Amazon Web Services made customer-managed runtimes for AWS Lambda generally available on November 18, 2026. The feature turns the runtime into its own Lambda resource, separate from the functions that use it. A platform team builds the runtime once as a container image, and application teams point their functions at it while still deploying ordinary .zip packages.
This matters for any organization that runs hundreds or thousands of functions with shared tooling baked in. Until now, updating a common SDK, a certificate bundle or a monitoring agent meant redeploying every function that carried it. With the new model, an update to the runtime reaches functions on their next cold start, with staged rollout and rollback built in.
What AWS Changed
Lambda Event Functions (the new name for standard Lambda functions) previously had two broad choices.
They could use a runtime that Lambda manages, such as Python or Node.js, or they could ship their own runtime inside the deployment package or a layer. AWS now renames that older do-it-yourself approach to "bring your own language" and recommends the new managed resource for most cases.
The key change is the split between two roles. The runtime publisher owns the runtime contents, updates, access rules and lifecycle. On the other side sits the function developer, who only selects the runtime and writes handler code.
- Each runtime is an OCI image kept in an Amazon ECR repository, so existing Docker and CI/CD pipelines work.
- A function references a runtime by alias and moves to a new version when Lambda recycles execution environments.
- Rollouts can be timed and phased, with health metrics and fast rollback.
- Runtimes can be shared with other accounts, whole AWS Organizations or OUs, or made public.
- The feature is available in all commercial AWS Regions at no extra charge.
How the Resource Model Works
AWS describes three resource types. A runtime is the top-level object with its own ARN and a resource-based policy that controls who can use it. A runtime version captures the contents at one point in time, is backed by a single ECR image, and cannot be modified after creation.
A runtime alias is a named pointer to a version, for example production, stable or php8.4.
Moving an alias to a newer version is how a publisher rolls out a patch, and moving it back is how they undo one. Runtimes are Regional, so AWS suggests keeping the same names in every Region and rolling out alias changes one Region at a time.
On the consumer side, the function configuration sets the runtime to custom and passes the runtime ARN plus the alias name. Logging, layers, monitoring and infrastructure-as-code tools behave as before. Each image also needs a startup script plus the Lambda client component that pulls invocations and hands them to your code.
Monitoring, Trust and Deletion Guardrails
Lambda publishes two CloudWatch metrics in the publisher's account, RuntimeInvocations and RuntimeInvocationErrors, broken down by runtime, version or alias. AWS suggests comparing the error rate of a new version against the old one during rollout and setting an alarm that halts the pipeline. Lambda does not patch these runtimes itself, so the publisher must rebuild from updated base images.
Because a publisher can push code into every function that uses its runtime, AWS stresses that consumers must trust the publisher as they would trust someone running code with their permissions. Organization admins can limit allowed runtimes with service control policies. Cross-account consumers also need the lambda:AssociateRuntimeAlias permission in an identity-based policy.
- Deleting a runtime, version or alias is scheduled with a waiting period of 7 to 30 days, 30 by default, and can be reversed.
- A version that an alias still points to cannot be deleted.
- Functions using a deleted alias or version go Inactive and fail on cold start, then recover automatically if the resource is restored.
- Publishers can mark a runtime as Preview, GenerallyAvailable or Deprecated and block new function creation after a set date.
What It Means for Developers
For platform and security teams, this is the biggest change to Lambda packaging in years.
Shared components such as internal SDKs, corporate CA certificates and observability agents can now be patched centrally, which shortens the window for fixing a critical library flaw. It also gives a clean path for languages Lambda does not offer natively, such as PHP, Bun or R, while keeping the familiar .zip workflow for developers.
Another practical use is extending a managed runtime past its deprecation date by forking it into your own runtime. That buys time for migrations, but it also moves patching duty onto your team, so plan for it. Treat runtime updates like library upgrades: test them in a dev alias first and watch error rates closely before moving the production alias.
Teams already using container image functions should weigh the trade-off. Container images give full control but force every team to rebuild for each shared fix, whereas a shared runtime handles that once.
If you are also exploring recent Lambda additions such as durable functions that can pause and resume, the runtime model works alongside them as a separate layer of control.
How to Get Started
Package the runtime as an image in ECR, starting either from an AWS-provided Lambda base or from an empty image, and allow the Lambda service to pull images from that repository. Then create the runtime, a first version that references the image, and an alias. Application teams set their functions to use that alias and keep deploying .zip files.
Management is available in the console and through the CLI, SDKs, CloudFormation or CDK. Pricing stays at standard Lambda rates for function execution, while runtime images incur normal ECR storage charges for as long as each version exists. Note that the feature applies to Event Functions and not to Lambda Web Functions.
Conclusion
Customer-managed runtimes separate runtime ownership from function code, letting one team patch shared components across a whole organization without touching individual deployments. The trade-off is that the publisher now carries patching and trust responsibilities that Lambda used to hold. For large serverless estates, that is likely a welcome shift.
Sources
- AWS What's New: AWS Lambda now supports customer-managed runtimes
- AWS Compute Blog: Control your AWS Lambda runtimes with customer-managed runtimes
- AWS Lambda Developer Guide: Customer-managed runtimes
See my articles more often when you search Google.
