io.net, the world’s largest decentralised GPU network, launched Agent Compute, a first-of-its-kind platform that lets AI agents provision their own computing infrastructure. Agents can independently spin up GPU clusters, run workloads, and scale resources up or down – without the enterprise onboarding and procurement processes that have kept cloud computing out of reach for smaller teams.
“The current cloud model is built for enterprise budgets,” said Gaurav Sharma, CEO of io.net. “If you’re a startup or solo developer building with AI agents, you’re either paying prohibitive hyperscaler rates or spending weeks navigating procurement. Agent Compute removes that barrier. An agent can independently find the best-priced GPU for the job, provision it, and manage the infrastructure end-to-end – so developers can spend their time building, not comparing cloud pricing or configuring servers.”
AWS and Google Cloud require lengthy onboarding, complex billing, and minimum commitments that price out small teams, startups and developers. Agent Compute sidesteps that entirely. Agents interact directly with io.net‘s marketplace of over 10,000 GPUs across 138 regions in 130+ countries – accessing compute at up to 70% lower cost than traditional cloud providers, provisioning resources on demand and releasing them when done.
The system uses the Model Context Protocol (MCP), which gives agents clear visibility into available compute resources – GPUs, costs, and specs – to make informed decisions rather than costly mistakes. Without these guardrails, AI agents have already caused real damage: Amazon’s AI shopping agent triggered a 13-hour outage after deleting a production environment, and OpenClaw users reported bills exceeding $3,600 a month from runaway workflows.
In practice, an agent analysing data can spin up a GPU cluster, process the job, and terminate resources when finished. No manual setup. No leftover costs. Developers can set spending limits and resource caps to stay in control.
“This is a step towards truly autonomous agents,” Sharma said. “Right now, agents still depend on humans for infrastructure. As they become more capable, that dependency becomes the bottleneck. If agents are going to operate independently – making decisions, executing tasks, scaling resources in real time – they need the ability to provision their own compute. That’s the future we’re building toward.”
The implications go beyond developer convenience. As agents gain the ability to manage their own compute requirements, the bottleneck shifts from infrastructure access to imagination. A solo developer can now build systems that were previously only possible with enterprise resources – agents that scale across continents, process massive datasets, or deploy models on demand. It’s a glimpse of a future where the limiting factor isn’t budget or team size, but what you can think to build.
The feature is available now in early access, with broader rollout to follow.