From Atomic to At-Risk

On-chain arbitrage felt like printing money with a safety net. You borrowed millions you didn’t have, swapped through a few pools, repaid the loan, and pocketed the difference — and if any step failed, the whole thing simply didn’t happen. That safety net had a name you probably never thought about: atomicity. This whole course is about what happens when you yank it away by crossing chains. Spoiler: every risk you thought arbitrage didn’t have comes flooding back.

Let’s start by appreciating the hero before we kill it off.

Atomicity, the guarantee you took for granted

Analogy. A same-chain atomic arb is a vending machine. You drop in a coin, the machine either dispenses your snack or spits your coin straight back. There is no universe where it keeps your coin and gives you nothing — the transaction commits in full or rolls back in full.

Definition. An atomic transaction is a sequence of state changes (borrow, swap, swap, repay) that the chain executes as a single indivisible unit: either every change is applied, or — if any sub-step reverts — none are. On the EVM, a revert anywhere bubbles up and unwinds the whole call tree.

This single property hands you four gifts, and the rest of this course is the story of losing all four:

Worked example. ETH trades at $1995 in pool X and $2000 in pool Y, same chain. You flash-borrow $1,995,000 USDC, buy 1000 ETH in X at $1995, sell those 1000 ETH in Y at $2000 for $2,000,000, repay the $1,995,000, and keep $2{,}000{,}000 - 1{,}995{,}000 =$ $5000 gross, minus gas. If pool Y had moved against you before your buy leg, the sell wouldn’t clear your repay and the transaction would revert — cost: gas only. You risked basically nothing but the gas.

When it matters

Atomicity matters most precisely when the opportunity is competitive and thin. Because downside is capped at gas, you can fire speculative bundles at slim edges and let the failures revert for free. That risk-free-retry economics is only available on a single ledger — and it’s the first thing you forfeit when you cross chains.

Two chains, no shared transaction

Analogy. Atomic arb is one cashier ringing up your whole basket in a single beep. Cross-chain arb is two stores in two cities with two cashiers and two cash drawers, connected by a courier on a motorbike. You hand over money in City A now, and a courier eventually tells City B what happened. Between “now” and “eventually,” anything can move.

Definition. A cross-chain arbitrage is necessarily two independent legs — sell on Chain A, buy on Chain B (or vice versa) — that are coordinated by a bridge message (a cross-chain communication that relays state or moves value), but are not bound into one transaction. Three consequences follow immediately:

• No cross-chain flash loan. A flash loan must repay in the same transaction; there is no single transaction here, so no one will lend you the notional risk-free. You bring your own capital.
• No cross-chain revert. If leg two becomes unprofitable, leg one has already settled on Chain A. You can’t undo it; you’re now holding inventory you didn’t want.
• A settlement gap in time. The two legs are separated by block inclusion, finality waits, and bridge relay — seconds to minutes during which you are exposed.

Worked example — the canonical trade. ETH is $2000 on Chain A and $2020 on Chain B: a gross gap of $20/ETH, or 1.0%. Atomic instinct says “buy A, sell B, done.” But you can’t. Instead:

1. Leg 1 (now, on A): buy ETH at $2000 (or sell, depending on direction) — this settles on Chain A’s clock.
2. The gap: you wait for A to include and finalize the block, then a bridge relays the message/value to B. With a 12s block time, 12 confirmations ≈ 144s of finality, plus a bridge relay ≈ 90s, for an exposure window of roughly 234s (~4 minutes) and total settlement near 250s.
3. Leg 2 (later, on B): now you execute at whatever $2020-ish has become. If ETH drifted, that 1% gap may have shrunk, vanished, or inverted.

When it matters

The settlement gap matters more the faster an asset moves and the thinner your gap. A 1% gap that takes 4 minutes to capture is fine for a sleepy stablecoin pair and terrifying for a volatile alt. We’ll quantify exactly how the gap erodes in the next lesson — but notice the mindset shift: you’re no longer asking “is there a gap?” but “will the gap survive until leg two?”

The non-atomic cousin: CEX–DEX arbitrage

Here’s the reframe that this whole lesson is building toward. Cross-chain arbitrage is not a sibling of atomic on-chain arb. Its real twin is the trade quant desks have run for years: CEX–DEX arbitrage — arbitraging price between a centralized exchange (CEX, an off-chain order-book venue) and an on-chain DEX pool.

Analogy. CEX–DEX arb is also two stores with a courier: the exchange and the chain. You keep cash in both registers (balances on the CEX and tokens on-chain), trade one side instantly, and later rebalance via a deposit or withdrawal that clears on its own slow schedule. Sound familiar? It’s the City-A/City-B courier story again — just with an exchange playing one of the cities.

Let’s line all three up across the axes that actually matter:

Read down the last two columns: CEX–DEX and cross-chain are the same animal. Both are non-atomic, capital-heavy, inventory-holding, latency-bound, and exposed to a counterparty (an exchange, or a bridge). The atomic column is the odd one out.

When it matters

If you already run a CEX–DEX desk, you have a head start: your risk plumbing — inventory limits, venue-credit caps, latency budgets, rebalancing logic — ports almost directly to cross-chain. If you come from pure atomic searching, the mental model you must unlearn is “the chain protects me.” Across chains, only your inventory management and your bridge choice protect you.

What you’re really running

Put it all together. Cross-chain arbitrage is a capital-intensive, latency-bound, tail-risky carry trade wearing an arbitrage costume.

Definition. A carry trade is a position you fund and hold to collect a spread that accrues (or is realized) over time, while bearing the risk that the underlying moves against you during the hold. Map it onto cross-chain:

• You fund it. Inventory must sit on both chains in advance. Canonical sizing: pre-position $2,000,000 on each side → $4,000,000 total locked and idle, earning nothing while it waits.
• You hold it. The spread (that $20/ETH, 1% gap) is only realized after the ~234s exposure window and ~250s total settlement — you’re long/short inventory the whole time.
• You bear tail risk. Crypto’s annualized vol ≈ 80%. Scaled to a 4-minute window via square-root-of-time, expected drift is ≈ ±0.2% — a fifth of your 1% gap, gone or doubled by noise alone. And that’s the calm case; reorgs and bridge halts are the fat tail.
• You pay carry costs. Rebalancing inventory back across the bridge costs ≈ 0.05% (5 bps) per trip, plus the opportunity cost of $4M sitting idle.

Worked example — the costume comes off. Your gross edge is the 1% gap on the traded notional. Say you move $2,000,000 of ETH: gross $0.01 \times 2{,}000{,}000 =$ $20,000. Now subtract the carry-trade realities:

The headline $20,000 is really $19,000 minus a ±$4000 coin-flip minus your funding cost — and that’s before a bridge has a bad day. Same gap, wildly different trade.

Now see the gap you’ve been reading about. Below is the at-risk window you’ll spend the next lessons quantifying. Drag the confirmation and bridge-latency sliders and watch the exposure window — and the illustrative price drift — grow. Note the atomic baseline chip: that’s the world you left behind, where the window was zero.

When it matters

This reframing matters from the very first sizing decision. If you size and hedge cross-chain arb like atomic arb (no inventory limits, no drift budget, assuming the gap is the P&L), you’ll be over-levered and under-hedged the instant volatility spikes. Treat it as a carry trade and the right questions appear on their own: How much inventory? For how long? Hedged how? Over which bridge? The remaining lessons answer each of those.

Recap

You walked in thinking arbitrage was inherently safe and walked out knowing that safety was on loan from atomicity — and that crossing chains calls the loan. Cross-chain arb keeps the name but inherits the risk profile of CEX–DEX arbitrage: capital-heavy, inventory-holding, latency-bound, and tail-risky.