On October 23, 2026 (Beijing time), the AWS Open Source Blog published a post by Shadaj Laddad introducing Infinity. It is open-sourced under Apache 2.0 as part of Hydro, a collaboration between AWS and UC Berkeley. The post frames Infinity as a research runtime for extreme agent concurrency: when an agent calls a tool, Infinity dispatches the call, saves the conversation, and suspends the agent; when the tool result arrives, it resumes from that saved state.
Because a suspended agent occupies little beyond its history, many agents can stay resident, and it can run on serverless: when Infinity suspends during a tool call, the serverless function can end, so nothing needs to keep running while a tool takes minutes or days. The blog also briefly covers synthetic tool calls for delivering async events into model-friendly conversation shape, a step mode for batch deliveries, and a reference AWS deployment using SQS FIFO, Lambda, Aurora DSQL, and DynamoDB.
On memory, AWS reports its own benchmark numbers—not a general SLA: about 100 KB per idle agent, and at that rate about 75,000 concurrent agents fit in 8 GB (Raspberry Pi 5 memory class). The run used 80,000 agents on a single thread of an AMD EPYC 9R14 with in-memory stores. The post calls Infinity an early research release and asks for feedback.
Infinity is agent runtime infrastructure in that suspend/resume sense. It is not the same as Agents API Scheduled Automations status quirks. As for whether those concurrency figures hold on your workloads, or how the research runtime behaves in practice—we don’t have that information yet. If you’ve used it, you can send feedback through the site-owner contact link in the footer.