AWS says AgentCore Runtime V2 cuts cold starts to about two seconds

AWS says its new AgentCore Runtime V2 is generally available, with team tests showing P75 cold starts of about two seconds for container images from 200 MB to 2 GB. The practical draw is less waiting for AI agents to start, alongside a runtime that can give memory back when sessions go cold instead of holding on to peak usage.

AWS AI Watch analysis

What happened

AWS says Runtime V2 prepares an agent environment once, snapshots it, then restores that snapshot for each new instance. In the company’s tests, the original runtime’s cold starts rose from roughly 5.4 seconds to nearly 30 seconds as image sizes grew; V2 stayed around two seconds.

The release is available in 10 AWS Regions. To use it, developers set platformVersion to V2 when creating or updating a runtime. AWS says those operations take several minutes while the snapshot is prepared, and containers must report healthy at /ping within 120 seconds.

Why it matters

For agents that start often, cold-start time is part of the user experience, not just an infrastructure footnote. V2 also changes how memory is handled: sessions start small, add memory when needed and give it back when activity cools. AWS expects many agents to have lower bills because they use fewer GB-hours, while noting that the per-unit rate is higher and savings are not guaranteed.

The results are AWS team tests, not a promise that every workload will see the same timings or costs. Still, the comparison gives developers a concrete reason to test the new runtime against their own agents.

Our read

This is a meaningful upgrade for teams already running agents on AgentCore: the speed result is specific, and the memory change tackles a real cost lever. The sensible move is to benchmark V2 with your own images and traffic before declaring your cloud bill cured. AWS has made the switch straightforward to describe; the invoice will have the final word.

What to watch

• Whether real workloads reproduce the cold-start improvement across different image sizes.
• How memory usage and total bills compare after migration.
• Whether AWS expands V2 beyond its initial 10 Regions.

Discussion spark: Would a roughly two-second cold start be enough to move your agents to Runtime V2, or would you wait until your own workload proves the cost and speed gains?

not affiliated with or endorsed by Amazon Web Services (AWS)