io.net announced Agent Cloud, describing it as MCP-native compute with sub-3-minute provisioning and pay-for-work billing, where AI agents can choose hardware, run jobs, pay through Stripe and x402, and clean up resources automatically.
The release gives io.net a direct product for autonomous AI-compute workflows rather than a broader infrastructure story alone. By letting agents provision compute quickly, pay only for completed work, and shut resources down on their own, it is aimed at bursty workloads that do not fit slower manual cloud processes.
That can support io.net's utility narrative around decentralized GPU infrastructure and AI agents, because the product ties compute access, task execution, and payment into one flow. A limitation is that the announcement confirms availability, but it does not show how widely agents or developers are using it yet.
AI agents don't work 9-5.
They fan out, burst, idle, and burst again.
But hyperscalers aren't built for that. They are built for humans who file tickets and wait 8 minutes for a cold start.
Agent Cloud fixes that:
- MCP-native. Infra is just another tool call.
- Sub-3-min…