AWS Lambda now allows functions on Managed Instances to run up to 90 minutes, extending serverless capabilities for AI inference and data processing.

AWS Lambda has expanded its runtime limits, now permitting functions on Managed Instances to execute for up to 90 minutes. This change broadens the scope of serverless computing, making it viable for longer‑running tasks such as AI inference, batch data processing, and complex orchestration.

Extended Timeout and Its Impact

The new 90‑minute timeout replaces the previous 15‑minute cap for Managed Instances, aligning Lambda more closely with traditional compute services while retaining its pay‑as‑you‑go pricing model. Developers can now design workflows that keep state within a single function invocation, reducing the need for external coordination.

By allowing longer execution, Lambda also supports emerging use cases like model serving for large language models, video transcoding, and multi‑step ETL pipelines that previously required container‑based solutions.

How the Feature Works

The extended timeout is available on the Managed Instance runtime, which runs Lambda functions on dedicated Amazon EC2 instances managed by AWS. Users configure the timeout via the standard Timeout setting in the Lambda console or through infrastructure‑as‑code tools.

Memory allocation, networking, and IAM permissions remain unchanged, ensuring that existing security and performance configurations continue to apply.

Considerations for Developers

While the longer window opens new possibilities, developers should be mindful of cost implications, as billing continues to accrue per 1‑ms of execution time. It is advisable to monitor function duration and set appropriate alarms to avoid unexpected charges.

  • Use provisioned concurrency for predictable latency
  • Leverage Lambda Layers to share large model binaries
  • Implement graceful shutdown handling for time‑outs

AWS also recommends breaking extremely long jobs into smaller, idempotent steps when possible, to benefit from Lambda’s automatic scaling and fault tolerance.

“The 90‑minute limit makes Lambda a realistic option for workloads that were previously out of reach for serverless.”

For a complete overview of the update, see the InfoQ article.

InfoQ coverage of AWS Lambda’s 90‑minute timeout