What does a crypto arbitrage bot actually do?
An arbitrage bot buys an asset where it is cheap and sells it where it is expensive, pocketing the difference. In crypto the classic forms are cross-exchange arbitrage (Bitcoin trades $0.20 higher on Kraken than Coinbase) and triangular arbitrage (a loop like USDT → BTC → ETH → USDT that ends with more USDT than you started).
The appeal is obvious. It looks market-neutral: you are not betting on direction, only on a gap that already exists. In theory you profit whether the market pumps or dumps.
Reality is harsher. A quoted gap is not a captured gap. Between seeing the price and getting both legs filled, the spread can vanish, one leg can slip, or the coin you need to move sits in a confirmation queue. The whole strategy lives or dies on the costs and delays between "I see the edge" and "both trades settled."
Cross exchange arbitrage: where did the edge go?
Cross exchange arbitrage means holding balances on two or more venues and trading the same pair when their prices diverge. The mechanics are simple; the problem is that everyone can see the same order books.
In crypto's early years, gaps of 1-5% between exchanges were common, and the famous "Kimchi premium" saw Bitcoin trade several percent higher on Korean exchanges for weeks. Those gaps existed because moving capital was slow and fragmented. In 2026, the top pairs on major centralized exchanges stay within a few basis points of each other almost all the time, because thousands of bots — many co-located next to the matching engine — arbitrage them continuously.
Two structural costs make cross-exchange arb hard for retail:
- You must pre-fund both sides. To sell BTC on Exchange A and buy on Exchange B at the same moment, you need BTC already sitting on A and USDT already on B. You cannot buy cheap and then ship the coin over in time.
- Rebalancing costs money. After enough trades your inventory drifts to one venue, and you must withdraw and redeposit, paying network fees and waiting for confirmations. Each rebalance eats the thin margins you just earned.
Speed decides everything here, and speed depends on how your orders reach the matching engine. Latency you can measure and cut with better hosting is one thing; the microseconds a co-located market-making desk enjoys is another game entirely.
Triangular arbitrage crypto: same problem, one venue
Triangular arbitrage crypto trades three pairs on a single exchange to exploit a pricing inconsistency — for example USDT → BTC → ETH → USDT. Because all three legs happen on one venue, you skip withdrawals and transfer delays entirely, which removes the slowest cost in cross-exchange arb.
That sounds better, and in one way it is. But you have swapped transfer risk for execution risk. All three legs must fill at the quoted prices, and each leg pays a taker fee. Three taker fills at 10 bps each is 30 bps of cost before you have made a cent, and triangular mispricings on liquid pairs are usually smaller than that. On thin pairs where gaps are larger, your own order moves the book, so the price you saw is not the price you get. Professional desks run these loops in microseconds; a retail bot polling a REST API is far too slow to be first.
What costs eat the arbitrage spread?
Every arbitrage number should be quoted net of costs, not gross. Here is what stands between a visible gap and realized profit.
| Cost | Typical size | Why it hurts |
|---|---|---|
| Taker fees | 2-10 bps per fill | You usually pay taker on at least one leg; multiple legs multiply it |
| Maker rebates | -1 to +2 bps | Helps, but a resting order may not fill before the gap closes |
| Withdrawal / network fees | Fixed per transfer | A flat cost crushes small cross-exchange trades |
| Transfer time | 1-30+ min | Coin in transit is exposed; the gap can close or reverse |
| Slippage | Varies | Thin books mean your fill is worse than the quote |
| Latency | Milliseconds matter | A slower bot arrives after the gap is already gone |
A single BTC arbitrage that shows a 15 bps gross gap can easily turn negative after two taker fees and a little slippage. A working rule of thumb: if your gross edge is not at least 2-3x your round-trip cost, assume it disappears in practice.
Is crypto arbitrage profitable in 2026?
Honestly, for most retail traders, no — not the pure cross-exchange or triangular versions on major pairs. The gaps are too small and too fast, and you are competing with firms that pay for co-location and see the book before you do. This is not a "just find a better bot" problem; it is a structural speed disadvantage. If a strategy's entire edge is being first, and you are not first, there is no edge.
That framing matters because arbitrage is a favorite hook for fraud. "Guaranteed 1% daily from risk-free arbitrage" is mathematically absurd once you understand the costs above — real arbitrage edges are tiny, contested, and shrinking. Any product promising fixed daily arbitrage returns is selling a story, not a strategy, and it sits near the top of most trading-bot scam red-flag lists for good reason.
Where arbitrage-like edges do still exist, they tend to be slower and messier — which is exactly why the fast desks ignore them.
What arbitrage still works for smaller operators?
The surviving edges are the ones that trade a real risk or friction, not pure speed:
- Funding-rate arbitrage. Hold spot and short the perpetual (or the reverse) to stay delta-neutral while collecting the funding payment. When funding is persistently positive, shorts get paid each interval. It is slower and capital-heavy, and it carries liquidation and basis risk, but it is not a pure latency race. Understand how funding rates actually work before trying it.
- On-chain / DEX arbitrage. Price differences between DEX pools, or between a DEX and a CEX, can be larger — but you fight gas fees, MEV bots, and failed transactions. The trade-offs between on-chain and CEX bots differ meaningfully.
- Market-making-adjacent flow. Quoting both sides and capturing the spread is a cousin of arbitrage, and a market making bot rewards patience and inventory management more than raw speed.
None of these are free money. Each swaps the "be fastest" requirement for a different risk you actually have to manage.
How do you test an arbitrage bot before funding it?
Paper-test with real costs baked in, then run tiny. Most arbitrage backtests look brilliant because they quote gross spreads and assume instant fills. Both assumptions are false.
- Log live order books from both venues and replay them, subtracting realistic fees, slippage, and transfer time on every simulated trade.
- Measure your actual round-trip latency from signal to fill, not the exchange's advertised numbers.
- Start with the smallest size the exchange allows, then compare realized PnL to your simulation. If reality is far worse than paper, your model ignored a cost.
Arbitrage is one of the strategies most flattered by careless simulation, so treat a clean-looking result with suspicion — it is a textbook case of a backtest that looks great, then dies the moment it meets real fees and real queues.
Frequently asked questions
Is crypto arbitrage still profitable in 2026?
For most retail traders, pure cross-exchange and triangular arbitrage on major pairs is not profitable after costs. The gaps are a few basis points and close in milliseconds, and professional desks with co-located servers capture them first. Slower variants like funding-rate or on-chain arbitrage still offer edges, but with real risk.
What is the difference between cross-exchange and triangular arbitrage?
Cross-exchange arbitrage trades the same pair across two venues, so you fight transfer times and withdrawal fees. Triangular arbitrage trades three pairs on one exchange (like USDT to BTC to ETH back to USDT), avoiding transfers but paying a fee on every leg. Both need speed most retail bots lack.
How much money do you need to start arbitrage trading?
There is no fixed minimum, but small balances are a real disadvantage. Fixed withdrawal and network fees consume a larger share of tiny trades, and pre-funding balances on multiple venues ties up capital. Below a few thousand dollars, flat costs usually swallow the thin margins arbitrage produces.
Are guaranteed-return arbitrage bots real?
No. Real arbitrage edges are small, contested, and shrinking, so any bot advertising a fixed daily return from "risk-free" arbitrage is a red flag. Genuine arbitrage cannot guarantee a rate, and promises like 1% per day are a common structure for scams and Ponzi schemes.
Is crypto arbitrage risk-free?
No. It is often called market-neutral, but you still carry execution risk (a leg not filling), transfer risk (a gap closing while coins move), counterparty risk (an exchange freezing withdrawals), and slippage. On funding or on-chain variants you also add liquidation, basis, and smart-contract risk.