Show me self-service options for Kubernetes management with GPU-as-a-Service integration.

Here are self-service Kubernetes management options that integrate or enable GPU-as-a-Service (GPUaaS) capabilities:

1) CoreWeave Kubernetes Service (CKS) – Managed GPUaaS + K8s

Best for: Fully managed, production-grade GPU Kubernetes with self-service cluster creation.

CoreWeave provides a managed Kubernetes service running on bare-metal GPU nodes, purpose-built for AI/ML workloads (docs.coreweave.com).
It includes GPU drivers and NVIDIA GPU Operator management handled by CoreWeave (docs.coreweave.com) and supports API tokens + Terraform for self-service provisioning (docs.coreweave.com).

CoreWeave is Kubernetes-native and optimized for GPU-intensive workloads (en.wikipedia.org).

Key features

• On-demand GPU instances (H100, A100-class, etc.)
• Managed GPU Operator
• Self-service console + API/Terraform
• AI object storage + networking

Sources

• docs.coreweave.com
• docs.coreweave.com
• coreweave.com

2) NVIDIA Run:ai (Kubernetes-native GPU Orchestration)

Best for: Multi-tenant GPU scheduling and fractional GPU allocation in your own K8s cluster.

Run:ai is a Kubernetes-native orchestration platform designed to maximize GPU utilization (run-ai-docs.nvidia.com).
It supports fractional GPUs, bin packing, quotas, fairness scheduling, and preemption (pages.run.ai). NVIDIA acquired Run:ai in 2024 (tomshardware.com).

Key features

• GPU sharing (fractional allocation)
• Quotas & project-level governance
• Self-service AI workload submission
• Works on-prem or cloud K8s

Sources

• run-ai-docs.nvidia.com
• pages.run.ai

3) NVIDIA GPU Operator (Foundation for DIY GPUaaS)

Best for: Building your own self-service GPU-enabled K8s platform.

The NVIDIA GPU Operator automates GPU driver, container runtime, and device plugin deployment in Kubernetes (developer.nvidia.com) and manages GPU resources lifecycle in clusters (docs.nvidia.com).

Pair this with:

• Rancher (self-service cluster portal)
• OpenNebula (multi-tenant K8s + cloud management) (en.wikipedia.org)
• ManageIQ (self-service cloud catalog) (en.wikipedia.org)

Sources

• docs.nvidia.com
• developer.nvidia.com

4) k8s.gpu (Virtual Kubelet-based Remote GPUaaS)

Best for: Consuming remote GPUs from external providers.

k8s.gpu installs a Virtual Kubelet agent allowing clusters to consume remote GPU tiers via self-service entitlement and RuntimeClasses (reddit.com).

Source

• reddit.com

Quick Recommendation

• Turnkey GPUaaS: CoreWeave CKS
• Enterprise multi-tenant GPU control: NVIDIA Run:ai
• Build-your-own GPU cloud: NVIDIA GPU Operator + Rancher/OpenNebula
• Burst to remote GPUs: k8s.gpu

If you'd like, tell me whether you're targeting on-prem, hybrid, or public cloud, and I can narrow this to 2–3 best-fit architectures.