The invoice arrives whether the node is full or not. Hardware lease or purchase, colo, transit, power. The RAM you have not sold yet still sits in that total. An empty node is not free capacity waiting for a better idea. It is a bill with no revenue line next to it.
Hosts already argue this in public. On LowEndTalk, a margin thread worked through what is left after hardware, colo, bandwidth, and power. One participant's guess, argued in the replies, landed around 5 to 20 percent gross, and less if the box is leased. Someone described another host at twenty-five cents per account. Whatever the exact number on your spreadsheet, the shape is the same. The fixed costs are real. The next VM is how you pay them down.
Why a cut still makes people flinch
Affiliate math is not new to this market. A thread collecting programs shows recurring payouts hosts already live with: several at 10 percent, PulsedMedia in the 7.5 to 9 percent band, Clouvider up to 15 percent, and other numbers beside those. That is a set of examples from one discussion, not a survey of the industry. It is enough to show that paying for a channel is ordinary.
It is also enough to show why a new cut feels expensive. In a thread about a proposed 15 percent affiliate program, the first reply is that those points have to come from somewhere, and prices end up about 20 percent higher. That objection is fair. Gross margin is thin. A large cut of a sale you already would have closed is a bad trade. A smaller cut of that same sale is still a bad trade.
The useful question is which sale this is.
The next guest is cheap. The empty node is not.
Once the node is paid for, the next guest is not free. It still burns power. It still needs an address. It still eats bandwidth. It still opens a support ticket and, sometimes, an abuse ticket. Those are real costs.
They are also smaller than the invoice you already accepted when you racked the machine. An empty 64 GB node and a half-full 64 GB node share most of the same monthly pain. The difference is whether someone is paying you for the second half.
On a ten dollar month, five percent is fifty cents. Next to the colo invoice, that is not the number that decides whether the node was a good purchase. What decides is whether the channel fills slots you were not going to fill yourself.
Plenty of hosts are not looking for a new billing system. One LowEndTalk request says it plainly: the poster wants a module, and wants to keep WHMCS. A marketplace that only works if you throw that out is a migration, not a channel. Keep the customers you already bill. Keep the panel you already operate. If a sale starts on the marketplace, the money lands in your Stripe account through Connect, and the platform takes its cut on that path. A customer you already have stays on the stack you already run.
If someone would have found you on LowEndTalk, opened a ticket, and paid you directly, sending that same person through a marketplace is a loss. The sale the channel is for is the one you did not have: a buyer who searched for a city and a size and found your plan because it was listed where they were looking.
What the empty slots are worth
Picture a node with room for twenty more of the plan you sell for ten dollars. That picture is a sketch. Addresses run out before RAM does, and a noisy neighbor can make the leftover cores unusable. Even so, the shape holds. Fill ten of those slots through a channel you did not have, and the fee on those ten months is small next to the hardware invoice that was already going out.
Fill none of them because every channel asked for a cut, and the invoice still goes out. Refusing is still the right call when the channel wants the keys to the cluster, or when it takes the customers you already had. On the idle-capacity question alone, an empty node is already the expensive option.
The invoice does not wait for the slots to fill.