NVIDIA NVKE for Self-Hosting AI Models: What It Gives You on Bare-Metal GPUs and What You Still Build
Blog post from Qovery
NVIDIA released NVKE and its companion build system, NVKR, as Apache 2.0 public-preview projects on November 10, 2026, offering a management plane for creating isolated Kubernetes clusters on NVIDIA GPU infrastructure. Using NVIDIA Infra Controller (NICo), NVKE provisions workload clusters with separate control planes and etcd, node attestation, and lifecycle features such as upgrades and node replacement, but NVIDIA does not recommend it for production yet; the preview also lacks management-cluster upgrade and backup procedures, and its API requires operators to enable authentication and secure communications. NVKE creates clusters rather than a complete model-serving platform, so teams must still provide GPU runtime components, serving engines such as vLLM or NVIDIA NIM, inference routing and autoscaling, model caching, and observability backends. The text presents managed EKS or GKE GPU node pools as a simpler production option for teams renting cloud GPUs, while NVKE may suit organizations with GPU racks, NICo deployments, and a need to offer multiple isolated clusters. It also recommends evaluating NVKE in a lab while keeping serving deployments portable, and notes that Qovery can provide an application-delivery layer for existing NVKE or cloud clusters without replacing their infrastructure or GPU runtime.
| Trend | Post Mentions | Total Month Mentions | Posts | Companies | MoM |
|---|---|---|---|---|---|
| Kubernetes | 39 | 4,377 | 419 | 114 | -15% |
| MCP | 10 | 9,757 | 959 | 203 | -18% |
| LLM | 7 | 3,674 | 823 | 199 | -44% |
| AI Agents | 4 | 6,233 | 1,230 | 245 | -11% |
| Observability | 4 | 2,390 | 523 | 156 | -23% |
| OpenTelemetry | 2 | 822 | 142 | 56 | -4% |
| Platform Engineering | 2 | 1,295 | 290 | 84 | -36% |
| Secrets Management | 2 | 2,876 | 539 | 148 | -7% |
Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.