Experiments · run on 11 September 2026
We built the “$68 into $750,000” arbitrage bot. It made $9.40.
The viral version: a student gives an AI coding assistant an afternoon, gets a bot that spots the same coin priced differently on two exchanges, and turns $68 into $750,000. We built that bot. Here is its trade log.
Read this first. These were paper-trading experiments: live market data, simulated money, no real orders. The figures are from our own run logs and describe what happened in a short window. They are not a forecast, a strategy or advice. Disclosures
What we built
The idea in the post is real and old: cross-exchange arbitrage. If a coin's best selling price on one exchange is above its best buying price on another, buy low, sell high, keep the difference.
Our version streamed live prices for ten coins on five exchanges — fifty markets — and checked every tenth of a second for two kinds of gap: the same coin priced differently on two venues, and three-way loops inside one venue. When a gap survived the fees on both sides, it placed a simulated trade. Three things made the simulation honest:
- Fees on both legs, at the rate an ordinary customer pays.
- Delay. The trade is priced a fraction of a second after the gap is seen, because that is when a real order would land.
- Inventory. To sell a coin on exchange B you must already hold it there. When the bot ran out, it had to wait for a transfer — ten minutes in our model — before it could trade that route again.
The third rule is the one the viral version leaves out. Without it, a simulator will happily book the same gap a thousand times.
What happened
| Setting | Trades | Traded | Profit |
|---|---|---|---|
| Ordinary customer fees (88-minute session) | 52 | $3,492 | +$9.04 |
| Zero fees — the fantasy setting (one minute) | 19 | $6,602 | −$0.06 |
| Ordinary fees, two separate trades | 2 | $200 | +$0.41 |
| All | 73 | $10,294 | +$9.40 |
One coin made all of it
Thirty-six of the 73 trades were a single coin, ATOM, mostly on one route between two exchanges. Those trades made $9.67. Everything else combined lost 27 cents. For most of the session, almost every other gap the scanner saw was smaller than the fees it would cost to trade it.
The profitable route is closed to Americans
The route that paid ran between two offshore exchanges that do not serve US customers. Of the 73 trades, two touched a US-regulated exchange. They made two cents. None ran between the two big US venues in the test.
With zero fees, it lost money
We switched fees off to see the dream version. The bot fired 19 trades in a minute, “won” 14 of them, and finished six cents down. With no fees the gaps it chases are a few hundredths of a percent wide and vanish in about half a second — by the time an order would land, the price has moved.
Why $68 does not become $750,000
Arbitrage profit is not limited by how many markets you scan. It is limited by three plain things:
- Fees. A gap has to be wider than the fees on both sides. At ordinary customer rates, that rules out nearly all of them.
- How much of the coin you already hold where you need to sell it. Each trade uses it up.
- How long it takes to move more there. Minutes, sometimes longer — while the gap closes.
Our best stretch earned about $9 in an hour and a half from one persistent gap in one small coin, and stopped each time the inventory ran dry. Firms that do this for a living pay a fraction of retail fees, keep large balances parked on every venue, and still fight over pennies.
What to take from it
The lesson is the one this whole site is built on: the fee you do not see decides the outcome. A 0.1% fee sounds like nothing until the thing you are chasing is 0.2% wide. The same arithmetic applies to slower money — a 1% trading fee or a 1% yearly custody fee compounding inside a retirement account.
Run your own numbers: the crypto IRA fee calculator shows what each provider's fees add up to over twenty years.
Method: live best bid and offer over each exchange's public data feed; simulated fills only; every trade written to a journal with the gap seen, the gap obtained, fees and slippage. Figures are totals from that journal for 11 September 2026. Published 4 October 2026.