above, billed per second.
GPU nodes
Nimbo Compute GPU rates, on-demand or reserved.
Create your first cluster.
Spin up a cluster with a CPU pool and a GPU pool in the console, or have an architect size it with you.