machine0
machine0 rents persistent cloud virtual machines built specifically to run long-lived AI agent workloads, billed by the minute.
🔗 Visit machine0
Description
If you've ever tried to leave an AI agent running unattended for hours — coding overnight, monitoring something, working through a long task queue — you've probably hit the problem machine0 is built for: most cloud sandboxes are made to spin up, do one thing, and disappear, not to stay alive and stateful for days. machine0 rents you a normal virtual machine that just keeps running, with your files, processes and installed tools intact, so an agent can pick up exactly where it left off instead of starting from a blank slate every time.
Technically, machine0 offers persistent CPU and GPU VMs with snapshotting/cloning, static IPs, HTTPS endpoints, and reproducible provisioning via NixOS flakes or Ansible — aimed squarely at teams running agent frameworks, MCP servers, or Claude Code-style workflows that need real infrastructure rather than an ephemeral sandbox. Billing is per-minute: CPU instances start around $0.013/hour (roughly $9/month if left running continuously), GPU instances range from $0.836/hour up to $39.336/hour depending on the card, and a suspended (stopped but not deleted) VM only costs storage at $0.078/GB/month. It's YC S26-backed, ships a CLI for scripted control, and supports five global regions.
💬 Our review
The short version: if you're building or operating AI agents that need to run for hours or days at a time — not just answer one prompt and exit — machine0 gives you a real, persistent VM instead of a throwaway sandbox, and prices it fairly at true pay-per-minute rates with a cheap suspend option for idle time. It's a good fit specifically because most "AI infra" cloud products (E2B, Modal) are optimized for short-lived, stateless execution, which is the opposite of what long-running agents need.
The honest caveat is that machine0 is infrastructure, not a platform — you still provision, configure, and manage the box yourself (NixOS/Ansible knowledge helps), and there's no built-in agent orchestration, monitoring dashboard, or team collaboration layer the way a more opinionated agent platform might offer. It also competes directly with general-purpose GPU/VM markets like Vast.ai, RunPod, and Lambda Labs on raw compute pricing, so it wins mainly on the persistence and reproducibility angle, not necessarily on being the cheapest GPU-hour on the market — worth comparing actual instance pricing for your specific card before committing.
💰 Pricing
📊 Global score
🤖 AI-enriched data
Facturation à la minute. CPU dès 0,013$/h (~9$/mois en continu). GPU de 0,836$/h à 39,336$/h selon la carte. VM suspendue (arrêtée) : stockage seul à 0,078$/Go/mois.
Pros
VMs vraiment persistantes (contrairement aux sandboxes éphémères E2B/Modal)
Facturation à la minute avec option suspension quasi-gratuite (stockage seul)
Provisioning reproductible via NixOS flakes / Ansible
IP statiques, endpoints HTTPS, snapshots/clonage de VM
Cons
Infrastructure brute — pas d'orchestration d'agents ni de dashboard intégré
Nécessite des compétences DevOps (NixOS/Ansible) pour en tirer parti
Concurrence directe sur le prix pur du GPU/heure face à RunPod, Vast.ai