AWS Expands Bedrock AgentCore Compute for Faster AI Agents
AWS announced on Thursday that the runtime layer of Amazon Bedrock AgentCore now supports up to four times more compute than the prior version. Each session can use up to eight virtual CPUs and 32 GB of memory, and container images up to 10 GB are accepted, allowing developers to run larger, more demanding agent workloads such as extensive code compilations, browser-driven tasks, and local model inference.
The update also improves memory efficiency. Unused memory is reclaimed within minutes after an agent releases it, and idle agents can scale to zero. For multi-turn agents that pause between interactions, the changes can lower memory consumption per session by as much as 68%, reducing overall operating costs for agents that spend time waiting on model responses, tool calls, or human input.
AWS introduced baseline pricing to support always-on agents. A minimum memory allocation per session is set, while workloads may burst up to four times that baseline when needed, giving enterprises predictable costs without paying for peak capacity in advance. The runtime now also extends isolation to data, letting developers dynamically mount Amazon S3 buckets via FUSE, eliminating the need for custom mechanisms to move files, credentials, and tool outputs between the isolated environment and external storage.
Industry analysts note that the larger session size and sub-two-second start-up times simplify development. Ashish Chaturvedi of HFS Research said the new limits allow complex workloads to stay within the secure boundary, removing the need to split tasks or move them outside the environment. Faster start-up also means developers can provision sessions on demand instead of maintaining warm pools of idle capacity, further cutting costs.
Why it matters
Enterprises can now deploy more sophisticated AI agents at lower cost and with predictable pricing, accelerating AI-driven automation projects.
FAQ
What compute resources does the new AgentCore runtime provide per session?
Each session can now use up to eight virtual CPUs and 32 GB of memory.
How large can container images be in the updated runtime?
Container images up to 10 GB are supported.
What memory savings can multi-turn agents expect?
Memory consumption per session can be reduced by up to 68%.
What pricing model does AWS introduce for always-on agents?
Baseline pricing sets a minimum memory allocation per session with the ability to burst up to four times that baseline.
Source: InfoWorld. See the original source for full details.
Scan with your phone
Scan with your phone