Upstream Kubernetes: Cloud Kubernetes Ledger
Upstream Kubernetes is building toward a standard scheduling and admission layer for accelerators and policy, with a release train that holds its dates.
Part of the Managed Cloud Kubernetes Ledger (edition 2026-10-13). How the scores work: methodology.
1. Snapshot
- Tier: Open-source engine
- Owner: Cloud Native Computing Foundation (Linux Foundation); open-source project, no single vendor
- Licence model: Apache-2.0; free, self-operated, no support contract
- Products scored: Kubernetes releases 1.35 to 1.37 as shipped by the project (API, kubelet, kubeadm, reference controllers); SIG projects named in the notes
- Latest release: v1.37.0, 2026-08-26 (v1.37.1 on 2026-09-15)
- Next signal: v1.38.0 planned 2026-12-16
- Archetype: Reference engine, DIY
- Evidence grade: A. Dates come from git tags, KEP files and the release schedules read in this run; ops and accel cells for 2016 are not confirmed and carry little weight.
2. Scorecard
| Dimension | Score | Direction |
|---|---|---|
| Capability today | 46.4 / 100 | higher is better |
| Momentum since end of 2022 | +8.0 points | higher is better |
| Credibility | 21.5 / 25 | higher is better |
| Commercial risk | not scored (engine: nothing to buy) | |
| Technical lock-in | 0.5 / 25 | lower is better |
| Integrator fit | 16.0 / 20 | higher is better |
| AI leverage | 5.0 / 20 | higher is better |
| Ledger Index | 65.5 / 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 |
|---|---|---|---|---|
| Versions, upgrades and support (12) | 2 | 2.5 | 2.5 | 2.5 |
| Nodes, autoscaling and compute choice (12) | 1 | 1.5 | 1.5 | 2 |
| Networking (10) | 1.5 | 2 | 2.5 | 3 |
| Storage and data protection (8) | 1 | 2 | 2.5 | 3 |
| Security and compliance (12) | 1.5 | 2.5 | 3 | 3.5 |
| Observability and day-2 operations (8) | 1 | 1 | 1.5 | 2 |
| Multi-cluster, hybrid and fleet (10) | 0.5 | 1 | 1.5 | 1.5 |
| Developer and app platform (8) | 1 | 1.5 | 2 | 2 |
| GPU and AI workloads (10) | 0.5 | 1.5 | 2 | 3.5 |
| Regions, sovereignty and SLA (10) | 0 | 0 | 0 | 0 |
| Total / 100 | 20.6 | 31.8 | 38.4 | 46.4 |
Year by year: 2016: 20.6 · 2017: 20.6 · 2018: 20.6 · 2019: 31.8 · 2020: 31.8 · 2021: 31.8 · 2022: 38.4 · 2023: 38.4 · 2024: 38.4 · 2025: 38.4 · 2026: 46.4
What moved the score
| Year | Domain | Change | Points | Trigger |
|---|---|---|---|---|
| 2019 | Developer and app platform | 1 → 1.5 | +0.8 | CRDs GA in 1.16 (2019-09-18) |
| 2019 | GPU and AI workloads | 0.5 → 1.5 | +2.0 | Device plugin API beta since 1.10 |
| 2019 | Multi-cluster, hybrid and fleet | 0.5 → 1 | +1.0 | Cluster API repository since 2018; not part of the release |
| 2019 | Networking | 1.5 → 2 | +1.0 | NetworkPolicy stable since 1.7; EndpointSlice alpha in 1.16 |
| 2019 | Nodes, autoscaling and compute choice | 1 → 1.5 | +1.2 | Windows node support production-level in 1.14 (2019-03-25) |
| 2019 | Security and compliance | 1.5 → 2.5 | +2.4 | RBAC stable in 1.8; Secret encryption alpha since 1.7 |
| 2019 | Storage and data protection | 1 → 2 | +1.6 | CSI GA in 1.13, local persistent volumes GA in 1.14 |
| 2019 | Versions, upgrades and support | 2 → 2.5 | +1.2 | kubeadm GA in 1.13 gave a documented upgrade path; still no automatic patching or long support |
| 2022 | Developer and app platform | 1.5 → 2 | +0.8 | Ephemeral containers and kubectl debug stable in 1.25 |
| 2022 | GPU and AI workloads | 1.5 → 2 | +1.0 | Device plugin API stable and DRA alpha, both in 1.26 |
| 2022 | Multi-cluster, hybrid and fleet | 1 → 1.5 | +1.0 | Cluster API v1.0.0 on 2021-10-06 as a SIG project |
| 2022 | Networking | 2 → 2.5 | +1.0 | EndpointSlice stable 1.21, dual-stack GA 1.23, NetworkPolicy endPort stable 1.25 |
| 2022 | Observability and day-2 operations | 1 → 1.5 | +0.8 | Ephemeral containers stable 1.25 for debugging; structured logging beta since 1.23 |
| 2022 | Security and compliance | 2.5 → 3 | +1.2 | Bound service account tokens stable 1.22; Pod Security Admission stable and PodSecurityPolicy removed in 1.25 |
| 2022 | Storage and data protection | 2 → 2.5 | +0.8 | Volume snapshots GA 1.20, volume expansion stable 1.24, generic ephemeral volumes GA 1.23 |
| 2026 | GPU and AI workloads | 2 → 3.5 | +3.0 | DRA core GA 1.34, CDI in device plugins stable 1.31, DRA device taints and extended resources stable 1.37 |
| 2026 | Networking | 2.5 → 3 | +1.0 | nftables kube-proxy, traffic distribution and multiple service CIDRs stable in 1.33; no CNI or Gateway API in the release |
| 2026 | Nodes, autoscaling and compute choice | 1.5 → 2 | +1.2 | Sidecars stable 1.33, swap stable 1.34, in-place resize stable 1.35, user namespaces 1.36; still no node autoscaling |
| 2026 | Observability and day-2 operations | 1.5 → 2 | +0.8 | API server and kubelet tracing stable in 1.34; no managed metrics, cost allocation or upgrade insights |
| 2026 | Security and compliance | 3 → 3.5 | +1.2 | KMS v2 1.29, ValidatingAdmissionPolicy 1.30, MutatingAdmissionPolicy and user namespaces 1.36; no image signing or threat detection |
| 2026 | Storage and data protection | 2.5 → 3 | +0.8 | VolumeGroupSnapshot and OCI volume source GA in 1.36; no drivers or backup tool shipped |
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 |
|---|---|---|---|
| 2023-12 | KMS v2 stable, in-tree cloud providers off by default | D9 Security | C |
| 2024-04 | ValidatingAdmissionPolicy stable; DRA structured parameters alpha | D9 Security | V |
| 2024-08 | In-tree cloud provider code removed; nftables kube-proxy beta | D11 Ecosystem openness | V |
| 2024-12 | DRA structured parameters beta; MutatingAdmissionPolicy alpha | D13 Accelerators | V |
| 2025-04 | Sidecar containers and nftables kube-proxy stable | D2 Compute and autoscaling | C |
| 2025-08 | DRA core GA; API server and kubelet tracing stable; node swap stable | D13 Accelerators | M |
| 2025-11 | Ingress NGINX retirement announced | D4 Networking | C |
| 2025-12 | In-place pod resize stable | D2 Compute and autoscaling | C |
| 2026-04 | Volume group snapshots, user namespaces, OCI volumes and MutatingAdmissionPolicy stable | D9 Security | C |
| 2026-08 | DRA device taints and extended resources stable | D13 Accelerators | M |
- Centre of gravity: Security (3), Accelerators (3), Compute and autoscaling (2)
- Driver mix: Vision 30%, Customer need 50%, Market window 20%, Upstream 0%, Portfolio 0%
5. Credibility: 21.5 / 25
| Criterion | Score | Basis |
|---|---|---|
| Cadence | 5 | Three minors a year; 10 of 11 releases from 1.27 to 1.37 shipped on the date published at cycle start |
| Roadmap transparency | 5 | Public KEP repository with commit history, sig-release schedules and an official CVE feed |
| Say-do | 4 | 20 of 24 stated targets met within 12 months; 16 of 24 on the stated date; sidecars, resize, swap and group snapshots slipped over a year |
| Velocity | 5 | 11 to 23 features reach stable in every release; 44 to 70 enhancements per release since 1.27 |
| Lifecycle stability | 2.5 | About 14 months per minor, no long-term support release; upgrade every minor, APIs get removed |
Say-do record, 2023 to 2026: Releases: 10 of 11 on date, 1.29 slipped 8 days. KEPs: 13 tracked, 6 on target, 7 slipped (sidecars 3 minors, in-place resize 5, node swap 4, volume group snapshot 5, OCI volume 2, user namespaces 1, KMS v2 1).
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: None: upstream is not a managed service. Official CVE feed lists 94 items, none in CISA KEV (catalogue 2026.10.08). Latest: CVE-2026-76654 on 2026-09-14.
7. Technical lock-in: 0.5 / 25 (lower is better)
| Criterion | Score | Basis |
|---|---|---|
| Formats | 0 | The reference API; workloads use standard Kubernetes objects |
| Export path | 0 | Nothing to export: runs anywhere, no egress fee, no vendor service |
| Stack coupling | 0 | No provider IAM, load balancer or storage class required |
| Hardware / cloud coupling | 0 | Runs on any cloud, bare metal or laptop |
| Skills and tooling coupling | 0.5 | Standard kubectl; operating etcd and kubeadm yourself needs scarcer skills |
8. Integrator fit: 16.0 / 20
| Criterion | Score | Basis |
|---|---|---|
| Partner programme openness | 4.5 | Vendor-neutral CNCF conformance programme with 131 listed products for 1.34 and 1.35 |
| Multi-tenancy and self-service | 2.5 | Namespaces, RBAC and quotas only; no organisations, projects or fleet isolation |
| API and infrastructure-as-code | 4 | Cluster API v1.14 and client libraries; no official Terraform provider |
| Skills and certification | 5 | CKA exam offered by the Linux Foundation and the largest public documentation and community |
9. AI leverage: 5.0 / 20
| Strand | Today | Latest step |
|---|---|---|
| AI workload support | 3.5 | 2026: DRA extended resources and device taints stable 1.37; agent-sandbox 1.0.0 on 2026-08-28 |
| AI platform stack | 0 | none |
| AI-assisted operations | 0 | 2023: The project ships no assistant; k8sgpt and kubectl-ai are third-party |
| Agent openness | 1.5 | 2016: Documented REST API; no official MCP server (the draft's kubernetes-sigs repository does not exist) |
AI say-do: 3 tracked, 2 GA, 1 pending (DRA consumable capacity slipped from 1.36 to 1.38). The project supplies scheduling primitives for GPUs; it has no assistant and no official MCP server.
10. Reading
Upstream Kubernetes is building toward a standard scheduling and admission layer for accelerators and policy, with a release train that holds its dates. The record shows reliable releases and features that often take two to five extra minors to reach stable. A buyer gets full portability and no fee, and pays with operations: no hosted control plane, no node provisioning, 14 months of support.
11. Open questions and evidence caveats
- Ops 2016 and accel 2016 cells are not confirmed in the research; both carry low scores with little effect.
- Cluster API, Kueue, JobSet and Gateway API are SIG projects outside the release; Cluster API is credited in fleet, the others only in AI infra.
- No official MCP server was found; community servers do not count under the rubric.
- CVE severity is missing for many feed items; the feed gives CVSS only for some entries.
More from Managed Cloud Kubernetes Ledger 2026
- Overview: all platforms side by side
- Cloud Kubernetes Ledger Methodology: How Every Score Is Built
- Akamai Linode Kubernetes Engine: Cloud Kubernetes Ledger
- Alibaba ACK: Cloud Kubernetes Ledger 2026
- Amazon Elastic Kubernetes Service: Cloud Kubernetes Ledger
- Azure Kubernetes Service: Cloud Kubernetes Ledger
- Civo Kubernetes: Cloud Kubernetes Ledger
- CoreWeave Kubernetes Service: Cloud Kubernetes Ledger
- DigitalOcean Kubernetes: Cloud Kubernetes Ledger
- Google Kubernetes Engine: Cloud Kubernetes Ledger
- Huawei Cloud Cloud Container Engine: Cloud Kubernetes Ledger
- IBM IKS / ROKS: Cloud Kubernetes Ledger 2026
- IONOS Cloud Managed Kubernetes: Cloud Kubernetes Ledger
- Open Telekom Cloud Cloud Container Engine: Cloud Kubernetes Ledger
- Oracle OCI Kubernetes Engine: Cloud Kubernetes Ledger
- OVHcloud Managed Kubernetes Service: Cloud Kubernetes Ledger
- Red Hat OpenShift Service on AWS: Cloud Kubernetes Ledger
- Scaleway Kubernetes Kapsule and Kosmos: Cloud Kubernetes Ledger
- STACKIT Kubernetes Engine: Cloud Kubernetes Ledger
- Tencent Kubernetes Engine: Cloud Kubernetes Ledger
- Vultr Kubernetes Engine: Cloud Kubernetes Ledger