Skip to main content
Most large GPU capacity never reaches a public catalog. It moves privately between operators — a rack frees up, a commitment ends, someone needs it gone — and it is spoken for before anyone else hears about it. Radar is where that capacity surfaces. We source it, verify it before it goes live, and you book it by inquiry rather than by clicking start.
Radar lives in the dashboard at /app/radar and is visible to signed-in accounts only. That is deliberate: these listings are not meant to sit on the open web.

​
Radar vs the catalog

Both are real capacity. They answer different questions. If you want a GPU now, use the catalog. If you want a specific machine — or several of them — for a long stretch, that is Radar.

​
What a listing tells you

Every listing carries the full sheet for one node, so you can size the cluster before talking to anyone:
GPU model, how many GPUs are in the node, and VRAM per GPU. CPU model and socket count.
RAM in GB and its type, storage in TB and its type.
The interconnect between the GPUs and the network uplink out of the node — the two numbers that decide whether multi-node training is worth attempting.
Price per hour, region, the date the capacity frees up, and the minimum and maximum term the operator will sign.
Listings say how many identical nodes are on offer, not just that one exists.
The sheet is folded on the card by default — a full node prints ten-plus rows, and three of those open at once stops the grid being scannable. Open the one you are actually considering.

​
Making an inquiry

Pick a listing and send an inquiry. We ask for:
​
string
required
Who we are talking to.
​
string
required
Where the confirmation goes.
​
string
Optional. Faster for large or time-sensitive requests.
​
string
required
Free text on purpose — “3 months”, “6–12 months”, “open-ended”. Terms on this side of the business are negotiated, not picked from a menu.
​
string
Anything that changes the answer: how many nodes, when you need them, what you are running.
Our team confirms the capacity is still there, agrees the term, and gets you set up. Nothing is charged at the inquiry stage.

​
Listing your own node

If you have a whole node to sell, propose it from Sell → Radar in the dashboard, or over the API. It lives on the sell side rather than on the page above, for the same reason listing a machine does: the Radar page is where you buy whole nodes, and the Sell tabs are where you offer them. A proposal is reviewed before anyone sees it. Buyers see nothing until we approve it, and if we turn it down you get a reason to act on rather than silence. Once it is live, two things are yours to do and they are not the same:
Takes the node off the grid straight away and keeps its approval, so putting it back needs no second review. This is the one to use when the hardware is sold elsewhere: it must stop being advertised now, not after someone reads a queue.
Approval describes specific text; once the text changes it no longer describes it. So an edit returns the node to review and it leaves the grid until approved again — worth knowing before you fix a typo on a live listing.
Listings do not carry a seller’s name. Radar is capacity we verified, and that is what a buyer is reading — so who proposed a node stays between you and us.

​
What Radar is not

  • Not self-serve. There is no button that starts a Radar node. If you want capacity in the next minute, that is the catalog.
  • Not a public board. Listings are visible to signed-in accounts only, and every one is checked by us before it appears — including the ones proposed by sellers.
  • Not a waitlist. Listings are capacity we have already sourced and checked. If something is on Radar, it exists.

Open Radar

Sign in to see the nodes currently on Radar.