Aurora Serverless now scales PostgreSQL to 384 ACUs, about 768 GiB of memory

Victoria Shinder29 October 2026, 09:191 min read

Amazon Aurora Serverless can now scale an Aurora PostgreSQL-Compatible database to 384 Aurora Capacity Units, up from a maximum of 256, AWS announced. Each ACU provides about 2 GiB of memory with matching CPU and networking, so a single serverless database can reach roughly 768 GiB of memory, and it still scales down to zero when idle.

The higher ceiling has conditions. It requires Platform Version 4 and one of five engine versions: Aurora PostgreSQL 18.4, 17.10, 16.14, 15.18 or 14.23. New clusters can set a maximum of up to 384 ACUs directly. Existing clusters already on Platform Version 4 change the maximum-capacity setting, and the engine version if needed, with no downtime. Clusters on older platform versions must first move to Platform Version 4 through a pending maintenance action. The announcement covers the PostgreSQL edition only.

AWS pitches the change at agentic AI workloads, which it describes as bursty, with long idle periods and unpredictable traffic.

What it means

Serverless capacity caps used to be the point where teams moved to provisioned instances for their largest databases. A 768 GiB ceiling pushes that point out considerably. The catch is in the prerequisites: the platform-version upgrade on older clusters is a maintenance action, so plan for it before relying on the new limit, and confirm the engine version, because only the five named releases qualify.