YC-backed machine0 launches persistent VM platform for AI agents
The new CLI-driven service provides programmable CPU and GPU compute designed for long-horizon autonomous workloads.
Barnaby Malet has launched machine0, a CLI-driven cloud platform providing persistent virtual machines (VMs) specifically engineered for long-horizon AI agents. Part of the Y Combinator S26 batch, the service allows both human users and autonomous agents to provision, snapshot, and manage dedicated compute resources via a terminal or the Model Context Protocol (MCP).
Technical specifications include CPU VMs with up to 60 vCPUs and 240 GB of RAM, with a claimed 99.99% VM-level uptime. For compute-intensive AI workloads, machine0 supports a broad array of GPUs, including the H100, H200, L40S, MI300X, and RTX 4000/6000 Ada. Every VM is equipped with a static IP and an HTTPS endpoint at <vm>.mac0.io. To ensure environment flexibility, the platform supports Ubuntu—pre-installed with Docker, Node, Python, Claude Code, and Codex—as well as NixOS for reproducible builds via flakes. Pricing follows a pay-as-you-go model billed by the minute, with CPU VMs starting at $0.013 per hour and GPU VMs starting from $0.836 per hour.
The shift to agentic infrastructure
This launch comes as the AI industry shifts from ephemeral, single-turn tasks toward long-running workloads. Modern coding agents and autonomous systems often require environments that persist for several hours or even operate 24/7. Traditionally, these environments required manual setup by human operators, creating a bottleneck for agents that need to maintain state or run complex background processes over extended periods.
Why programmable compute matters
By treating the VM as a programmable resource, machine0 lowers the barrier for deploying "agentic" infrastructure. The integration with MCP and a dedicated CLI enables a closed loop where agents can independently provision their own compute and configure their environments through code. According to founder Barnaby Malet, the platform is specifically designed for this non-human user base, stating, "Most users have one thing in common: they're not humans, they're agents. And machine0 is built to be used by them."
What to watch
As AI agents gain more autonomy over their own infrastructure, the industry will likely see a move toward more isolated, reproducible environments to mitigate security risks. The adoption of NixOS on machine0 suggests a push toward "infrastructure as code" for agents, ensuring that an agent's environment can be perfectly replicated across different instances. The primary metric for success will be how effectively these agents can manage their own resource lifecycles without human intervention.