machine0 review, pricing and verdict

Pick machine0 for a machine that stays on, sleeps at storage-only cost and can be rebuilt from a NixOS flake. Skip it if you need managed databases, Kubernetes or compliance certificates, none of which the vendor site lists. GPU rates are now public, and several rose on 1 August 2026.

machine0 is a command-line cloud service from Fundamental Software that rents persistent KVM virtual machines, billed per minute and driven from a CLI, SDK or MCP server. It lists 14 CPU sizes up to 60 vCPUs and nine GPU sizes across five regions, with 99.99% uptime stated by the vendor.

About machine0

machine0 is a cloud VM service from Fundamental Software that gives developers persistent, real KVM virtual machines controlled from the command line. It launched in 2026 and is backed by Y Combinator; the founder, Barnaby Malet, was previously co-founder and CTO of Upflow. The pitch is a machine that stays on, holds a static IP and answers a terminal command in seconds, aimed at long-running coding agents such as Claude Code and at engineers who would rather script infrastructure than click through a console.

Every machine is a full KVM instance rather than a container, and both base images, NixOS and Ubuntu, are published as open source on GitHub. NixOS flakes give reproducible builds and one-command rollbacks, while the Ubuntu image ships with Docker, Node, Python and the Claude Code and Codex agents already installed. Every machine is reachable on a public IP, with an HTTPS address under mac0.io, and the vendor states 99.99% uptime across five regions: US East, US West, UK, EU and Asia. Suspending a machine stops compute billing, which suits agents that work in bursts. Profiles inject credentials, MCP servers, prompts and environment variables into new machines, and a remote MCP server lets an agent create, suspend and snapshot machines itself.

The product has moved quickly since the June listing. The changelog records a US West region on 10 July 2026, NVMe sizes the same day, dedicated-vCPU premium sizes on 2 August, published GPU rates that rose on 1 August after DigitalOcean raised its own, a TypeScript SDK on 12 October, standing orders for out-of-stock sizes on 13 October and scheduled actions, still experimental, on 15 October. The same changelog admits that GPU sizes were wrongly shown as out of stock from 25 August until a fix on 12 October, so check capacity before planning around a size. GPU sizes are offered in four of the five regions; US West is CPU only.

Where machine0 is the wrong pick: it lists no managed databases, load balancers, object storage or Kubernetes, and no compliance certificates, so a team that needs those next to its servers will be better served by Hetzner. CoreWeave is built around cluster-scale GPU capacity and suits that job better. For short batch jobs rather than a machine kept for weeks, a serverless platform such as Modal fits better, and sandbox products like Daytona or Podflare cover throwaway agent environments. Most of the GPUs on offer are NVIDIA parts, with AMD as the alternative.

Pricing

Charges accrue by the minute against a prepaid balance (minimum top-up $5, unused credit refundable), and prices are identical in every region. CPU sizes run from small (1 vCPU, 1 GB) at $0.013/hr to 6xl (60 vCPU, 240 GB, 900 GB) at $3.714/hr. The three -nvme sizes cost about 17% more per hour for NVMe storage, and the two -premium sizes (xl-premium $0.236/hr, xxl-premium $0.473/hr) use dedicated vCPUs.

GPU sizes run from gpu-4000ada-1 at $0.836/hr to gpu-h200-8 (8x H200) at $39.336/hr; the MI300X, H100 and H200 rates rose on 1 August 2026 after DigitalOcean raised its own GPU prices. A suspended machine pays only $0.078/GB/month for image storage. There is no free tier.

Rates read from machine0.io on 2026-10-19.

Key Features

• Suspend and resume billing: Suspending a machine freezes its state and stops compute charges, leaving only image storage billed, so idle gaps between agent runs cost little.
• Real KVM machines, not containers: Sizes run from a one-vCPU box up to the largest 6xl, including NVMe-disk variants and dedicated-vCPU premium sizes, in five regions.
• Open-source NixOS and Ubuntu images: Both base images are public on GitHub, and NixOS flakes give reproducible builds with one-command rollbacks and no configuration drift.
• Static IP and HTTPS address on every machine: A public IP and an HTTPS endpoint at its own mac0.io subdomain come with every machine, with no NAT, tunnels or reverse proxy to set up.
• Built for agents: CLI, SDK and MCP: Every operation is a CLI command with JSON output, a remote MCP server and a TypeScript SDK expose the same controls, and profiles inject credentials, MCP servers and prompts.
• Nine GPU sizes with published rates: Sizes range from entry RTX cards to eight-GPU machines, each with an hourly rate on the pricing page, in four regions.
• Standing orders for out-of-stock sizes: The order command queues a machine for a size with no capacity and creates it automatically, with an email, when capacity appears; orders last up to 30 days.

Strengths

• Suspend billing fits bursty agent work: a sleeping machine costs storage only, not compute you are not using.
• GPU rates are now public, so cost can be compared with other GPU clouds before signing up, which was not possible at the June listing.
• Open-source base images and a flake-based build path make it easier to rebuild a machine elsewhere, which reduces lock-in.
• The CLI, SDK and MCP server cover the same operations, so automation does not depend on the web dashboard.

Weaknesses

• No managed databases, Kubernetes, load balancers, CDN or object storage appear on the public site, so teams that outgrow raw VMs must run those themselves or add a second provider.
• The product changes fast: GPU rates rose in August 2026 and GPU sizes showed as out of stock by mistake until a fix in October, so check rates and capacity before committing.
• No compliance certificates (SOC 2, ISO 27001, HIPAA) are listed on the vendor site, which can block procurement in regulated teams.
• There is no free tier or trial, and a minimum top-up is needed before first use, although unused credit is refundable.