How Earnings Work
How your payout is calculated, what share you keep, and how you get paid.
The short version
Your node earns a share of what your cluster makes renting out compute. How big that share is depends on how many qualifying GPUs your node has and how reliably it stays online.
How your share is calculated
- Your cluster earns revenue over a payout period from the workloads it hosts.
- The 30% network fee is deducted, leaving 70% to distribute.
- That 70% is split across qualifying nodes in proportion to each node's score.
Why there's a 30% network fee
The fee isn't charged on top of your work — it pays for the shared infrastructure your node needs in order to earn anything at all. Running a cluster means running machines and services that stay up whether or not any individual node does.
Kubernetes control plane
Schedules and manages every workload that lands on your machine.
Headscale coordination server
The private network that lets a machine in your home be reachable as a provider, without exposing anything to the open internet.
Monitoring and security
The uptime tracking that determines your payout, plus the detectors that keep abusive workloads off the network.
Dashboard, API, and payouts
Everything you use to manage your node, including the rails that move funds to you.
What counts
Qualifying GPUs
These GPU models count toward your multiplier. A node with two of them scores twice what a node with one does, all else equal.
If your GPU isn't on this list, contact support to confirm how your node is scored.
Eligible uptime
Uptime is counted in cycles rather than raw hours. A cycle only counts toward your score if your node was online for at least 80% of it, and an outage longer than 10 minutes breaks the cycle instead of being averaged away. Sustained availability is therefore worth more than the same number of hours scattered across interruptions. A cycle also has to run at least 12 hours to count at all. A node that restarts every few hours can finish a period with no qualifying time even though it was online for most of it.
A worked example
One week, three nodes. The cluster earns $930. The 30% network fee is $279, leaving $651 to distribute.
| Node | Qualifying GPUs | Qualifying hours | Score | Share | Payout (USD value) |
|---|---|---|---|---|---|
| Node A | 2× RTX 4090 | 168 (full week) | 336 | 51.6% | $336 |
| Node B | 1× RTX 4090 | 168 (full week) | 168 | 25.8% | $168 |
| Node C | 1× RTX 3090 | 147 (87.5% — 21h maintenance window) | 147 | 22.6% | $147 |
| Total | 651 | $651 | |||
In this example each point of score is worth exactly $1. Node A earns double Node B on identical uptime, because GPU count multiplies your score. Node C's downtime was a single 21-hour block, so the stretches either side each still qualified on their own and it lost exactly the hours it was offline. The same 21 hours split into many short outages would have cost more, because every break of more than 10 minutes ends a cycle.
Every amount above is a USD value. The payout itself is sent in AKT — see “What you're paid in” below.
What you're paid in
Your share is calculated in US dollars. When a payout is issued, that amount is sent to you as AKT, converted at the market rate at the time of distribution.
Why your payout might be lower than you expected
- Downtime broke cycles rather than reducing them proportionally — an interruption can cost you a whole cycle, not just the minutes you were offline.
- Your node restarted often enough that no stretch of uptime reached the 12-hour minimum, so none of it qualified.
- Your node joined partway through the payout period.
- Your payout account or identity verification isn't finished yet, so funds haven't been released.