machine0 - Persistent CPU and GPU VMs from the CLI

TL;DR

TL;DR: machine0 is a CLI-first cloud VM service with up to 60 vCPUs, 240 GB RAM, and H100/H200 GPU options — built for AI agents that need persistent, resumable infrastructure billed by the minute.

Source and Accuracy Notes

⚠️ This section is MANDATORY. All links must be verified from actual source, not guessed.

• Project page: machine0.io — verified via direct fetch
• Source repository: github.com/fdmtl/machine0-nixos and github.com/fdmtl/machine0-ubuntu — open-source VM images
• HN launch thread: news.ycombinator.com/item?id=49348136 — verified YC S26 batch
• License: Proprietary SaaS (VM images are open source, control plane is not)

What Is machine0?

machine0 is a Y Combinator S26-backed cloud infrastructure service that lets you spin up persistent virtual machines directly from the CLI. Unlike ephemeral cloud dev environments, machines stay alive until you explicitly stop them — billed per minute rather than per hour.

The product is designed around the AI agent workflow: every operation is a CLI command with --json output, and a remote MCP server lets code agents like Claude Code and Codex drive everything programmatically. VMs survive restarts, get static IPs and HTTPS endpoints at [name].mac0.io, and can be suspended to freeze state and pause billing.

Setup

Install the CLI

curl -LsSf https://machine0.io/install.sh | sh

After installation, authenticate:

machine0 auth login

Create Your First VM

# Spin up a 4 vCPU, 8 GB RAM machine in us-east
machine0 new myvm --size xxl --region us-east

# SSH in immediately
machine0 ssh myvm

Configure MCP Integration

Agents like Claude Code and Codex auto-discover MCP servers injected via profiles:

# Add a profile with your MCPs and credentials
machine0 integrations connect claude-code --profile default
machine0 integrations add linear --url https://mcp.linear.app/mcp --profile default

# Create a VM with that profile — Claude Code inherits everything
machine0 new agentbox --profile default

Suspend and Resume

# Freeze state and stop billing
machine0 suspend myvm

# Restore later
machine0 start myvm

Instance Options

machine0 offers CPU and GPU instance sizes. All are billed per minute with no minimum top-up below $5.

CPU Instances

| Size | vCPU | RAM | Disk | $/hour | |----------|------|-------|-------|--------| | small | 1 | 1 GB | 25 GB | $0.013 | | medium | 2 | 2 GB | 60 GB | $0.034 | | large | 2 | 4 GB | 80 GB | $0.052 | | xxl | 8 | 16 GB | 320 GB| $0.208 | | 4xl | 32 | 128 GB| 480 GB| $1.980 | | 6xl | 60 | 240 GB| 900 GB| $3.714 |

The -nvme variants use newer-generation CPUs with NVMe storage for higher disk IOPS at the same vCPU/RAM. The -premium sizes use dedicated (non-shared) vCPUs for faster single-thread performance.

GPU Instances

| Size | GPU | VRAM | vCPU | RAM | $/hour | |----------------|-----------|----------|------|-------|--------| | gpu-4000ada-1 | 1x RTX 4000 Ada | 20 GB | 8 | 32 GB | $0.836 | | gpu-l40s-1 | 1x L40S | 48 GB | 8 | 64 GB | $1.727 | | gpu-h100-1 | 1x H100 | 80 GB | 20 | 240 GB| $4.851 | | gpu-h200-1 | 1x H200 | 141 GB | 24 | 240 GB| $4.917 | | gpu-h100-8 | 8x H100 | 640 GB | 160 | 1.9 TB| $38.808 |

GPU instances are available in us-east, uk, eu, and asia regions.

Key Features

Persistent VMs. Unlike cloud dev environments that terminate when idle, machine0 VMs run until you suspend or delete them. State survives reboots.

Profile-based agent configuration. Inject MCP servers, credentials, environment variables, and prompts into VMs via named profiles. Claude Code and Codex pick them up automatically on SSH.

Reproducible builds with NixOS. NixOS images are fully deterministic — one-command rollbacks, no configuration drift. Ubuntu images come pre-installed with Docker, Node, Python, Claude Code, and Codex.

Free suspend. Run a task, machine0 suspend, and billing stops entirely. You only pay for image storage ($0.078/GB/month) while suspended.

5 regions. us-east, uk, eu, asia, plus a default region. Static IPs and HTTPS endpoints ([vm].mac0.io) on every machine.

FAQ

Q: Is the control plane open source? A: No. The VM images (NixOS and Ubuntu configurations) are open source at github.com/fdmtl/machine0-nixos and github.com/fdmtl/machine0-ubuntu, but the cloud infrastructure and orchestration layer is proprietary.

Q: How does suspend billing work? A: When you run machine0 suspend, the VM state is frozen and compute billing stops. You continue to pay for image storage at $0.078/GB/month. There is no charge for the suspend/resume cycle itself.

Q: What makes this agent-native? A: Every CLI command outputs --json for structured parsing. The remote MCP server exposes all machine0 operations as MCP tools, so agents can create, configure, and destroy VMs programmatically without parsing terminal output.

Q: Are dedicated (non-shared) vCPUs available? A: Yes. The -premium sizes (xl-premium, xxl-premium) use dedicated vCPUs with faster single-thread performance. All other sizes use shared vCPU pools.

Conclusion

machine0 targets a specific gap in cloud infrastructure: AI agents and CLI-first developers who need persistent, GPU-capable VMs without the overhead of managing a cloud dashboard. Per-minute billing, suspend-to-billing-stop, and MCP integration distinguish it from both ephemeral dev environments and traditional cloud VMs. The NixOS image reproducibility is a bonus for teams that want deterministic infrastructure definition in code.

For agents running long tasks — automated research, code generation at scale, model inference — machine0’s model is worth evaluating against a租用 GPU cloud like RunPod or Lambda Labs, which bill similarly per second but lack the CLI-first agent UX and suspend model.

Source and Accuracy Notes

• Project page: machine0.io
• Source repository: github.com/fdmtl/machine0-nixos, github.com/fdmtl/machine0-ubuntu
• HN launch thread: news.ycombinator.com/item?id=49348136 (YC S26)
• License: Proprietary SaaS; VM images are MIT (per repo)
• Source last checked: 2026-08-19 (direct from machine0.io homepage + pricing page)