AWS launches EC2-backed AgentCore runtime instances for 14-day sessions
Amazon Bedrock AgentCore now offers runtime instances, an EC2-backed compute option for long-running, hardware-specific, and multi-agent workloads. Sessions can run for up to 14 days, compared with up to eight hours for the existing microVM option, while AWS manages provisioning, patching, scaling, and teardown.
AWS made AgentCore runtime instances generally available on August 6, adding an EC2-backed compute option to Amazon Bedrock AgentCore. The launch targets agents that need longer sessions, specialized hardware, shared local state, or multiple collaborating processes on one host.
The new option complements AgentCore's existing microVM runtime. AWS says microVM sessions can run for up to eight hours, while runtime-instance sessions can run for up to 14 days. Both use the same AgentCore invocation surface, but the underlying compute and operating model differ.
Managed EC2 in the customer's account
Runtime instances use Amazon EC2 managed instances provisioned inside the customer's AWS account. Teams create a capacity provider that defines allowed instance types, networking, storage, and the infrastructure role. AgentCore then handles provisioning, operating-system and runtime patching, scaling, and teardown.
The launch supports Linux on x86_64 and Arm, code packages or container images, and selected GPU and Inferentia2 instance families. AWS also documents persistent EBS-backed storage across session stops. Multiple agent runtimes tied to the same capacity provider can use one session ID to share an instance and filesystem, allowing a writer, reviewer, test runner, or other specialist agents to work against the same artifacts.
Where it changes the architecture decision
This is not simply a longer serverless timeout. Runtime instances trade the microVM model's rapid, consumption-based scaling for more direct control over EC2 capacity, operating environment, persistent volumes, and hardware selection. They are aimed at multi-day automation, compilation, security scanning, media processing, specialized inference, and other workloads that need sustained state or compute.
For shorter API-driven agents, AWS says the existing microVM runtime is generally the better fit. Runtime-instance customers pay the underlying EC2 and storage charges plus an AgentCore management fee, so idle capacity and retained persistent resources need explicit lifecycle controls. The practical evaluation is therefore workload-specific: session duration, startup tolerance, hardware requirements, collaboration model, and utilization determine which runtime is the better fit.
Key Points
- 1AgentCore runtime instances use AWS-managed EC2 infrastructure in the customer's account and support sessions lasting up to 14 days.
- 2Capacity providers define instance types, networking, storage, and infrastructure roles; multiple agents can share one session and filesystem.
- 3The EC2-backed option adds specialized hardware and persistent compute but also adds EC2, storage, and AgentCore management costs.
Scoring Rationale
The launch materially expands AgentCore from short-lived isolated microVM sessions to managed EC2 capacity for multi-day, multi-agent, and specialized-hardware workloads. It is directly relevant to teams choosing production agent infrastructure, though its practical value depends on workload duration and utilization.
Sources
Primary source and supporting public references used for this report.
Practice with real Retail & eCommerce data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card
See all Retail & eCommerce problems
