Container runtimes in 2026: what was measured, what was surveyed, and what is missing
Figures for "containerd’s share in 2026" circulate, but no measurement stands behind them: the last telemetry is November 2023 and the series has been discontinued. The CNCF survey of September 2025, source of the "containerd 74%" figure, asks no runtime question at all.
There is, however, a fact stronger than any percentage: on managed Kubernetes across all three major clouds, Docker Engine is technically unavailable as a runtime from version 1.24. Below: what is solidly known, and five circulating figures that appear in no report.
Free online tool — HTTP header checker: instant results, no signup.
No 2026 measurement of runtime shares exists
Worth saying at once, because "containerd’s share in 2026" figures do circulate while no measurement stands behind them.
The last telemetry is Datadog’s report marked November 2023: over 2.4 billion containers across Datadog customers, containerd in 53% of organisations (23% a year earlier), Docker in 65% (down from 88%). The total exceeds a hundred because an organisation can run several runtimes at once. The series has been discontinued: no 2024, 2025 or 2026 edition was published.
Sysdig’s 2026 report (published 16 April) contains no runtime telemetry at all — the section that carried this data until the 2023 edition has been removed entirely. Image sizes, base images, layer counts and container density went with it.
The last survey is CNCF’s, fielded September 2025 and published January 2026, with 628 respondents. That is self-report rather than measurement, and recruitment ran through CNCF’s own community.
Why CNCF’s figures cannot be read as market shares
The usual citation from CNCF is "containerd 74%, cri-o 34%". Those numbers are real, but they come from the question "which graduated CNCF projects is your organization using in production or evaluating?" — not "what runtime does your cluster use". There is no runtime question in the survey at all.
Three things follow:
- Docker Engine is not a CNCF project, so it physically cannot appear in that question. The word "Docker" occurs once in the whole report, in a passing phrase, with no percentage attached.
- These are not shares: an organisation may tick both containerd and cri-o.
- Comparing containerd against Docker Engine from CNCF data is impossible. Anyone doing so is filling in the blanks.
What the report does establish: containers for most or all production applications at 56% against 41% in 2023; Kubernetes in production among container users at 82% against 66%.
A fact stronger than any survey: Docker Engine cannot be chosen
There are no 2026 shares, but there is a documented circumstance that answers the question better than surveys do: on all three major clouds, Docker Engine is technically unavailable as a runtime.
| Platform | State |
|---|---|
| Kubernetes | dockershim deprecated in 1.20 and removed in 1.24 (May 2022) |
| Google GKE | from 1.24 only containerd node images are supported; remaining Docker pools were migrated by force from 1 September 2023 |
| Amazon EKS | from 1.24 the Amazon-provided node images support containerd only (announced 15 November 2022) |
| Microsoft AKS | containerd on Linux from 1.19; on Windows Server it is the only option from 1.23 |
This is stronger than percentages: the question is not how many people say they migrated, but that managed Kubernetes no longer offers the choice.
Docker as a developer tool is a separate quantity, and the two get confused. By the Stack Overflow survey (fielded 29 May – 23 June 2025, 49,009 responses) Docker is used by 71.1% of respondents, Kubernetes by 28.5%, Podman by 11.1%. But "I build images with Docker" coexists perfectly well with containerd in the cluster — that is a tool on a desk, not a runtime in production.
Figures in circulation that are not in the reports
Checking against the full report texts produces an unpleasant list:
| Claim | What is actually the case |
|---|---|
| "containerd is the default runtime in 95% of clusters, CNCF 2025" | the CNCF report contains no "95%" line at all |
| "34% of organisations run more than one runtime, CNCF 2025" | absent from the report |
| "53% containerd, per CNCF 2025" | that is Datadog telemetry from November 2023, attributed to the wrong source and aged by three years |
| "Stack Overflow 2026: Docker 71.1%" | the 2026 survey results do not exist, the page returns 404; 71.1% is 2025 |
| "Sysdig: image sizes grew fivefold" | the report says bloat — the share of unused packages — grew fivefold, not megabytes; Sysdig has never measured image size |
The common signature: none of these figures names a collection period, a population or a unit of count. If any of the three is missing, the figure is not worth carrying over.
Sources: CNCF survey, Datadog report, Kubernetes on the dockershim removal.
Learn more
Frequently Asked Questions
Should I migrate dockerd → containerd?
Managed K8s (GKE, EKS, Yandex) already all on containerd. Self-hosted 1.24+ too. If on 1.22/1.23 — yes, migrate.
containerd vs CRI-O for self-hosted?
containerd — universal, easier to get CNCF support. CRI-O — best if you are on Red Hat OpenShift.
Podman vs Docker Desktop?
Podman — daemonless, rootless by default. Free. Docker Desktop — polished UI, paid for enterprises. Podman better for CI/CD containers.
Monitor K8s runtime?
Enterno HTTP for ingress endpoints. Security scan for exposed kubelet.
Try the live tool that powered this guide
Free plan — 10 monitors, checks every 5 min, no card required. Upgrade for 1-minute interval and multi-region monitoring.