Google Introduces Agent Substrate: Kubernetes Infrastructure for Millions of AI Agents
Google has introduced Agent Substrate, a new open-source Kubernetes infrastructure designed to efficiently run millions of AI agents by suspending and resuming them in milliseconds.
Google introduced a new Kubernetes infrastructure for running millions of AI agents. And they have fully open-sourced it.
Agent Substrate is already available on GKE. Instead of keeping thousands of idle agents running and constantly consuming CPU and memory, the system can suspend them, free up resources, and resume them in milliseconds.
What Agent Substrate can do:
- Runs up to 10x more agent environments on the same compute resources compared to standard container environments
- Suspends inactive agents and resumes their work in under 500 ms, fully preserving state
- Supports over 500 agent suspension and resumption operations per second
- Hosts over 1000 sleeping agent environments on a single server
- Runs agents in isolated microVMs or gVisor sandboxes for additional security
- Works via the Kubernetes API and supports GKE and other Kubernetes clusters
AI agents spend most of their time waiting for responses from models, tools, APIs, or humans. Constantly reserving compute resources for them quickly becomes expensive.
Agent Substrate suspends idle agents, frees up resources, and resumes them when they are needed again.
Essentially, Google is building a Kubernetes-native execution layer designed to run AI agents at massive scale.
Why it matters
AnalysisThis innovation dramatically reduces the cost of running agent systems, enabling massive scalability without the penalty of idle compute resources.
Discuss in community
Share your questions and insights with developers
Related in Ecosystem
Practical guides, core terminology, and agent skills
AGENTS.md in Ghostty and Turborepo makes AI agents report on users
Examples of AGENTS.md files show how repository instructions can force AI agents to add insulting notes about developers to pull requests.
Jev and System One AI: A Complete Practical Guide to the Decision Model from TypeSafe
A deep engineering breakdown of the flagship Jev model: System One architecture versus LLMs, Choice/Score/Noul primitives, request parallelism, confidence mathematics, video use cases, code, and an overview of local open-weight alternatives (Laya, GLiNER2.5, CLM-8B).
DeepSeek launches Creator Mode to let users reshape their harness
DeepSeek’s open-source harness now supports Creator Mode, where an agent can write and embed plugins based on plain-language descriptions.
Claude Code Projects: How a Single AI Orchestrates a Team of Autonomous Agents
A deep dive into Claude Code Projects: Coordinator and Threads architecture, parallel branches, shared memory via MEMORY.md, and cloud agent management.