Cloud & VPS
What Is machine0 Pricing for Persistent AI Agent VMs?
machine0 pricing starts at $0.013 per hour for CPU VMs and $0.836 per hour for GPU VMs, billed per minute, with suspend-to-stop billing across five regions, verified on the official page as of August 2026. The YC S26 startup sells persistent cloud VMs for AI agents, operated over CLI and MCP.
What matters
- CPU VMs start at $0.013/hr (1 vCPU, 1 GB RAM) and scale to $3.714/hr for 60 vCPU and 240 GB, per the official pricing page as of August 2026.
- GPU VMs start at $0.836/hr (RTX 4000 Ada) and reach $39.336/hr for 8x H200; GPU sizes are offered in us-east, UK, EU, and Asia.
- Billing is per minute: suspend a VM to stop compute billing and pay only $0.078/GB/month for stored images.
- machine0 targets long-running agent workloads (hours to days), not ephemeral sandbox jobs; Modal and E2B cover the short-lived end.
What is machine0?
machine0 is a cloud VM service built for AI agents, launched on August 18, 2026 by Y Combinator S26 startup founder Barnaby, previously CTO of Upflow (YC W20). The command machine0 new mybox provisions an SSH-ready persistent VM with a static IP and HTTPS endpoint in about a minute, per the launch post on Hacker News.
Every machine is a full KVM virtual machine, not a container or sandbox. VMs run NixOS or Ubuntu with Docker, Node, Python, Claude Code, and Codex pre-installed, and can be built reproducibly from Nix flakes or Ansible playbooks. Agents drive everything themselves through the CLI or a remote MCP server, including creating, snapshotting, and tearing down machines.
machine0 advertises 99.99% VM-level uptime across five regions: US East, US West, UK, EU, and Asia. Static IPs and HTTPS endpoints at <vm>.mac0.io come standard, with no NAT and no tunnels, per the official site.
machine0 pricing for CPU VMs
All machine0 pricing is pay-as-you-go, billed per minute, with the same rates in every region. The cheapest CPU VM is small at $0.013/hr (1 vCPU, 1 GB RAM, 25 GB disk), roughly $9 per month if run continuously. The largest standard CPU tier, 6xl, costs $3.714/hr (60 vCPU, 240 GB RAM, 900 GB disk), about $2,711 per month.
Three disk classes exist: standard, -nvme (newer-generation CPUs with NVMe storage for higher IOPS at the same vCPU, RAM, and disk), and -premium (dedicated, non-shared vCPUs with the fastest single-thread performance). Pricing on the -nvme tiers runs from $0.061/hr for large-nvme up to $0.243/hr for xxl-nvme.
machine0 pricing for GPU VMs
GPU machines start at $0.836/hr for a single RTX 4000 Ada (20 GB VRAM, 8 vCPU, 32 GB RAM), and top out at $39.336/hr for 8x H200 (1,128 GB VRAM, 192 vCPU, 1,920 GB RAM, 2 TB disk), per the official pricing table. Available GPUs are RTX 4000 Ada, RTX 6000 Ada, L40S, MI300X, H100, and H200.
GPU sizes are offered in us-east, UK, EU, and Asia, which means US West is CPU-only. Single-GPU tiers range from $0.836/hr (RTX 4000 Ada) to $4.917/hr (H200, 141 GB); multi-GPU configurations reach 8x MI300X at $22.792/hr.
How machine0 billing works
Billing is per minute, and the headline mechanism is suspend-to-stop-billing: a suspended VM stops accruing compute charges and instead costs only $0.078/GB per month for its stored image. Resume picks up from the exact state, which makes the model well suited to agents that work in bursts overnight.
- Minimum top-up is $5, and unused credits are refundable.
- Snapshots and golden images let you clone a fleet from one VM state.
- Persistent block storage volumes range from 10 GB to 16 TB, per the launch post.
- Profiles inject credentials, MCP connections, prompts, and environment variables at VM creation.
Who should use machine0?
machine0 is aimed at teams running long-horizon agents: coding agents on complex features that run 6-8 hours, agent-orchestrated training and RL environments that run for days, and always-on assistants. The launch post describes three real usage patterns: agent fleets (a pilot agent scopes work and delegates to sub-agents, each on its own VM), model optimization and RL environments (a suspended H100 pointed at overnight inference-speed work), and product infrastructure (each user session gets a fresh VM from a versioned image).
Skip it if your workloads are short-lived function calls, where per-second serverless sandboxes are cheaper; if you need Windows guests; or if you want a managed dashboard rather than a CLI your agents operate.
machine0 vs raw VPS vs sandboxes
The official FAQ draws the line against both neighbors. Compared to a raw VPS, machine0 adds an agent-operable CLI and MCP server, NixOS declarative provisioning, suspend/resume with full state snapshots, per-minute billing, and pre-built images. Compared to sandboxes like E2B or Modal, machine0 is persistent rather than ephemeral: static IPs, real KVM, and data that persists, built for workloads that run hours or days instead of seconds.
Budget VPS providers such as RackNerd remain the cheaper option for plain always-on servers, while Docker sandboxes for AI agents cover the ephemeral end of the spectrum.
machine0 limitations and the honest catch
Three caveats matter before you buy. First, GPU machines are not available in US West, so teams that need both low-latency West Coast GPU work and CPU fleets must split regions. Second, there is no free tier: the entry point is a $5 minimum top-up, and the pricing page lists only pay-as-you-go per-minute rates with no annual plans. Third, suspension is not free: a suspended VM still accrues $0.078/GB per month for its stored image, which on a 900 GB 6xl machine adds up to roughly $70/month even while idle.
The company is also young. machine0 launched on August 18, 2026 as a YC S26 startup, so the 99.99% uptime figure is a stated SLA rather than a long track record. For teams that want a mature, dashboard-driven managed VPS instead of an agent-native CLI, established providers remain the safer default.
machine0 alternatives
If machine0 does not fit, the realistic alternatives split by workload shape:
- RackNerd: budget KVM VPS with fixed monthly plans, the low-cost choice for a simple always-on server without agent tooling.
- Modal: serverless, per-second GPU and CPU compute for short jobs; no persistent VMs or static IPs.
- E2B: sandboxes designed for AI agents, ephemeral and provider-managed, suited to short task execution.
- Standard cloud VPS from AWS, Google Cloud, or DigitalOcean: full control, monthly billing, but dashboard-heavy and without the agent-native CLI, MCP, and snapshot workflow.
No free tier is listed for machine0; the entry cost is the $5 minimum top-up.
When comparing, the buyer question is persistence: sandboxes win for tasks measured in seconds, budget VPS providers win for static always-on servers, and machine0 targets the hours-to-days agent workloads in between.
At a glance
| Tier | Specs | $/hour | ~$/month |
|---|---|---|---|
| small (CPU) | 1 vCPU / 1 GB / 25 GB | $0.013 | $9 |
| medium (CPU) | 2 vCPU / 2 GB / 60 GB | $0.034 | $25 |
| xl (CPU) | 4 vCPU / 8 GB / 160 GB | $0.104 | $76 |
| xxl (CPU) | 8 vCPU / 16 GB / 320 GB | $0.208 | $152 |
| xxxl (CPU) | 16 vCPU / 64 GB / 200 GB | $0.825 | $602 |
| 4xl (CPU) | 32 vCPU / 128 GB / 480 GB | $1.980 | $1,445 |
| 6xl (CPU) | 60 vCPU / 240 GB / 900 GB | $3.714 | $2,711 |
| gpu-4000ada-1 | 1x RTX 4000 Ada / 8 vCPU / 32 GB | $0.836 | $610 |
| gpu-l40s-1 | 1x L40S / 8 vCPU / 64 GB | $1.727 | $1,261 |
| gpu-mi300x-1 | 1x MI300X / 20 vCPU / 240 GB | $2.849 | $2,080 |
| gpu-h100-1 | 1x H100 / 20 vCPU / 240 GB | $4.851 | $3,541 |
| gpu-h200-1 | 1x H200 / 24 vCPU / 240 GB | $4.917 | $3,589 |
| gpu-h200-8 | 8x H200 / 192 vCPU / 1,920 GB | $39.336 | $28,715 |
FAQ
Is machine0 free?
No free tier is listed on the official site. The entry point is a $5 minimum top-up, and unused credits are refundable. Billing is per minute, so a suspended VM costs only its stored image ($0.078/GB per month).
How is machine0 different from Modal or E2B?
Per machine0's own FAQ, sandboxes like E2B and Modal are ephemeral containers the provider controls, optimized for short-lived tasks. machine0 is persistent VMs you own, with static IPs, real KVM, and data that persists, built for workloads running hours or days.
Can I host a website on machine0?
Yes. Every VM gets a public static IP and an HTTPS endpoint at <vm>.mac0.io, and the official site lists web apps and databases as a supported use case. NixOS or Ubuntu images come with Docker, Node, and Python pre-installed.
Related reading
Docker sandboxes for AI agents, ToolBistro Radar index, RackNerd budget VPS