AWS Extends Lambda Managed Instances to 90-Minute Runs

A longer runway for Lambda workloads

AWS has expanded the execution window for functions running on Lambda Managed Instances to 90 minutes. The new ceiling is six times the former 15-minute limit and gives teams another option for tasks that need sustained processing without shifting immediately to separately managed infrastructure. The change applies where Lambda Managed Instances are offered, while the timeout for conventional synchronous Lambda requests stays at 15 minutes.

The prior cap had become a practical boundary for work that could not reliably finish within a short invocation. Examples cited around the change include media jobs, financial computations, ETL and broader data-processing runs, AI inference, scraping activity, and large-file movement. For these workloads, developers have often needed additional mechanisms to break up, coordinate, or otherwise accommodate work that exceeded a single Lambda run. The longer window may allow some of that supporting complexity to be removed.

But a longer timeout is not simply a matter of letting existing code run for more time. AWS stresses that applications must account for connections and temporary credentials that could expire during a lengthy execution. Teams also need to plan for retries and duplicate processing. Lambda does not provide exactly-once processing, and the possibility of another delivery or rerun becomes more significant as the operating window grows.

That makes idempotency a central design consideration. Operations with consequences outside the function, such as charging a payment or updating a database record, should produce a safe outcome if they are performed more than once. AWS points developers to Powertools for AWS Lambda as one way to add idempotency handling in application code. Durable Functions similarly require workloads to tolerate steps being rerun after a failure.

The announcement further differentiates the forms Lambda now offers. Event-driven functions retain their 15-minute limit. MicroVMs, aimed at user- or AI-generated code, can run for as long as eight hours. Managed Instances sit between those models for steady-state work: they can serve multiple requests on an instance and offer EC2-based compute choices and pricing without requiring users to manage the underlying infrastructure.

Reaction reflects both the appeal and the boundaries of the new capability. Supporters see an opportunity to retire workaround-heavy designs created by the old limit, including in agentic workflow scenarios. Others caution against treating 90 minutes as a default architecture. If a function spends substantial time waiting rather than doing productive computation, Lambda economics may be less attractive than ECS Tasks or AWS Batch. The decision therefore remains workload-specific: the expanded timeout broadens the available design space, but it also raises the importance of reliability discipline and cost evaluation.

The release follows another Managed Instances update, support for Graviton5-powered EC2 instances. Together, the changes position this Lambda option for a wider range of sustained workloads while preserving a distinct model from standard short-lived function invocations.