Pay for what you
actually use.
Every VM and bare-metal host is billed by the second. No subscriptions, no monthly minimum, no contact-sales gating. The monthly numbers below show what you'd pay running 24/7 — the actual bill scales down with anything less.
Pay for what you
actually use.
- + 1 vCPU
- + 1 GB ECC
- + 10 GB network NVMe
- + 10 TB transfer
- + IPv6 /64 (IPv4 +$2/mo)
- + 2 vCPU
- + 8 GB ECC
- + 80 GB network NVMe
- + 10 TB transfer
- + Snapshots (50% of NVMe rate)
- + MFA + audit log
- + 4 vCPU
- + 16 GB ECC
- + 160 GB network NVMe
- + 10 TB transfer
- + Snapshots (50% of NVMe rate)
- + MFA + audit log
- + Single-tenant host
- + ECC memory
- + Local NVMe + network NVMe
- + 50 Gbps host uplink
- + Console + rescue mode
- + Per-second billing
Common questions, straight answers.
How does per-second billing work?
Every VM and bare-metal host is metered to the second from the moment it boots until it's destroyed. Run a VM for an hour and pay for an hour. Run it for 11 seconds during a CI job and pay for 11 seconds. No monthly minimum, no rounding up. The monthly numbers on the pricing page are just the equivalent if you ran the resource 24/7 for the month.
Can I pick my own vCPU / RAM / disk shape, or am I stuck with presets?
Either. The tiles above are presets sized for common workloads, but the dashboard lets you dial CPU class, vCPU, RAM, boot disk, GPU, and public IPv4 independently and see a live per-second quote as you tweak. Same per-unit prices for everyone (about $2.50/mo per vCPU on Intel Cascade Lake, $1.25/mo per GB RAM, $0.05/mo per GB NVMe, $2.00/mo per public IPv4; AMD Milan vCPU runs a little more) whether you pick a preset or go fully custom.
Do you have GPUs?
Yes — NVIDIA RTX 4090 via whole-card PCIe passthrough. You get a dedicated physical GPU, not a time-sliced fraction. Attach one or more per VM; the card's vCPU and RAM are reserved alongside it, so GPU VMs start at 4 vCPU / 8 GB. Billed per second at $0.60/GPU-hour (≈$440/mo) plus the vCPU and RAM you configure. The RTX 4090 is the first card in the catalog — more GPU models are on the roadmap.
Do you charge for egress / bandwidth out?
Bare-metal leases include unlimited bandwidth — no transfer cap and no egress fees. VMs include a 10 TB monthly transfer allowance (in and out). VM overages bill at $0.0098/GB, a fraction of what the hyperscalers charge. We don't think bandwidth should be a tax.
Can I switch a VM to a bigger size without rebuilding?
Yes. Vertical resize hot-plugs vCPU and memory live where the guest supports it; otherwise it's a quick stop-edit-start. Either way, your disks, network attachments, public IPs, and snapshots come along untouched — no data migration needed.
What does the 50% beta discount cover?
Any SKU, for the first three months, for the first 100 users on the waitlist. After three months you roll to the standard rate. You can cancel any time and you're never on the hook for more than what you've already used.
Where are your data centers?
DC-1 is live in Dallas. Additional regions are on the public roadmap — for now everything runs in DC-1.
Do you offer managed Kubernetes or container hosting?
Not as a product on this site directly. For container workloads we run a provider on the Akash Network — same Metalhost hardware, deploy by SDL spec, pay by credit card on the Akash console. See the 'Compute options' section on the home page.
Why two different cloud frontends?
Different workloads, different economics. VMs and bare metal on this site are right when you want familiar tooling, USD invoices, and direct customer support. Akash deploys are right when you want bid-based pricing or short-lived container workloads. The same Metalhost hardware powers both — pick the frontend that fits.
Get on the metal,
at half price.
Public launch june 22, 2026. Drop your email — first hundred founders get half off everything for the first three months. No credit card. No "schedule a call."
waitlist closes when the first 100 fill