We run infrastructure for clients, we do not rent out nodes, so the question "why pay you when the hoster's box is 93 dollars a month" reaches us often. We answer it with numbers. The 93 dollars is one of seven lines. The other six hide in the bill, and the operational ones drain a team that decided to run nodes on its own. We know these numbers from our own polygon: a fleet across 12 countries where we sharpen the discipline before we bring it to clients.
Of the seven lines, four are hardware: rent or capex, power, rack and transit, parts replacement. They stay with the hoster or with you as the node owner. Three are operational: on-call, observability, and the discipline around keys and incidents. Those you either staff yourself or hand to a team like us. The line between doing it yourself and outsourcing sits exactly here, and below we show why the cheap hardware line is misleading.
Seven lines and who carries them
We cost per slot-month, not per node: one validator slot, fully kitted and serviced, for a calendar month.
Hardware lines, which stay with the hoster or node owner:
- Hardware or rent. For owned gear, the amortization of capex over 36 months. For rented boxes (Hetzner, Latitude, and others), the monthly price.
- Power and cooling. For colo, part of the rack contract. For rented boxes, already folded into the plan.
- Rack, cross-connect, transit. Space in the rack, cross-connects to the uplinks, transit itself.
- Parts replacement (RMA). Contractually the hoster's job, not yours.
Operational lines, which you staff yourself or we carry:
- On-call. Rotation and escalations. Covered separately below, because this is the line people understate most.
- Observability. Monitoring, alerting, the log pipeline, dashboards.
- Keys and incidents. HSM, rotation, slashing-aware playbooks, incident reviews.
"Uptime percentage" does not enter this model: it is a derived metric that lives on a dashboard, not in a budget.
Hardware: the cheap half
The classic budgeting mistake reads like this: "Hetzner's AX102 is 93 dollars, we need 10 nodes, so the budget is 930 a month." Six months in, the real bill lands near 1,400 and the owner genuinely cannot see where it came from.
Part of it comes from hardware lines no colo landing page prints. On our polygon, one Validator-S node (L1 validator, non-archival, no GPU; 8c/16t Zen4-class, 64 GB ECC, 2x2 TB NVMe) works out like this:
- AX102 rent at Hetzner, transit included: 93 dollars.
- Traffic above quota on a busy node: about 12.
That is the cheap half, around 105 dollars per slot-month. Parts replacement does not enter it as your line: NVMe under validator load burns through wearout faster than a desktop profile, but replacing it is the hoster's contractual job. The question is not the price of the drive but who makes the hoster's "obligation" turn into "fixed in 4 hours" rather than "we'll get to it next week." That is already an operational line, and it has nothing to do with hardware.
Operational costs: what the price list omits
Now the second half, the one a node owner does not see until they start running nodes themselves. It does not behave like hardware. Hardware scales linearly: ten nodes cost about ten times one. A team does not scale that way.
A sane on-call rotation is a minimum of three engineers: one on shift, one on backup, one resting, around the clock. Fewer than three means either gaps in coverage or one burning-out admin who becomes a single point of failure. An engineer who can actually fix a validator that dropped out of consensus at 3 a.m. costs an employer, fully loaded (salary plus employer taxes and benefits), from 9 thousand dollars a month in the EU to 16 to 26 thousand in the US, and more at blockchain grade. A rotation of three runs from 26 thousand a month in the cheapest jurisdiction to 78 thousand in the US, and that floor does not move between 6 nodes and 60.
On top of the rotation sit two more operational lines: observability (the monitoring stack, alerts on p99 attestation rather than on ping) and the discipline around keys and incidents (HSM, rotation, slashing-aware playbooks, incident reviews). Both are nearly independent of whether you run 6 nodes or 60: the same team carries them.
Why you cannot divide the operational half per node
The temptation is to take the cost of that team, divide it by the fleet, and call it "the cost of a node." On a large fleet the number looks fine; on a small one it is ruinous. Take the low end, 26 thousand a month: across 200 nodes that is 130 dollars per node, across 20 it is 1,300, across 6 it is over 4,000. Same team, same month, only the denominator changes. In the US, triple it.
That math is the whole fork in the road. For the per-node operational line to get small, the fleet has to run into the hundreds; until then you either keep a team at an absurd per-node price or put one person on call and get a single point of failure in your own ops. Outsourcing does not erase the cost of the team: it lets several fleet owners share one already-staffed rotation, so you pay for a slice of the team rather than three full salaries.
What this means for you
If you are comparing options, treat the operational half as a fixed team cost, not something you divide per node. The hardware lines are linear and stay with the hoster. The operational ones start at 26 thousand a month for sane 24/7 coverage if you hire it yourself, and below a couple hundred nodes your own team is the most expensive option, not the cheapest.
This is where our line sits. We keep the same 24/7 team of three to four engineers for around 15 to 20 thousand dollars a month, flat and at any fleet size: cheaper than building a rotation in even the lowest-cost jurisdiction, and it is already staffed and already on shift. What that coverage includes is laid out in operate.
Want to run the numbers against your specific fleet: get in touch and we will work them out together.