Mr.PlanB Logo
    On-Prem Kubernetes Ledger 2026
    Ledger
    On-Prem Kubernetes Ledger 2026

    Google Distributed Cloud, software only: On-Prem Kubernetes Ledger

    Google sells the on-premises clusters as an extension of its cloud: fleet, billing, identity and the install and upgrade path all run through a Google…

    Edition October 13, 20266 min read

    Part of the On-Prem Kubernetes Ledger (edition 2026-10-13). How the scores work: methodology.

    1. Snapshot

    • Tier: Hyperscaler on-prem
    • Owner: Alphabet Inc. (Google Cloud), public company
    • Licence model: Commercial software billed per vCPU in user clusters through a Google Cloud billing account; list price $0.03288 per vCPU hour on bare metal and VMware; VM Runtime included
    • Products scored: GDC software only for bare metal (ex Anthos on bare metal) and for VMware (ex GKE On-Prem); the GDC connected and air-gapped hardware products are not scored
    • Latest release: Bare metal and VMware 1.36.0-gke.532 (Kubernetes 1.36), 2026-08-25; patch 1.36.100-gke.144 on 2026-10-07
    • Next signal: No dated roadmap; 1.37 not yet released (upstream Kubernetes 1.37 shipped 2026-08-26)
    • Archetype: Cloud-tethered hybrid kubernetes
    • Evidence grade: B. Release notes, version history, support policy, pricing page and Wayback snapshots read directly; no public roadmap, support policy before 2023 and partner facts not confirmed, and no CNCF conformance listing exists to cross-check.

    Not ranked: Inclusion test 4 not met on the evidence: cncf/k8s-conformance has no Anthos or Google Distributed Cloud folder for Kubernetes 1.24 to 1.37 (only the hosted GKE service is listed). Test 3 is met only in part: the management plane is Google Cloud with no self-hosted option, and creating or upgrading clusters needs a connection to it.

    2. Scorecard

    DimensionScoreDirection
    Capability today64.8 / 100higher is better
    Momentum since end of 2022+6.0 pointshigher is better
    Credibility16.0 / 25higher is better
    Commercial risk9.0 / 25lower is better
    Technical lock-in13.0 / 25lower is better
    Integrator fit12.0 / 20higher is better
    AI leverage4.0 / 20higher is better
    Ledger Index56.2 / 100balanced weights

    3. Capability by domain

    Scores 0 to 5 against fixed anchors; total is weighted to 100.

    Domain (weight)201620192022Today
    Install, upgrade and lifecycle (12)n/a233.5
    Bare metal, edge and on-prem infrastructure (10)n/a1.533
    Multi-cluster and fleet management (10)n/a1.533.5
    Networking (10)n/a244
    Storage and data protection (10)n/a233
    Security and compliance (12)n/a233.5
    Developer platform (8)n/a111.5
    Observability and day-2 operations (8)n/a233.5
    VMs on Kubernetes (VMware exit) (10)n/a033
    GPU and accelerator workloads (10)n/a033.5
    Total / 100not released28.458.864.8

    Year by year: 2019: 28.4 · 2020: 32.4 · 2021: 37.4 · 2022: 58.8 · 2023: 61.2 · 2024: 63.0 · 2025: 64.0 · 2026: 64.8

    What moved the score

    YearDomainChangePointsTrigger
    2020GPU and accelerator workloads0 → 1+2.0Bare metal 1.6 lists NVIDIA GPU support; VMware GPU attach in beta 2020-08-20
    2020VMs on Kubernetes (VMware exit)0 → 1+2.0Bare metal 1.6 (2020-11-30): KubeVirt VM management in preview
    2021Bare metal, edge and on-prem infrastructure1.5 → 3+3.0Bare metal GA 2020-11-30 and 1.8 (2021-06-21): no hypervisor, OpenStack, edge profile; cloud connection mandatory since 1.7
    2021Networking2 → 3+2.0Dataplane V2 GA on VMware 2021-03-25; multi-NIC GA in bare metal 1.9; BGP load balancer preview
    2022GPU and accelerator workloads1 → 3+4.01.13 and 1.14 (2022): NVIDIA device plugin with MIG manager, Tesla T4, GPUs in VMs preview
    2022Install, upgrade and lifecycle2 → 3+2.4Bare metal 1.14 (2022-12-13): bmctl, automatic health checks, upgrades through a bootstrap cluster; each minor supported 12 months
    2022Multi-cluster and fleet management1.5 → 3+3.01.13 (2022-09-29): fleet registration, Connect gateway with Google Groups, admin cluster managing user clusters
    2022Networking3 → 4+2.0Bare metal 1.11 (2022-03-31): bundled BGP and L2 load balancing, SR-IOV, dual-stack GA; 1.13 flat IP with BGP
    2022Observability and day-2 operations2 → 3+1.61.13 (2022-09-29): Managed Service for Prometheus GA, Cloud Logging, node problem detector, sample dashboards
    2022Security and compliance2 → 3+2.41.14 (2022-12-13): seccomp, SELinux on RHEL, Workload Identity, audit logs, LDAP identity, Azure AD groups
    2022Storage and data protection2 → 3+2.0CSI v1.0 drivers from Ready storage partners, snapshots, bmctl control-plane backup GA in 1.10 (2021-12-10); no bundled storage or data backup
    2022VMs on Kubernetes (VMware exit)1 → 3+4.0VM Runtime GA 2022-08-23 (1.12): Windows guests, VLAN and IP stickiness, VM HA, no extra charge; live migration and snapshots absent
    2023Install, upgrade and lifecycle3 → 3.5+1.21.15 (2023-04-27) in-place upgrade default; 1.16 (2023-08-25) parallel node-pool upgrades; GKE On-Prem API enrollment; lag 133 days
    2023Security and compliance3 → 3.5+1.21.16 (2023-08-25) Binary Authorization and VPC Service Controls GA; 1.29 Identity Service v2; no FIPS, STIG or SBOM found, no KEV in own components
    2024Multi-cluster and fleet management3 → 3.5+1.01.30 (2024-08-29): one admin cluster manages several user-cluster versions; 100 user clusters per admin cluster tested
    2024Observability and day-2 operations3 → 3.5+0.8Alerting policies, periodic health checks, Managed Collector GA; no cost or capacity reporting found
    2025GPU and accelerator workloads3 → 3.5+1.01.33 (2025-09-02): bundled NVIDIA GPU Operator and Dynamic Resource Allocation in preview (half step); still preview in 1.36
    2026Developer platform1 → 1.5+0.8No registry, CI or portal in the 173 bare-metal doc pages; Config Sync and console workload views are outside the product

    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).

    DateItemThemeDriver
    2023-041.15: in-place upgrade default and parallel node-pool upgrades in previewD1 Install and lifecycleC
    2023-081.16: Binary Authorization and VPC Service Controls GAD9 SecurityC
    2023-081.16: GKE On-Prem API enrollment from console, gcloud and Terraform GAD3 Fleet managementP
    2023-11GKE Enterprise edition GA, combining GKE and AnthosD12 Packaging and licensingP
    2023-121.28: version numbers aligned with GKE; pause and resume upgrades in previewD1 Install and lifecycleU
    2024-041.29: GKE Identity Service v2 GAD9 SecurityC
    2024-081.30: node-pool upgrade rollback GAD1 Install and lifecycleC
    2025-051.32: separate control-plane and data-plane L2 load balancingD4 NetworkingC
    2025-091.33: Node Agent, bundled NVIDIA GPU Operator and Dynamic Resource Allocation in previewD13 AcceleratorsM
    2025-09GKE consolidated into a single paid offering with standalone SKUsD12 Packaging and licensingP
    2025-121.34: skip-minor cluster upgrades GAD1 Install and lifecycleC
    2026-051.35: cgroup v2 required, containerd 2.1D1 Install and lifecycleU
    2026-081.36: vertical Pod autoscaling GA, layer 4 gateway controllerD8 Observability and operationsC
    • Centre of gravity: Install and lifecycle (5), Security (2), Packaging and licensing (2)
    • Driver mix: Vision 0%, Customer need 54%, Market window 8%, Upstream 15%, Portfolio 23%

    5. Credibility: 16.0 / 25

    CriterionScoreBasis
    Cadence4A minor every four to five months since 1.28, 106 to 146 days after upstream; final patch dates and a written support policy
    Roadmap transparency2.5Detailed dated release notes and version history; no public roadmap, tracker or direction beyond Next blog posts
    Say-do3.514 of 19 previews reached GA within 12 months (median 121 days); 5 stay in preview 13 to 30 months; mostly small features
    Velocity3One or two substantive capabilities per release: parallel upgrades, Identity Service v2, Node Agent, skip-minor upgrades, VPA
    Lifecycle stability312-month minors with in-place upgrades, but renamed three times since 2019 and folded into GKE Enterprise in 2023; no extended support

    Say-do record, 2023 to 2026: 19 items 2023-2026, built from preview-to-GA entries in the release notes plus the GKE Enterprise keynote promise: 14 shipped within 12 months (78 to 357 days), 5 pending beyond 12 months (Dynamic Resource Allocation, bundled GPU Operator, keyless mode, custom scheduler, intermediate CA), 0 dropped. No roadmap or keynote promises exist for the software-only line.

    6. Commercial risk: 9.0 / 25 (lower is better)

    CriterionScoreBasis
    Price and licence volatility2Renamed Anthos to GDC Virtual 2022-06-17 and into GKE Enterprise 2023-11-15 with no on-prem price change; GKE editions consolidated September 2025
    Purchase constraints1.5Pay as you go per vCPU hour through a Google Cloud billing account; custom quotes for commitments; no minimum seen
    Channel and access1.5Documentation and release notes open; partner programme exists; billing and fleet registration require a Google Cloud project
    Owner stability1.5Alphabet is stable, but the Anthos brand and product line were restructured twice and the software-only line is a minor part of the portfolio
    Cost of staying supported2.5Free updates but a 12-month window per minor, mandatory upgrades every year and new OS floors (cgroup v2, RHEL 8.10, Ansible 2.18)

    Security record of the product itself: No CISA KEV entry and no critical flaw in GDC's own components (bmctl, Connect, Identity Service, VM Runtime) found in 171 Google bulletins read. Critical containerd CVE-2026-50195 (CVSS 9.9, GCP-2026-051, 2026-07-31) and runc breakouts are upstream runtimes patched by every distribution, so no deduction.

    7. Technical lock-in: 13.0 / 25 (lower is better)

    CriterionScoreBasis
    Formats2Workloads use upstream APIs; VMs and cluster lifecycle use vm.cluster.gke.io and baremetal.cluster.gke.io CRDs; no CNCF conformance listing found
    Export path2.5Standard manifests move; bmctl backup restores control-plane state into GDC only; no documented migration path off the platform
    Stack coupling2Bring your own OS and any CSI; Cilium-based Dataplane V2, MetalLB and Config Sync are the supported network and policy stack
    Hardware / cloud coupling4Docs call a permanent Google Cloud connection the design; create and upgrade need it, fleet registration and metering are mandatory
    Skills and tooling coupling2.5bmctl, gkectl, gcloud and Google Cloud console skills on top of kubectl

    8. Integrator fit: 12.0 / 20

    CriterionScoreBasis
    Partner programme openness3Google Cloud Partner Advantage and storage Ready partners; hybrid specialisation details not opened
    Multi-tenancy and self-service2.5Admin and user clusters, namespaces, RBAC, fleet; no hosted control planes
    API and infrastructure-as-code4GKE On-Prem API with gcloud and Terraform resources for bare metal and VMware clusters; Terraform provider has 2.6 billion downloads
    Skills and certification2.5Google Cloud certifications exist; none specific to GDC found

    9. AI leverage: 4.0 / 20

    StrandTodayLatest step
    AI workload support2.52025: 1.33 (2025-09-02): bundled GPU Operator and Dynamic Resource Allocation in preview; vLLM serving guides
    AI platform stack0none
    AI-assisted operations0none
    Agent openness1.52023: GKE On-Prem API (gcloud, Terraform, REST) enrollment GA in 1.16, 2023-08-25; no official MCP server for GDC

    AI say-do: 2 announced, 0 GA, 2 preview, 2 pending, 0 dropped; GPU support is real but the AI platform, assistant and MCP work Google announced is for its connected and air-gapped hardware, not for software only.

    10. Reading

    Google sells the on-premises clusters as an extension of its cloud: fleet, billing, identity and the install and upgrade path all run through a Google Cloud project. The release notes show steady, dated delivery and a solid VM Runtime, but no public roadmap, GPU features stuck in preview and no CNCF conformance listing. It suits teams already on Google Cloud that accept a standing connection; it does not suit disconnected sites.

    11. Open questions and evidence caveats

    • No Anthos or Google Distributed Cloud folder exists in cncf/k8s-conformance for 1.24 to 1.37: confirm certification with Google before ranking.
    • Create and upgrade need a live connection to Google Cloud; the docs say disconnected operation is not a nominal mode.
    • Support policy text before 2023 and partner programme details were not confirmed.
    • Scores rest on release notes; there is no public roadmap, so the say-do record covers small features only.

    More from On-Prem Kubernetes Ledger 2026