Atomic Bundles & Flash Loans

You found a 0.5% loop across three pools. Now comes the terrifying middle: you’ve bought the first leg, you’re holding inventory, and the price could move against you before you close the trade. On a normal exchange that gap is where dreams go to die. On-chain, there’s a trick that makes the gap literally impossible — and a second trick that means you don’t even need the money to do it. This lesson is about both.

The bundle: one transaction, all legs

Think of a relay race where the runners are handcuffed together: either the whole team crosses the line or none of them move at all. There’s no scenario where one runner finishes and the rest are left stranded mid-track. That’s atomic execution.

An atomic bundle packs every leg of your arbitrage — buy on A, sell on B, and any extra hops in between — into a single transaction that the EVM runs as one indivisible unit. The contract executes the legs in order, then checks: did I end up with more than I started? If yes, the transaction commits and you keep the profit. If no, the transaction reverts — every state change is undone as if nothing happened.

The sequence inside one atomic transaction looks like this:

1. Buy ETH on pool A (cheap).
2. Sell that ETH on pool B (dear) — and any intermediate hops if it’s a triangular loop.
3. Check the final balance: is profit greater than zero (after costs)?
4. If yes, commit. If no, revert() — the buy on A is unwound too, and you never held the inventory.

The magic is step 4. Because the whole thing reverts on a failed profit check, you can never get stuck having done leg one but not leg two. The searcher never holds inventory between blocks, never wakes up to a position that moved against them. The risk that haunts every off-chain trader — “I bought, then the market ran away before I could sell” — is engineered out of existence.

When it matters

Atomicity is what turns a fragile multi-step arbitrage into a single safe bet. It’s why on-chain searchers can chase razor-thin spreads that no off-chain trader would touch: the downside of a missed leg isn’t a loss, it’s a wasted gas fee. Every strategy in the rest of this course assumes the legs are bundled — it’s the bedrock.

Flash loans: rent the whole stake for one block

Imagine a bank that hands you a million dollars at the door, on the condition that you walk a loop through the building and hand it all back — plus a small tip — before you reach the exit. If you can’t return it by the exit, the security cameras rewind and it’s as if you never walked in. No collateral, no credit check, because the bank physically cannot lose: the door won’t open until you’ve repaid.

A flash loan is exactly that. You borrow a large sum with no collateral, on the sole condition that it’s repaid before the same transaction ends. If repayment fails, the entire transaction reverts — the loan, your trades, all of it — as if it never happened. Because repayment is enforced atomically (same trick as the bundle), the lender takes on zero default risk and can lend to a complete stranger.

The economic consequence is enormous: capital stops being the barrier to entry. You no longer need a war chest to run a big arbitrage. You borrow the entire stake for the duration of one block, run your atomic loop, repay the loan out of the proceeds, and keep whatever’s left. What decides the winner is no longer how much money you have — it’s speed and search quality: who spots the opportunity and lands the bundle first.

Lenders don’t do this for free. A typical flash-loan fee is 0.05% — that’s 5 basis points — charged on the borrowed amount and paid back alongside the principal inside the same transaction.

When it matters

Flash loans democratized MEV. A teenager with a clever bot and a laptop can deploy the same capital as a hedge fund — for the length of one transaction. That’s why on-chain arbitrage competition is brutal and margins are thin: the moat that protected the well-capitalized everywhere else simply doesn’t exist here.

Worked: capital-free arbitrage

Let’s run the numbers from the triangular-arbitrage loop you priced earlier, now financed entirely with borrowed money. You bring almost none of your own capital to the table.

Walk it through: you borrow $100,000 of USDC with no money down. You run the atomic loop and it returns 0.5%, which is $500 gross. You repay the principal plus the 0.05% fee — that fee is $50 on a $100,000 loan. What’s left in your pocket, before paying gas, is about $450.

The headline isn’t the $450 — it’s whose $450 it is. You captured a $450 edge having staked almost none of your own capital. The flash loan supplied the entire $100,000; your only out-of-pocket cost is the gas to submit the transaction. That is the whole reason on-chain arbitrage looks nothing like its off-chain cousin: the capital requirement collapsed to roughly zero.

When it matters

This is the math that explains why MEV is a high-volume, low-margin grind. When capital is free, everyone can chase the same $450, so the edge gets bid down to the thinnest sliver that still beats gas and the loan fee. The winner isn’t the richest — it’s the one whose bundle is cheapest to run and fastest to land.

What atomicity does NOT protect you from

Here’s where overconfidence kills accounts. It’s tempting to read “atomic = the whole thing reverts if it’s not profitable” as “atomic = risk-free.” That’s only half true, and the missing half is where searchers actually lose money. Atomicity is a fortress against exactly one danger — inventory risk — and offers nothing against the others.

Three risks survive atomicity entirely:

1. Gas on a reverted attempt. Revert refunds your trades, not your gas. Every failed bundle still burns the gas spent computing it. Fire a hundred speculative attempts that all revert and you’ve paid a hundred gas bills for zero profit.
2. Outbid or not included. A block builder chooses which bundles go in and in what order. A competitor offering a juicier payment can outbid you, or the builder may simply leave your bundle out. Atomicity guarantees your bundle is correct, not that it’s chosen.
3. The opportunity vanishes. If a competitor’s transaction lands first and changes the pool state, the gap you targeted may be gone by the time your bundle runs. Your profit check then fails — correctly — and you revert, having paid gas to discover the edge evaporated.

So atomicity kills inventory risk stone dead. It does nothing about competition risk or gas-on-failure. Confusing “my trade can’t get stuck half-done” with “my trade can’t lose” is the classic rookie error.

When it matters

This is the line between a searcher who survives and one who quietly goes broke. The survivor prices in gas-on-failure and competition, simulates aggressively before submitting to avoid doomed bundles, and only fires when the expected win clears the expected cost of trying. “It’s atomic, so I can’t lose” is the epitaph on a lot of dead bots.

Recap