How to Become a Vast.ai Host: The Complete Walkthrough

What you need before you start

• A dedicated machine. One or more NVIDIA GPUs in a box that will do nothing else. Renters expect the machine to be available around the clock during a rental, so a computer you also game on does not qualify.
• Ubuntu. The standard host OS. A clean Ubuntu Server install is the least painful path.
• A network you control. Renters connect directly to your machine, so you need the ability to open a range of ports on your router. Solid upload bandwidth makes your listing more attractive for data-heavy jobs.
• A separate hosting account. Vast.ai asks hosts to create a hosting account distinct from any renting account, and to accept the hosting agreement before installing anything.

Phase 1: prepare the machine

Vast.ai expects you to arrive with a working GPU box: Ubuntu installed, disk partitioned with plenty of space for renter images, NVIDIA drivers loaded, and Docker running with GPU support. In order:

1. 1Install Ubuntu Server and give the machine a static LAN address so port forwarding does not silently break after a DHCP renewal.
2. 2Install the NVIDIA driver and confirm every card shows up in nvidia-smi. A card the driver cannot see is a card you cannot rent out.
3. 3Install Docker and the NVIDIA container toolkit so containers can reach the GPUs. Our open source notes on Docker with NVIDIA GPUs walk through this part.
4. 4Disable unattended upgrades. A background update that restarts Docker will kill every active rental and dent your reliability score. The full reasoning and commands are in our auto updates guide.

The complete OS-level checklist, including networking and filesystem choices, lives in Ubuntu server setup for GPU hosting.

Phase 2: install the Vast.ai host software

With the machine ready, the marketplace side is short. From your hosting account, open the host setup page at vast.ai/hosting and follow the guided install. It gives you a command to run on the machine that installs the Vast daemon, benchmarks your hardware, and registers the machine to your account. Your machine typically appears on the platform within minutes.

Then open the ports. Clients need open ports to connect directly to the machine for most jobs, so forward the port range the setup page assigns you from your router to the machine's static LAN address. Test from outside your network before moving on; a machine with dead ports will fail jobs and collect bad reliability stats.

Phase 3: verification, and how long it takes

Verified machines rank better and win more rentals. Verification is automated for most hardware, and the timeline depends on what you are listing:

• High-end machines (think H100 or A100 80GB class) are usually tested within about a day.
• Strong multi-GPU consumer rigs, like 8x RTX 4090, typically get tested within a week.
• Everything else gets random automatic verification passes, roughly weekly. Repurposed mining rigs may never be verified at all.

You can rent out an unverified machine in the meantime, just expect less demand. Use the waiting period to burn the machine in and fix anything flaky before renters find it for you.

Phase 4: day-one listing settings

Five settings decide how your listing competes:

• On-demand price per hour. The headline number renters sort by. Start near what comparable rented machines are paying, not what unsold listings are asking.
• Interruptible minimum bid. The floor for spot rentals that fill gaps between on-demand jobs. A sensible discount below your on-demand price keeps the machine earning when demand is thin.
• Min GPU grouping. The smallest slice of the machine a renter can take. Smaller slices rent more often; whole-machine rentals pay more per job.
• Storage price. Charged per GB per month. Renters compare this too, and an outlier number here quietly filters you out of searches.
• Offer end date. How long you commit the listing. Remember that a rental contract locks in your terms at the moment of rental, so you cannot change the price on a job already running.

Here is the trap: those prices are not settings you tune once. The market moves every hour. The competitor sitting one cent below you at lunch will be undercut by someone else at midnight, and your carefully chosen price slides from competitive to invisible while you sleep.

Your first week as a host

Expect a quiet start. Reliability history and verification both take time, and rentals compound as your stats improve. Spend the first week watching for failed jobs, confirming ports stay reachable, and resisting the urge to yank the machine offline for tweaks while a renter is on it.

Two follow-ups worth doing this week: create the API key that lets tooling read your machine data, covered in how to create a Vast.ai API key, and then connect it in the GPU Autopilot quick start so your pricing is handled from day one.