Independent perspective. Global context.

Technology

AI gets more time. AWS raises the limit.

On-demand Lambda Event Functions can run for 30 minutes instead of 15. Longer tasks gain room, but the change does not make the code faster.

Amazon Web Services office signage in Houston, Texas.
Archive photo: Tony Webster / Wikimedia Commons · Source and rights

AWS has doubled the maximum runtime for on-demand Lambda Event Functions from 15 to 30 minutes. The change, announced on 19 November, gives developers more room for work that cannot finish inside the previous limit.

The extension covers synchronous and asynchronous invocations and event source mappings on the on-demand service. It also applies to synchronous invocations on Lambda Managed Instances. Asynchronous and event-source work on Managed Instances retains its separate 90-minute limit.

A clock can shape an application

A timeout decides how long an individual invocation may run before it is stopped. Developers facing that boundary may have to divide a task, coordinate several steps or move the work to another computing service.

A longer window can simplify some of those decisions. AI requests involving several operations, large media conversions and calculations are examples of work where continuous execution time can matter. It does not mean the underlying operation runs twice as fast.

The limit is configurable. Existing functions do not automatically acquire a longer runtime merely because the ceiling has changed, and related services can have their own timeouts.

More time is also more work to manage

Consider a task that needs 20 minutes and fails near the end. Allowing it to run longer solves one boundary, but not the cost of repeating almost-completed work. Breaking the job into recoverable steps may still be useful.

There is also a capacity question. If requests arrive at the same rate but remain active longer, more work overlaps. A team needs to consider that overlap alongside the new runtime rather than treat the larger number as a performance improvement on its own.

The practical gain is choice. Some jobs can stay within the same event-driven architecture instead of being redesigned around a 15-minute stop. Whether that is the right choice still depends on reliability, cost and the rest of the application.