VMware vSphere / VCFGA
Discover GPU hosts, carve VMs, keep vMotion and existing tooling
Nimbo Fleet
Every GPU you run, in one control plane.
Fleet connects your VMware, Proxmox and OpenShift clusters and your EKS, GKE and AKS accounts to Nimbo capacity. Teams get self-service GPUs; platform teams get one set of quotas, policies and costs.
What Fleet does
Built for the platform team that has to say yes to every AI project
Self-service GPUs
Data scientists request notebooks, training jobs, Slurm clusters or inference endpoints from a catalog. The right hardware is chosen by policy.
Quotas and fair share
Guaranteed and burstable quotas per team, preemption rules, and fractional GPUs where the hardware supports it.
Burst to Nimbo
When on-prem queues back up, jobs that policy allows run on reserved Nimbo capacity instead of waiting.
Chargeback
Cost per team, project and model across your hardware, cloud accounts and Nimbo, in one report.
Multi-tenancy
Hard isolation between tenants with dedicated nodes and network segmentation; soft isolation inside a tenant.
Day-2 operations
GPU health, driver and firmware updates, and cluster upgrades run as audited workflows with approvals.
Integrations
Keep VMware. Add Proxmox. Burst to Nimbo.
Fleet manages hypervisors and Kubernetes side by side, so a VMware exit can happen at your pace.
Proxmox VEGA
Clusters, GPU passthrough and vGPU, the landing zone after VMwareRed Hat OpenShift / OpenShift AIGA
Operator-based install, GPU Operator, llm-d and vLLM servingAmazon EKSGA
Attach existing clusters, schedule into your own GPU node groupsGoogle GKEGA
Attach clusters; schedule TPU and GPU node pools in your GCP accountAzure AKSGA
Attach clusters, quota and chargeback across subscriptionsNVIDIA Hopper and BlackwellGA
H200, B200 and B300 on Nimbo capacity; MIG and full-node tenancyAMD InstinctGA
MI300X and MI355X on Nimbo capacity, ROCm images maintainedGoogle Cloud TPUPreview
Orchestrated in your own GCP project through GKE; billed by GoogleNimbo RackGA
Managed on-prem GPU racks that join Fleet on day oneKubernetes and Slurm as a serviceGA
Managed clusters with CPU and GPU node poolsGAgenerally availablePreviewavailable to design partnersRoadmapannounced, not yet available
Deployment
Runs where your policy says it must
Use Fleet as a managed service, or run the control plane in your own data center, including air-gapped sites.
- ManagedNimbo operates the control plane; agents in your clusters connect outbound only.
- Self-hostedThe full control plane on your hardware, updated through signed releases.
- Air-gappedFor sovereign and defense workloads: no outbound connection, offline updates. Available on request.
Map your estate in one session.
Bring an inventory of clusters and GPUs. We will show what Fleet would schedule where, and what bursting would cost.