What It Costs to Run a CoW Protocol Solver

Search for what it costs to run a solver and you get validator hardware guides and RPC price lists. None of them are about solvers.

The question is really two questions. What does it cost to take part? And what does it cost to win? The answers are an order of magnitude apart, and most conversations about this mix them up.

What it costs to take part

Less than most people expect. We launched Kaisersolver on CoW Protocol for under $1,500, and that covered everything, not a month.

Here is what taking part actually needs:

• A company with KYC. The protocol asks for incorporation details and passports for the people behind the solver. This is the part most builders do not have, and it is what an operator of record carries.
• A bond, or a vouch. Posting your own bond takes serious capital. Being vouched into the CoW DAO bonding pool takes none, but someone has to stand behind you.
• A gas float on each chain. The protocol recommends starting balances that come to a few hundred to a couple of thousand dollars per chain, depending on where you deploy.
• Somewhere to run the engine, and access to the chains it reads. Public RPC providers are enough to start.

The gas float is not an expense

This is the most common mistake in a solver's numbers, and we made it ourselves.

Funding the address that submits your settlements looks like spending money. It is not. The ETH sits there and is used as gas only when one of your winning settlements lands, and the protocol pays that gas back through its weekly accounting. The balance stays yours the whole time.

It is working capital, not a cost. Count it as a cost, and the business looks far worse than it is, especially in the first months.

What comes out before anyone is paid

A solver's reward number is not its income. Before anything reaches the people running it, several things come off:

1. The bonding pool's service fee, if you are in someone else's pool rather than your own.
2. Gas, netted through the same weekly accounting.
3. Penalties, when a settlement fails or lands late. Budget for one. Ours came twelve days after launch and cost $566.62.
4. Infrastructure: nodes, RPC, hosting and monitoring.
5. Administration: accounting, compliance, and anything else the company needs to stay in good standing.

Only what is left gets divided between the operator of record and the builder. If you read a solver's gross rewards as profit, you are off by a wide margin, and the gap is widest early, when the fixed costs are large next to everything else.

What it costs to win

This is the other number, and it is a different kind of number.

The solvers at the top of the field are market makers trading from their own inventory. Public research from Sprinter put a production-grade solver operation at five figures a month, before it earns anything, and found most solvers unprofitable. Sprinter later left the competition, saying solvers working from public liquidity cannot compete with those that have private market maker liquidity.

Money alone does not buy that liquidity. It is a relationship business at exactly the point where it matters.

So most people who try this end up in between: in the competition, technically correct, earning something, and nowhere near the top.

Two checks for any solver's numbers

Whenever you read a solver's results, ours included, do two things first.

Separate the float from the costs. Gas sitting at the submitter is working capital. It belongs on the balance sheet, not in the expenses.

Take the token out of the results. Rewards are paid in COW. When COW rises, an unchanged solver looks better. Count rewards in COW before you count them in dollars, and you see what the solver actually won.

Fixed costs get cheaper as you grow

A node, or any other fixed monthly cost, has to lift revenue enough to pay for itself. How much it has to lift depends on how much revenue there is.

Early on, almost nothing fixed pays for itself. As rewards grow, the same cost needs a much smaller lift to be worth it. Nothing about the cost changes. The business around it does. If you shelved infrastructure early because the numbers did not work, run them again.

We are running that test ourselves now, with our own nodes built on Nodevin.

Find out before you spend anything

The cheapest way to learn what a solver will earn is to test it before it goes live. cow-backtester replays real, archived auctions against your engine and shows where you would have won and lost, for free, without onboarding, bonding or funding anything.

If the answer is that your engine can compete and the company side is what is missing, that is the part we carry. Get in touch.