Standing Orders: Get a GPU VM the Moment Capacity Appears
GPU sizes sell out. Until now that meant retrying machine0 new by hand until a slot opened. With machine0 order you place a standing order instead: the VM waits in an unbilled ORDERED state, machine0 checks for capacity every minute, and the moment a slot appears the VM is created and you get an email.
## The problem with on-demand GPUs
machine0 bills GPU VMs by the minute with no commitment, from a single RTX 4000 Ada at $0.836 per hour up to eight H200s at $39.336 per hour. The trade-off of buying capacity this way is that a size can be sold out when you ask for it. When that happens, machine0 new fails with an out-of-stock error, and the only option used to be trying again later.
That is fine for a person at a keyboard. It is not fine for an agent that needs an H100 at 3am, or for a training job you want to kick off the moment hardware is free.
## What a standing order does
machine0 order takes every flag machine0 new takes, so the command you already know becomes an order:
machine0 order my-gpu --size gpu-h100-1 --region tor1If the size is in stock, order simply creates the machine, exactly like new. If it is not, the machine is created in the ORDERED state and shows up in machine0 ls right away. Three things are true of an ordered machine:
- It bills nothing. You pay only once it is running.
- machine0 checks for capacity every minute. As soon as a slot appears the VM is created with the name, size, region and key you asked for.
- You get an email when it is running, so nothing needs to poll.
The out-of-stock error from machine0 new now points at the order command too, so you will not miss it.
## Setting an expiry
An order stays open for seven days by default. --order-expiry changes that, as a duration or an ISO timestamp. The minimum is ten minutes and the maximum is thirty days.
machine0 order my-gpu --size gpu-h100-1 --region tor1 --order-expiry "2 days"
machine0 order my-gpu --size gpu-h100-8 --order-expiry 2026-11-01T00:00:00ZIf capacity never appears before the deadline, the order flips to ERRORED with the reason "order expired" and you get an email. From there machine0 start places the order again, or machine0 rm removes it.
## Managing an open order
machine0 get shows a Status Info row explaining that the machine is waiting for stock, plus an Order Expires row with the deadline.
To cancel, destroy the machine:
machine0 rm my-gpuTo try to fill the order immediately, for example after you hear that stock came back, start it:
machine0 start my-gpuIf the size is still sold out, start fails fast with a clear error instead of leaving the machine stuck, and the order keeps waiting.
## A note on idle detection
Orders combine well with idle detection, but the clock starts when the order fills, not when you place it. An order that fills unattended at night gets its idle action about one timeout after boot. For that reason prefer --on-idle suspend over destroy on ordered machines: a suspended VM keeps its disk, stops machine billing, and comes back with machine0 start whenever you are ready to use it.
## Where this fits
Standing orders turn a sold-out size from a failure into a queue position. The pattern we see most is an agent or a script that orders the GPU it needs, does CPU work in the meantime on a large or xl VM, and picks up the GPU job when the email arrives. Everything else about the machine is unchanged: per-minute billing, a static IP, HTTPS at its own hostname, and suspend to stop paying when the job is done.
The command shipped in CLI version 1.0.172. Update with the install script, or install it fresh:
curl -LsSf https://machine0.io/install.sh | sh