Become a Community Supplier

Earn money renting out your own GPU machines on GPU.ai. Community Cloud is self-serve: enroll your org, register a machine, install the agent, pass an automated hardware verification, and your machine appears in the customer catalog. No sales call, no contract negotiation.

This guide walks the whole lifecycle: enrollment → agent install → verification → pricing → rentals and earnings → payouts → delisting.

Everything below can be done from the dashboard (Cloud → Supply), the gpu CLI, or the public API with an API key holding the community scope.

1. Enroll as a supplier

Any org can enable the supplier role — it is instant and reversible:

gpu community enable
gpu community status

2. Register a machine

Register each physical machine you want to list. All spec fields are optional at registration (the agent reports and verification confirms them later), but a machine needs a price, region, and GPU spec before it can appear in the catalog:

gpu community machines register \
  --gpu-type "RTX 4090" --gpu-count 2 --vram-gb 24 \
  --price-per-hour 0.80 --region us-east --tier on_demand

Registration prints a one-time enrollment token (64 hex chars). Save it immediately — it is shown exactly once and can never be retrieved again. If you lose it, delist the machine and register a replacement.

Notes:

• --tier spot lists the machine as cheaper, interruptible capacity you can

take back at any time (see Reclaim); the default on_demand promises the renter guaranteed capacity.

• There is a per-org machine cap (default 20 active machines). Delisted

machines free their slot.

3. Install the agent

On the machine itself (Linux, NVIDIA driver, Docker 24+, and nvidia-container-toolkit required):

curl -fsSL https://gpu.ai/gpud-install | sudo GPUD_ENROLLMENT_TOKEN=<token> bash

The installer sets up gpud as a systemd service; it starts heartbeating to GPU.ai within seconds. See the gpud install reference for prerequisites, flags, and troubleshooting (journalctl -u gpud -f is your friend).

4. Automated verification

Your machine is not listed until it passes verification — this is what keeps fake listings out of the catalog, with no human review to wait on:

1. Uptime window — the machine must heartbeat continuously for about an

hour after its first heartbeat.

1. GPU benchmark — a server-issued benchmark challenge runs on your GPU;

the measured performance must match the GPU model you'll be listed as.

1. Bandwidth probe — a minimum network throughput check.
2. GPU uniqueness — each physical GPU (by hardware UUID) can back only one

listing across the whole platform.

Expect 1–2 hours from agent install to listed. Verification then re-runs periodically (roughly daily) and immediately after anything suspicious (GPU swap, spec change). A machine that fails re-verification stays out of the catalog until it passes again.

Check progress any time:

gpu community machines list
gpu community machines get <machine-id>

Why is my machine not visible in the catalog?

A machine appears to customers only when all of these hold — the list view shows you which one is missing:

• STATUS is listed (not pending verification, delisted, draining, or

suspended)

• VERIFIED is verified
• ONLINE is yes (heartbeated within the last 60 seconds)
• OCCUPIED is no (a whole-node machine already renting is not a live offer)
• price, region, GPU type, and GPU count are all set

5. Pricing

gpu community machines price <machine-id> 1.85

• The price is the whole-machine hourly rate in USD — never per-GPU. An

8-GPU machine priced at 4.00 rents all eight GPUs for $4.00/hr total.

• There is a platform minimum; prices below it are rejected (the error shows

the current floor).

• Repricing never changes an in-flight rental — the renter keeps the rate they

launched at. The catalog updates within about 30 seconds.

6. Rentals, reliability, and earnings

When a customer rents your machine, their workload runs in a container the platform manages; billing is per-second and your cut accrues automatically:

gpu community earnings

Your machine also carries a reliability score (0–100), computed from uptime, abrupt disconnections, and provisioning failures. Machines that fall below the platform threshold are automatically delisted (drain-first — a running rental is never killed) with a visible reason. Keeping the machine online and stable is the whole game.

Payouts

Payouts are initiated by GPU.ai staff once your unpaid balance clears the documented minimum ($50) and cadence (weekly). Two identity steps are required first, in order:

1. Submit payout details — legal name, country, and payout method type

(never account numbers) — from the dashboard's Earnings page, with gpu community kyc, or via PUT /v1/community/suppliers/me/kyc.

1. Verify your identity. Document verification is required before your

first payout: a government-ID + selfie check completed in a browser, hosted by our verification partner (we never see or store your documents — only the pass/fail result). Start it from the Earnings page or with gpu community verify-identity; your status updates automatically once the check completes. If a check fails, the reason (for example an expired document) shows in gpu community status and on the Earnings page, and you can retry.

Until verification completes, earnings still accrue normally — only the payout itself is locked. Verification is currently unavailable in a small set of countries; if yours is one of them the dashboard and API will say so when you submit payout details, and payouts cannot be enabled for the account.

7. Delisting and reclaim

gpu community machines delist <machine-id>     # any tier
gpu community machines reclaim <machine-id>    # spot tier only

Both are drain-not-kill: they stop new rentals immediately, but a rental already running is never terminated — an occupied machine shows draining until the rental ends, then leaves supply. Reclaim is the spot-interruption path: the renter is preemption-warned and keeps a grace window.

8. Suspension

GPU.ai can suspend a machine or an entire supplier account for abuse, fraud, or sustained unreliability. Suspension is visible (with the reason) in gpu community machines list and the dashboard. A suspended machine's rentals drain rather than being killed. Reinstatement puts the machine back through the same automated verification every new machine passes — there is no shortcut back to listed. To appeal, contact support with your machine id.

Command reference

Task	Command
Enable the supplier role	gpu community enable
Check supplier status	gpu community status
Submit payout details	gpu community kyc
Verify identity for payouts	gpu community verify-identity
Register a machine	gpu community machines register
List your machines	gpu community machines list
Inspect one machine	gpu community machines get
Set the price	gpu community machines price
Remove from supply	gpu community machines delist
Reclaim a spot machine	gpu community machines reclaim
View earnings	gpu community earnings