Upstream Kubernetes: On-Prem Kubernetes Ledger 2026
Upstream Kubernetes is the reference line for every other entry: three releases a year on schedule, public enhancement proposals, and nothing to buy.
Part of the On-Prem Kubernetes Ledger (edition 2026-10-13). How the scores work: methodology.
1. Snapshot
- Tier: Open-source engine
- Owner: Cloud Native Computing Foundation / Linux Foundation (Graduated project, open governance by SIGs)
- Licence model: Apache-2.0; nobody sells it; no commercial terms
- Products scored: Kubernetes 1.37, kubeadm, Cluster API v1.14 with the kubeadm bootstrap and control-plane providers
- Latest release: Kubernetes 1.37.0 on 2026-08-26 (patch 1.37.1 on 2026-09-15); Cluster API v1.14.0 on 2026-08-11
- Next signal: Kubernetes 1.38.0 proposed for 2026-12-16 (SIG Release schedule)
- Archetype: Reference engine
- Evidence grade: A. Dates come from the GitHub releases API, the release announcements, the release history page and git history of the enhancements repository; only a few 2016 and 2019 cells rest on release announcements that give less detail.
2. Scorecard
| Dimension | Score | Direction |
|---|---|---|
| Capability today | 54.6 / 100 | higher is better |
| Momentum since end of 2022 | +2.6 points | higher is better |
| Credibility | 21.5 / 25 | higher is better |
| Commercial risk | not scored (engine: nothing to buy) | |
| Technical lock-in | 1.5 / 25 | lower is better |
| Integrator fit | 16.5 / 20 | higher is better |
| AI leverage | 4.5 / 20 | higher is better |
| Ledger Index | 66.3 / 100 | balanced weights |
3. Capability by domain
Scores 0 to 5 against fixed anchors; total is weighted to 100.
| Domain (weight) | 2016 | 2019 | 2022 | Today |
|---|---|---|---|---|
| Install, upgrade and lifecycle (12) | 1 | 3 | 3.5 | 3.5 |
| Bare metal, edge and on-prem infrastructure (10) | 1.5 | 2.5 | 3 | 3 |
| Multi-cluster and fleet management (10) | 1 | 1.5 | 2 | 2 |
| Networking (10) | 1.5 | 2.5 | 3 | 3 |
| Storage and data protection (10) | 2 | 2.5 | 3 | 3 |
| Security and compliance (12) | 1.5 | 2.5 | 3 | 3.25 |
| Developer platform (8) | 1 | 1.5 | 2 | 1.5 |
| Observability and day-2 operations (8) | 1 | 2 | 2 | 2.5 |
| VMs on Kubernetes (VMware exit) (10) | 0 | 0.5 | 1.5 | 2 |
| GPU and accelerator workloads (10) | 0 | 2 | 2.5 | 3 |
| Total / 100 | 21.2 | 41.8 | 52.0 | 54.6 |
Year by year: 2016: 21.2 · 2017: 21.2 · 2018: 21.2 · 2019: 41.8 · 2020: 42.8 · 2021: 42.8 · 2022: 52.0 · 2023: 52.0 · 2024: 54.0 · 2025: 55.0 · 2026: 54.6
What moved the score
| Year | Domain | Change | Points | Trigger |
|---|---|---|---|---|
| 2019 | Bare metal, edge and on-prem infrastructure | 1.5 → 2.5 | +2.0 | kubeadm on any Linux host with HA topologies; in-tree cloud providers; Cluster API still pre-0.3 |
| 2019 | Developer platform | 1 → 1.5 | +0.8 | custom resources GA in 1.16; Dashboard |
| 2019 | GPU and accelerator workloads | 0 → 2 | +4.0 | hardware accelerator support alpha in 1.9; device plugin framework beta since 1.12 |
| 2019 | Install, upgrade and lifecycle | 1 → 3 | +4.8 | kubeadm GA in 1.13 (2018-12-03) with kubeadm upgrade; 9-month support window documented |
| 2019 | Multi-cluster and fleet management | 1 → 1.5 | +1.0 | federation only in the 1.5 release notes; Cluster API pre-0.3 |
| 2019 | Networking | 1.5 → 2.5 | +2.0 | Network Policy API stable in 1.7; CoreDNS default in 1.13; IPVS kube-proxy GA in 1.11; implementation left to CNI plugins |
| 2019 | Observability and day-2 operations | 1 → 2 | +1.6 | overhauled metrics in 1.16 |
| 2019 | Security and compliance | 1.5 → 2.5 | +2.4 | RBAC stable in 1.8, node authorizer 1.7, Network Policy stable |
| 2019 | Storage and data protection | 2 → 2.5 | +1.0 | CSI GA in 1.13 (2018-12-03); local persistent volumes GA in 1.14 |
| 2019 | VMs on Kubernetes (VMware exit) | 0 → 0.5 | +1.0 | KubeVirt accepted to CNCF 2019-09-06 (community add-on, counted half) |
| 2020 | Storage and data protection | 2.5 → 3 | +1.0 | volume snapshots stable in 1.20 (2020-12-08) |
| 2022 | Bare metal, edge and on-prem infrastructure | 2.5 → 3 | +1.0 | Cluster API v1.3 (2022-12-01) with vSphere, OpenStack, Metal3 and cloud providers (listed by the providers' own teams) |
| 2022 | Developer platform | 1.5 → 2 | +0.8 | server-side apply GA (1.22), ephemeral containers stable (1.25); Dashboard |
| 2022 | GPU and accelerator workloads | 2 → 2.5 | +1.0 | device plugin framework stable in 1.26; Dynamic Resource Allocation alpha (2022-12-09) |
| 2022 | Install, upgrade and lifecycle | 3 → 3.5 | +1.2 | Cluster API v1.0 (2021-10-06, v1beta1) and v1.3 (2022-12-01); one-year support since 1.19; three releases a year since 2021 |
| 2022 | Multi-cluster and fleet management | 1.5 → 2 | +1.0 | Cluster API v1.3 management clusters; Karmada and Open Cluster Management Sandbox projects [calibrated: capped at 2, upstream ships no add-ons] |
| 2022 | Networking | 2.5 → 3 | +1.0 | Ingress GA (1.19), dual-stack GA (1.23), EndpointSlices; Gateway API v0.5.0 (2022-07-13) |
| 2022 | Security and compliance | 2.5 → 3 | +1.2 | Pod Security admission stable and PodSecurityPolicy removed in 1.25 (2022-08-23); release artifacts signed (beta, 1.26) |
| 2022 | VMs on Kubernetes (VMware exit) | 0.5 → 1.5 | +2.0 | KubeVirt CNCF Incubating 2022-04-19 (community add-on) |
| 2024 | Observability and day-2 operations | 2 → 2.5 | +0.8 | structured logging, API server tracing stable in 1.30 (2024-04-17); no bundled metrics store, dashboards or alerting |
| 2024 | Security and compliance | 3 → 3.5 | +1.2 | KMS v2 GA (1.29), ValidatingAdmissionPolicy stable (1.30), structured authorization stable (1.32), SPDX SBOMs; no FIPS, STIG, scanner or runtime detection |
| 2025 | GPU and accelerator workloads | 2.5 → 3 | +1.0 | core of Dynamic Resource Allocation GA in 1.34 (2025-08-27), more DRA features stable in 1.37; drivers and GPU operators come from vendors [calibrated: capped at 3, upstream ships no add-ons] |
| 2026 | Developer platform | 2 → 1.5 | -0.8 | Kubernetes Dashboard archived (post of 2026-06-01); Headlamp is the successor but not installed by kubeadm |
| 2026 | Security and compliance | 3.5 → 3.25 | -0.6 | 3.5 earned (structured authentication 1.34, user namespaces 1.36) less 0.25 for CVE-2024-9486, critical in Image Builder for Cluster API images (fixed 2024-10-14) |
| 2026 | VMs on Kubernetes (VMware exit) | 1.5 → 2 | +1.0 | KubeVirt (still Incubating, v1.9.1 2026-10-12) with live migration and snapshots, counted half; nothing in kubeadm or Cluster API releases |
4. Direction, 2023 to 2026
Each item carries one theme and one driver (V vision, C customer need, M market window, U upstream, P portfolio).
| Date | Item | Theme | Driver |
|---|---|---|---|
| 2023-08 | Kubelet skew widened to n-3 (1.28) | D1 Install and lifecycle | C |
| 2023-12 | KMS v2 encryption at rest GA (1.29) | D9 Security | U |
| 2024-04 | ValidatingAdmissionPolicy and API server tracing stable (1.30) | D9 Security | U |
| 2025-04 | Sidecar containers and nftables kube-proxy stable (1.33) | D4 Networking | U |
| 2025-08 | Dynamic Resource Allocation core GA (1.34) | D13 Accelerators | M |
| 2025-12 | In-place pod resize stable (1.35) | D1 Install and lifecycle | C |
| 2026-03 | AI Gateway Working Group formed | D13 Accelerators | M |
| 2026-03 | ingress-nginx retired; Gateway API is the path forward | D4 Networking | P |
| 2026-04 | User namespaces and mutating admission policies stable (1.36) | D9 Security | U |
| 2026-06 | Kubernetes Dashboard archived; Headlamp named successor | D6 Developer platform | P |
| 2026-08 | Pod certificates and more DRA features stable (1.37) | D9 Security | U |
- Centre of gravity: Security (4), Install and lifecycle (2), Networking (2)
- Driver mix: Vision 0%, Customer need 18%, Market window 18%, Upstream 45%, Portfolio 18%
5. Credibility: 21.5 / 25
| Criterion | Score | Basis |
|---|---|---|
| Cadence | 5 | three minors a year on a published calendar, every minor from 1.26 to 1.37 on schedule; 1.38.0 proposed for 2026-12-16 |
| Roadmap transparency | 5 | public enhancement proposals with milestones, open issue tracker, release schedule and release history with end-of-life dates |
| Say-do | 3 | 8 of 13 tracked KEPs delivered within 12 months of the stated release (62%); 4 later, 1 dropped |
| Velocity | 5 | 17 enhancements to stable in 1.30, 23 in 1.34; several substantive capabilities every release |
| Lifecycle stability | 3.5 | patch support about 14 months with clean in-place upgrades and n-3 skew, but no long-term version and forced removals (dockershim 1.24, PodSecurityPolicy 1.25, ingress-nginx 2026, Dashboard 2026) |
Say-do record, 2023 to 2026: 13 KEPs tracked from kep.yaml history: 6 shipped in the promised release, 2 within 12 months (KMS v2 +120 days, user namespaces +126), 4 later (sidecar containers +371, structured authentication +379, node swap +743, in-place resize +609), 1 dropped (classic DRA, withdrawn 2024-10-04).
6. Commercial risk: not scored
This is an open-source engine: nobody sells it, so there are no commercial terms to score. Its project health is described below.
Security record of the product itself: No CISA KEV entry for Kubernetes or its sub-projects (file dated 2026-10-08). One critical flaw in an own component (CVE-2024-9486, Image Builder, 9.8, fixed v0.1.38 3 days after the advisory) subtracts 0.25; CVE-2025-1974 (ingress-nginx, 9.8) is not part of kubeadm or Cluster API and is not counted.
7. Technical lock-in: 1.5 / 25 (lower is better)
| Criterion | Score | Basis |
|---|---|---|
| Formats | 0 | reference APIs; every conformant distribution uses the same objects |
| Export path | 0 | this is the standard other platforms export to; Velero and standard tools apply |
| Stack coupling | 0.5 | kubeadm needs a CRI, CNI and CSI choice made by the operator |
| Hardware / cloud coupling | 0 | any Linux or Windows host, fully self-hosted |
| Skills and tooling coupling | 1 | kubectl and Cluster API skills; assembly is left to the operator |
8. Integrator fit: 16.5 / 20
| Criterion | Score | Basis |
|---|---|---|
| Partner programme openness | 4 | open Certified Kubernetes programme and training partners; no vendor partner tiers |
| Multi-tenancy and self-service | 3 | namespaces, RBAC, quotas and network policy; hosted control planes only through third-party Cluster API providers |
| API and infrastructure-as-code | 4.5 | Cluster API and kubeadm are the standard infrastructure-as-code path; Terraform and Helm are third-party |
| Skills and certification | 5 | CKA, CKAD and CKS certifications from CNCF and the Linux Foundation |
9. AI leverage: 4.5 / 20
| Strand | Today | Latest step |
|---|---|---|
| AI workload support | 2.5 | 2026: image volumes stable in 1.36, Inference Extension v1.0.0, AI Gateway Working Group (2026-03-09), gang scheduling beta in 1.37 |
| AI platform stack | 0 | none |
| AI-assisted operations | 1 | 2025: Headlamp AI Assistant plugin (2025-08-07), a community plugin for the SIG UI project, counted half |
| Agent openness | 1 | 2016: documented Kubernetes REST API with OpenAPI; no official Kubernetes MCP server exists |
AI say-do: 3 announced, 1 shipped on time, 1 shipped 364 days late, 0 preview, 0 pending, 1 dropped. Upstream ships GPU scheduling (scored under accelerators) and a few AI-adjacent pieces, but no AI platform, assistant or MCP server.
10. Reading
Upstream Kubernetes is the reference line for every other entry: three releases a year on schedule, public enhancement proposals, and nothing to buy. What it ships is an engine and a lifecycle tool, not a platform, so storage, networking, observability, developer tooling and virtualization are left to community add-ons or to the distributions built on it. The record is strong on cadence and openness and mixed on say-do, where six of thirteen tracked proposals reached stable later than stated and one was dropped.
11. Open questions and evidence caveats
- Scored as kubeadm plus Cluster API only; a team that adds Cilium, Velero, KubeVirt and an observability stack gets more than the scores show, and pays for the assembly in skills.
- Support is about 14 months per minor with no long-term version; Kubernetes 1.34 reaches end of life on 2026-10-27.
- Say-do is measured from enhancement-proposal milestone changes, not from announced dates; delays of one to six minors are common.
- Community add-ons (KubeVirt Incubating, Karmada Graduated, Metal3 Incubating) are counted at half; Gateway API and Cluster API providers are maintained by separate teams.
More from On-Prem Kubernetes Ledger 2026
- Overview: all platforms side by side
- On-Prem Kubernetes Ledger Methodology: How Every Score Is Built
- Amazon EKS Anywhere: On-Prem Kubernetes Ledger
- Canonical Kubernetes: On-Prem Kubernetes Ledger
- DaoCloud Enterprise: On-Prem Kubernetes Ledger
- Deckhouse Kubernetes Platform: On-Prem Kubernetes Ledger
- Giant Swarm managed platform: On-Prem Kubernetes Ledger
- Google Distributed Cloud, software only: On-Prem Kubernetes Ledger
- Kubermatic Kubernetes Platform: On-Prem Kubernetes Ledger
- KubeSphere: On-Prem Kubernetes Ledger
- Microsoft AKS on Azure Local: On-Prem Kubernetes Ledger
- Mirantis Kubernetes Engine: On-Prem Kubernetes Ledger
- Nutanix Kubernetes Platform: On-Prem Kubernetes Ledger
- OKD: On-Prem Kubernetes Ledger 2026
- Oracle Cloud Native Environment: On-Prem Kubernetes Ledger
- Platform9 Managed Kubernetes: On-Prem Kubernetes Ledger
- Red Hat OpenShift Container Platform: On-Prem Kubernetes Ledger
- Sidero Labs Talos Linux and Omni: On-Prem Kubernetes Ledger
- Spectro Cloud Palette: On-Prem Kubernetes Ledger
- SUSE Rancher Prime: On-Prem Kubernetes Ledger
- VMware vSphere Kubernetes Service: On-Prem Kubernetes Ledger