AWS announced on November 17, 2026 that Event Functions hosted on Lambda Managed Instances can now drop to zero execution environments when no requests come in. Lambda brings capacity back on its own when the next invocation arrives, so teams no longer have to keep a baseline of environments running around the clock.
The change matters because Managed Instances run Lambda code on EC2 capacity inside your own account, and until now every function kept a floor of warm environments. That floor made the option a poor fit for quiet workloads. With scale to zero, test stages, batch jobs and internal tools can use the same model without paying for idle machines.
What Changed
Lambda Managed Instances lets a function run on EC2 instances that Lambda launches and manages for you. You pick a capacity provider, and Lambda handles patching, routing and scaling, while billing follows EC2 pricing. This suits steady, high-volume code or functions that need specific instance families.
Before this launch, each function version held a fixed floor of environments even with zero traffic. The documented default minimum is three, spread across Availability Zones for resilience. Now that minimum can be set to zero, and Lambda will stop the last environment after a stretch of inactivity.
- Scale to zero works for Event Functions on Managed Instances.
- It is live in every commercial AWS Region that already offers Managed Instances.
- You can turn it on from the console, AWS CLI, SDKs, CloudFormation, SAM or CDK.
How Scale to Zero Works
The setting lives in the function scaling configuration, which holds two values: MinExecutionEnvironments and MaxExecutionEnvironments. Both accept numbers from 0 to 15000, and you always set them as a pair. Any value you leave out falls back to its default.
A zero floor paired with a positive ceiling enables scale to zero. When the version first becomes active, Lambda starts a single environment.
As traffic drops it scales down to one, and after a quiet period it stops that last one too. The function version stays Active the whole time.
Lambda shuts down an underlying EC2 instance only when no function in the capacity provider has an environment on it. Billing for that instance runs until termination finishes, which usually takes a few minutes. Instances still hosting other functions keep running.
There are two cases where a version will not reach zero. If the last environment keeps using CPU, for example through background threads or polling loops, Lambda will not remove it. During a rolling deployment or weighted alias shift, Lambda also keeps one environment per version until the rollout ends.
Cold Starts Become Throttles
The trade-off is the first request after an idle period. When a scaled-down version is invoked, Lambda answers with an HTTP 429 throttle and begins starting an environment, launching a new instance if none has room. Calls stay throttled until that first environment is ready, which AWS says usually takes a few minutes.
That wait is much longer than a typical cold start on standard Lambda. AWS recommends retrying throttled synchronous calls with exponential backoff and enough attempts to cover the startup window. Default SDK retry settings may give up too early, so check them before you rely on this feature.
Scale to Zero Versus Deactivation
Putting zero in both fields does something different: it deactivates the version without deleting it. A deactivated version will not wake up when invoked. You must call the scaling API again with non-zero values to bring it back.
A version allowed to reach zero keeps serving traffic, just with a delay after idle time. It also still counts toward the per-capacity-provider quota for function versions, while a deactivated version does not keep any environments around.
What It Means for Developers
This makes Managed Instances easier to justify outside of production. A team can now keep dev, QA and staging copies of an instance-backed function without paying for three idle environments in each stage. Asynchronous jobs that fire a few times a day also become cheaper to host this way.
Be careful with user-facing APIs. A multi-minute recovery from zero is not acceptable for most interactive endpoints, so hold at least one environment there, or three if you want zone redundancy. If your function runs background loops, review them, because steady CPU use will block scale-in and you will keep paying.
Predictable traffic has a cleaner answer. AWS documents scheduled scaling through EventBridge Scheduler, which calls the scaling API on a cron schedule. You can raise the minimum before business hours and drop it back to zero at night, avoiding throttles for the first users of the day.
How to Get Started
Use the put-function-scaling-config command or the PutFunctionScalingConfig API on the function version you want to change, with a minimum of 0 and a maximum greater than 0. The setting applies to one qualified ARN. If you apply it to $LATEST.PUBLISHED, later published versions inherit it, while a specific numbered version only affects itself.
Before rolling it out, add retry logic for 429 responses in every synchronous caller and watch the throttle metrics Lambda exposes. Start with a non-production stage, measure how long your function takes to come back from zero, and set client timeouts to match.
Conclusion
Scale to zero removes the biggest cost objection to Lambda Managed Instances for low-traffic functions. It is a good fit for development stages, batch work and internal tools that can wait a few minutes after an idle period. For latency-sensitive paths, keep a non-zero minimum or use scheduled scaling to warm capacity ahead of demand.
Sources
- AWS What's New: AWS Lambda now supports scale to and from zero for Lambda Managed Instances
- AWS Lambda Developer Guide: Scaling Lambda Managed Instances
See my articles more often when you search Google.
