Mr.PlanB Logo
    Managed Cloud Kubernetes Ledger 2026
    Ledger
    Managed Cloud Kubernetes Ledger 2026

    Google Kubernetes Engine: Cloud Kubernetes Ledger

    Google is building GKE into the place to run very large AI training and inference fleets on its own TPUs and NVIDIA GPUs, while folding Autopilot and…

    Edition October 13, 20266 min read
    Open the interactive LedgerMethodology

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

    1. Snapshot

    • Tier: Hyperscaler
    • Owner: Alphabet Inc. (Google Cloud), public company
    • Licence model: Pay as you go: $0.10 per cluster-hour (one zonal or Autopilot cluster covered by a $74.40 monthly credit), Autopilot billed per pod resource, +$0.50 per cluster-hour in extended support
    • Products scored: GKE Standard and Autopilot on Google Cloud, with fleet features, Backup for GKE and GKE multicloud as scored add-ons; Google Distributed Cloud is scored in the On-Prem Ledger
    • Latest release: Kubernetes 1.36 in Regular channel 2026-06-09; 1.37 in Rapid channel 2026-09-02
    • Next signal: 1.37 predicted in Regular channel 2026-10-14; No-channel clusters moved to Stable after 2027-06-14
    • Archetype: AI-scale Kubernetes reference
    • Evidence grade: A. Release notes (2014 to 2026), release schedule with Wayback history, pricing and SLA archives and security bulletins read directly; gaps are a few pre-2020 GA dates, compliance scope and partner facts.

    2. Scorecard

    DimensionScoreDirection
    Capability today92.0 / 100higher is better
    Momentum since end of 2022+9.4 pointshigher is better
    Credibility21.0 / 25higher is better
    Commercial risk5.5 / 25lower is better
    Technical lock-in8.0 / 25lower is better
    Integrator fit18.0 / 20higher is better
    AI leverage15.0 / 20higher is better
    Ledger Index82.4 / 100balanced weights

    3. Capability by domain

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

    Domain (weight)201620192022Today
    Versions, upgrades and support (12)33.544.5
    Nodes, autoscaling and compute choice (12)33.54.55
    Networking (10)23.54.55
    Storage and data protection (8)1.523.55
    Security and compliance (12)234.54
    Observability and day-2 operations (8)2344.5
    Multi-cluster, hybrid and fleet (10)13.54.55
    Developer and app platform (8)2.5344
    GPU and AI workloads (10)0345
    Regions, sovereignty and SLA (10)32.53.54
    Total / 10040.861.882.692.0

    Year by year: 2016: 40.8 · 2017: 39.8 · 2018: 48.8 · 2019: 61.8 · 2020: 63.8 · 2021: 70.2 · 2022: 82.6 · 2023: 85.0 · 2024: 90.4 · 2025: 92.0 · 2026: 92.0

    What moved the score

    YearDomainChangePointsTrigger
    2017Regions, sovereignty and SLA3 → 2.5-1.0SLA withdrawn 2017-11-28 when the fee was removed; internal goals only until 2020
    2018GPU and AI workloads0 → 3+6.0NVIDIA GPU node pools GA 2018-06-18; Cloud TPU in GKE beta 2018-07
    2018Networking2 → 3.5+3.0Alias IP VPC-native GA 2018-05-22 and private clusters GA 2018-10-02; Calico network policy
    2019Developer and app platform2.5 → 3+0.8Cloud Build, Marketplace apps; Config Connector in beta
    2019Multi-cluster, hybrid and fleet1 → 3.5+5.0Anthos 2019 with GKE On-Prem and config management; Config Sync GA 2020-01-29
    2019Nodes, autoscaling and compute choice3 → 3.5+1.2Node auto-provisioning and Vertical Pod Autoscaler GA 2019-10-11; GPU nodes GA 2018-06
    2019Observability and day-2 operations2 → 3+1.6Node auto-repair, usage metering GA 2019-06, Kubernetes monitoring in Stackdriver
    2019Security and compliance2 → 3+2.4RBAC, private clusters GA 2018-10; Binary Authorization and Shielded nodes in beta
    2019Storage and data protection1.5 → 2+0.8Still in-tree Persistent Disk; CSI driver only in beta 2020-02; no managed backup
    2019Versions, upgrades and support3 → 3.5+1.2Release channels GA 2019-09-16 and regional multi-master clusters; but 1.12 to 1.15 reached GA 153 to 177 days after upstream
    2020Regions, sovereignty and SLA2.5 → 3.5+2.0SLA back 2020-06-04: 99.95% regional, 99.5% zonal; Autopilot 99.95% from 2021-02
    2021Multi-cluster, hybrid and fleet3.5 → 4.5+2.0Multi-cluster Ingress GA 2020-04, Multi-cluster Services GA 2021-02, clusters on AWS and Azure, attached clusters
    2021Networking3.5 → 4.5+2.0Dataplane V2 eBPF GA 2021-05-12, multi-cluster Ingress and Services, Gateway GA 2022-11, dual-stack 2022-12
    2021Nodes, autoscaling and compute choice3.5 → 4.5+2.4Autopilot 2021-02 with Google-managed nodes, Windows nodes GA 2020-04, Spot VMs GA 2022-05, 15,000-node clusters
    2022Developer and app platform3 → 4+1.6Config Connector GA 2020-01-29, Artifact Registry, Cloud Deploy for GKE delivery
    2022GPU and AI workloads3 → 4+2.0Multi-Instance GPU GA 2021-09, GPU time-sharing GA 2022-06, TPU node pools
    2022Observability and day-2 operations3 → 4+1.6Managed Prometheus GA 2022-03, cost allocation GA 2022-10-14, deprecation insights in preview
    2022Security and compliance3 → 4.5+3.6Workload Identity GA 2020-03-16, GKE Sandbox GA 2020-01, Confidential nodes GA 2022-06-15, published bulletins
    2022Storage and data protection2 → 3.5+2.4Backup for GKE GA 2022-11-16 and Persistent Disk CSI; no object storage CSI yet
    2022Versions, upgrades and support3.5 → 4+1.2Surge upgrades GA 2020-02-11, maintenance windows and exclusions; no extended support yet
    2023Observability and day-2 operations4 → 4.5+0.8Infrastructure dashboards and metrics packages GA 2023-11-09, managed GPU metrics 2024-07
    2023Storage and data protection3.5 → 4.5+1.6Cloud Storage FUSE CSI GA 2023-07-13, Hyperdisk Extreme and Throughput GA 2023-06
    2024GPU and AI workloads4 → 5+2.0TPU v5e GA 2023-11, Trillium GA 2024-12, 65,000-node clusters; DRA GA 2025-09, Inference Gateway GA 2025-12
    2024Networking4.5 → 5+1.0FQDN network policies GA 2024-02, transparent inter-node encryption, managed DRANET GA 2026-04
    2024Nodes, autoscaling and compute choice4.5 → 5+1.265,000-node clusters 2024-11; Autopilot compute in Standard clusters 2025-09; Arm ComputeClasses
    2024Versions, upgrades and support4 → 4.5+1.2Extended channel GA 2024-07-31 with up to 24 months; Rapid within a week of upstream, Regular median 66 days
    2025Multi-cluster, hybrid and fleet4.5 → 5+1.0Fleet, team scopes, Config Sync and Policy Controller included in GKE from 2025-09-02; on-prem via Google Distributed Cloud
    2025Regions, sovereignty and SLA3.5 → 4+1.043 regions; SecNumCloud 3.2 for partner S3NS 2025-12; Google Cloud Dedicated with T-Systems announced 2025-05
    2025Security and compliance4.5 → 4-1.2Policy Controller included 2025-09, TDX nodes; less 1.0 for GCP-2025-001 firewall exposure and GCP-2026-058 cross-project flaw
    2025Storage and data protection4.5 → 5+0.8Hyperdisk ML GA 2024-08, Parallelstore CSI 2024-10, cross-project backup and restore GA 2025-08-25

    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
    2023-07Cloud Storage FUSE CSI driver GAD5 Storage and data protectionC
    2023-11Cloud TPU v5e GA in GKED13 AcceleratorsV
    2023-11GKE Enterprise edition GAD3 Fleet and hybridP
    2024-02FQDN network policies GAD4 NetworkingC
    2024-07Extended release channel with $0.50 per hour feeD1 Versions and lifecycleC
    2024-1165,000-node clustersD2 Compute and autoscalingM
    2024-12Trillium TPU v6e GA in GKED13 AcceleratorsV
    2025-05Container-optimized compute platform for AutopilotD2 Compute and autoscalingV
    2025-08Backup for GKE cross-project backup and restore GAD5 Storage and data protectionC
    2025-09Autopilot features in Standard clustersD2 Compute and autoscalingV
    2025-09Dynamic Resource Allocation GA with 1.34D13 AcceleratorsU
    2025-09GKE Enterprise features folded into GKE and standalone SKUsD12 Packaging and pricingP
    2025-12GKE Inference Gateway GAD13 AcceleratorsM
    2026-04Gemini Cloud Assist investigations limited to Premium SupportD7 AI-assisted operationsP
    2026-04Managed DRANET GAD4 NetworkingU
    • Centre of gravity: Accelerators (4), Compute and autoscaling (3), Storage and data protection (2)
    • Driver mix: Vision 27%, Customer need 27%, Market window 13%, Upstream 13%, Portfolio 20%

    5. Credibility: 21.0 / 25

    CriterionScoreBasis
    Cadence5Weekly dated release notes and a published release schedule with channel dates per minor, kept since 2021
    Roadmap transparency3.5Release schedule predicts dates by quarter; no public roadmap board, direction comes from Next and KubeCon posts
    Say-do3.510 of 13 resolvable promises 2023-2026 reached GA within 12 months (77%); Gemini Cloud Assist and RayTurbo slipped
    Velocity5Several GA capabilities per quarter in 2024-2026: TPUs, DRA, Inference Gateway, Autopilot in Standard, backup features
    Lifecycle stability4Extended channel to 24 months since 2024-07; but GKE Enterprise edition launched 2023 and dissolved 2025, No channel removed 2027

    Say-do record, 2023 to 2026: 14 items 2023-2026 from preview launches and Next keynotes: 10 shipped, 1 partial (Ironwood), 3 not delivered (Duet AI / Gemini Cloud Assist for GKE still preview since 2023-08, Cloud Assist investigations gated 2026-04, RayTurbo on GKE not found). No public roadmap board exists.

    6. Commercial risk: 5.5 / 25 (lower is better)

    CriterionScoreBasis
    Price and licence volatility2.5Since 2023: Enterprise edition at $0.00822 per vCPU-hour added then dissolved into SKUs 2025-09; extended fee $0.50 per hour from 2024-07
    Purchase constraints0.5Pay as you go, per-second cluster fee, monthly credit covers one zonal or Autopilot cluster
    Channel and access0Self-service sign-up, public docs, pricing and release notes, 43 regions
    Owner stability0.5Alphabet, profitable and stable; Google Cloud profitable
    Cost of staying supported2About 14 months standard support with forced auto-upgrade after; extended support costs $0.50 per cluster-hour

    Security record of the product itself: GCP-2025-001 (2025-01-08): Multi-Cluster Gateway created allow-all ingress firewall rules in customer VPCs 2024-12-04 to 2025-01-04, fixed by Google. GCP-2026-058 (2026-09-02, Critical): missing permission check in GKE Multi-Cloud APIs allowed cross-project Workload Identity token minting, fixed server-side, no exploitation. GCP-2023-047 and GCP-2025-033 are GKE-specific but need prior compromise; not deducted.

    7. Technical lock-in: 8.0 / 25 (lower is better)

    CriterionScoreBasis
    Formats1.5CNCF-conformant; Autopilot, ComputeClasses and Config Connector CRDs are Google-specific but optional
    Export path1.5Exit data transfer fees waived since 2024-01-11; Backup for GKE is Google-only, Velero works
    Stack coupling2.5Workload Identity to Google IAM, Gateway on Cloud Load Balancing, Persistent Disk and Hyperdisk storage classes
    Hardware / cloud coupling1Fleet runs GKE on AWS and Azure, attached clusters and Google Distributed Cloud on premises
    Skills and tooling coupling1.5kubectl, Terraform and gcloud; Autopilot and fleet concepts are GKE-specific

    8. Integrator fit: 18.0 / 20

    CriterionScoreBasis
    Partner programme openness4.5Google Cloud Partner Advantage with large SI and MSP base; programme details not re-read
    Multi-tenancy and self-service4.5Projects, folders and fleet team scopes with namespaces; multi-team management included since 2025-09
    API and infrastructure-as-code5hashicorp/google provider 2.65 billion downloads; GKE module 63 million; Config Connector; gcloud and SDKs
    Skills and certification4Professional Cloud Architect and DevOps Engineer certifications cover GKE; no GKE-only certification

    9. AI leverage: 15.0 / 20

    StrandTodayLatest step
    AI workload support52025: DRA GA 2025-09, Inference Gateway GA 2025-12, Inference Quickstart, DRANET GA 2026-04
    AI platform stack32025: Cluster Director for GKE GA at Next 2025, Inference Gateway with model-aware routing
    AI-assisted operations2.52025: Gemini Cloud Assist for GKE in preview since 2023 (half step); investigations read-only, Premium-only from 2026-04
    Agent openness4.52026: Remote GKE MCP server listed without preview label 2026, IAM roles and audit logging documented; tools read-only

    AI say-do 2023-2026: 8 tracked, 4 GA, 1 partial (Ironwood), 3 not delivered (Duet AI / Gemini Cloud Assist for GKE, Cloud Assist investigations, RayTurbo). The AI that ships is in the workload layer (TPUs, DRA, Inference Gateway); the operator assistant is read-only and still preview.

    10. Reading

    Google is building GKE into the place to run very large AI training and inference fleets on its own TPUs and NVIDIA GPUs, while folding Autopilot and fleet features into one product. The release notes show steady weekly delivery, fast Rapid-channel adoption of new Kubernetes minors and most previews reaching GA within a year, but packaging moved twice and the operator AI assistant has stayed in preview since 2023. Buyers get the broadest managed feature set with modest lock-in, at the cost of a per-cluster fee and paid extended support.

    11. Open questions and evidence caveats

    • Two provider-side security flaws with cross-tenant or VPC-wide exposure in 2025-2026 (GCP-2025-001, GCP-2026-058), both fixed by Google.
    • Regular-channel lag was 153 to 296 days for 1.12 to 1.19; it is 42 to 120 days since 2023, with Rapid under a week.
    • Incidents: europe-west9 water leak and fire 2023-04-25, 1.25 stuck-volume defect for 15 days 2023-09, VPC controller automation bug 2024-05-16.
    • No financially backed SLA from 2017-11-28 to 2020-06 while the control plane was free.
    • FedRAMP and other compliance scope for GKE not re-verified in this pass; SecNumCloud is through partner S3NS, not Google itself.

    More from Managed Cloud Kubernetes Ledger 2026