AWS Extends Lambda Managed Instance Jobs to 90 Minutes
Asynchronous Lambda Managed Instance invocations can now run six times longer. The extension reduces forced rewrites, but long-running work still needs idempotency and recovery design.
AWS announced on September 9, 2026 that Lambda functions running on Lambda Managed Instances can use a timeout of up to 90 minutes for asynchronous and event source mapping invocations. The previous per-invocation ceiling was 15 minutes.
What changed
The new limit is 5,400 seconds and applies only to Lambda Managed Instances in the supported invocation modes. Synchronous invocations retain the 15-minute maximum. AWS says existing runtimes, handlers, IAM roles, VPC settings and observability integrations continue to work.
The feature can be combined with Lambda durable functions. A durable execution may span much longer, while the 90-minute setting controls each individual invocation. AWS states there is no separate fee for the longer timeout; standard Lambda Managed Instances pricing applies.
Why it matters
Data processing, media work, AI inference and batch jobs often exceed 15 minutes. Teams previously had to split work, move it to another compute service or accept a more complex orchestration design solely because of the timeout.
A longer ceiling removes that forced boundary. It does not make a 90-minute process automatically resilient. A host failure late in a job can still waste significant work unless the task is idempotent or checkpointed.
Design checks
- Confirm the invocation is asynchronous or event-source driven.
- Make queue visibility timeouts and retry behavior consistent with the function timeout.
- Use idempotency keys when a repeated message could duplicate side effects.
- Add checkpoints when replaying from the beginning is expensive.
- Measure steady-state cost against containers, batch services or dedicated workers.
- Set an intentional timeout rather than using 90 minutes as a default.
BronzeDev view
This change widens the useful serverless workload envelope, especially for teams already operating event-driven systems. Architecture should still be chosen around failure recovery, concurrency and cost—not only the maximum runtime shown in a console.
Primary source
AWS Compute Blog: Announcing 90-minute function timeout on AWS Lambda Managed Instances
