What is a flash loan? Zero-collateral crypto loans

Flash loans let anyone borrow tens of millions of dollars with no collateral, no credit check, and no identity, on one condition: the loan must be repaid within the same transaction that created it. This guide explains how that is even possible, what flash loans are legitimately used for, how attackers weaponize them, and why the strangest loan in finance is also one of its most honest.

Imagine walking into a bank and asking to borrow $50 million. You have no collateral, no credit history, and you decline to give your name. In traditional finance the conversation ends there. In decentralized finance, the loan is approved instantly, at a fee of a few basis points, by a lending pool that has never heard of you and never will. The only condition is the strange one: you must pay the entire loan back before you leave the building, and if you cannot, it will be as if you never entered at all.

That is a flash loan, and it is one of the few financial primitives that genuinely could not exist before blockchains. Flash loans have funded some of the most elegant arbitrage trades in crypto and some of its most devastating exploits, including attacks that drained protocols of tens or hundreds of millions of dollars in a single block. They are cited in nearly every post-mortem of a DeFi disaster, which has given them a sinister reputation, and they are also used thousands of times a day for entirely mundane purposes that make markets more efficient.

This guide explains flash loans from the ground up: the property of blockchains that makes an uncollateralized loan safe for the lender, the mechanics of a flash loan transaction step by step, the four legitimate jobs flash loans do, a worked example of how an attack actually uses one, why the loan itself is almost never the vulnerability, what they cost and where they live, and the honest debate about whether DeFi would be better off without them.

The trick that makes it possible: atomicity

Everything about flash loans follows from one property of blockchain transactions called atomicity. A transaction on a network like Ethereum is all-or-nothing: either every operation inside it completes successfully, or the entire transaction fails and the chain’s state reverts to what it was before, as if the transaction had never been attempted. There is no such thing as a half-finished transaction. The word comes from the Greek atomos, indivisible, and it is the same guarantee that prevents a token transfer from debiting your wallet without crediting the recipient’s.

Now consider what atomicity does for a lender. In ordinary finance, a lender demands collateral because of time: between the moment money leaves the lender and the moment it returns, days or years pass in which the borrower can default, disappear, or go bankrupt. Collateral is a hostage against that passage of time. But what if the loan and its repayment happened inside a single atomic transaction? Then there is no passage of time in which anything can go wrong. The lending contract checks, at the end of the transaction, whether the borrowed amount plus fee has been returned. If it has, the transaction completes and everything inside it becomes permanent. If it has not, the entire transaction reverts, including the original lending step, and the pool’s money never actually left. The lender is not trusting the borrower; it is trusting mathematics. Default is not forbidden; it is impossible, because a defaulted flash loan is a transaction that never happened.

This is why flash loans need no collateral, no credit check, and no identity. All the machinery banks use to manage repayment risk exists because repayment is uncertain, and atomicity deletes the uncertainty. The trade-off is equally absolute: a flash loan cannot outlive its transaction. You cannot borrow via flash loan to buy something and repay next week. Whatever you intend to do with the money, borrowing it, using it, and repaying it, must all fit inside one transaction, which on Ethereum means inside roughly twelve seconds of block time and, more precisely, inside a single bundle of code execution.

Anatomy of a flash loan, step by step

A flash loan transaction is a small program the borrower writes and submits, and its skeleton is always the same five steps.

Step one, the request: the borrower’s smart contract calls a lending protocol, such as Aave, and requests a loan, say 10 million USDC. Step two, the transfer: the protocol sends the full amount to the borrower’s contract, optimistically, before any repayment. Step three, the payload: the borrower’s contract does whatever it came to do with the money, and this is the only step that varies; it might buy a token on one exchange and sell it on another, repay a loan on a second protocol, or swap collateral. Step four, the repayment: the borrower’s contract returns the 10 million USDC plus the fee, on Aave about 0.05%, so $5,000 on this loan, to the lending pool. Step five, the check: the protocol verifies the repayment landed; if yes, the transaction finalizes and the borrower keeps whatever profit the payload generated; if no, everything reverts and the borrower has lost only the network gas fee for the failed attempt.

Notice what the borrower risked: gas, typically a few dollars to a few hundred depending on complexity and congestion. Notice what the lender risked: nothing, by construction. And notice the skill that actually matters: writing the payload. Flash loans are permissionless, but they are not point-and-click for most purposes; using one means deploying a contract that orchestrates every step, which is why their users are overwhelmingly bots, developers, and MEV searchers, not casual traders, and why several services now sell no-code flash-loan tooling of wildly varying quality.

What flash loans are actually for

Four legitimate jobs account for the vast majority of flash-loan volume, and each exists because DeFi is a fragmented landscape of pools and protocols whose prices and terms drift apart.

The first is arbitrage. If ETH trades at $1,780 on one decentralized exchange and $1,790 on another, anyone who can buy on the first and sell on the second captures the spread, and the profit scales with size. Flash loans remove the capital barrier: a trader with $500 can borrow $5 million, execute both legs, repay, and keep the difference, minus fees and the slippage that large orders inevitably incur. This is the wholesome face of flash loans; arbitrageurs doing this thousands of times daily are the reason prices across DeFi venues stay closely aligned, an alignment everyone else gets for free.

The second is collateral swapping. Suppose you have a loan on a lending protocol backed by ETH and you want the collateral to be staked ETH instead, without closing the position. A flash loan lets you borrow enough to repay the loan, withdraw your ETH collateral, swap it, redeposit the new collateral, reborrow, and repay the flash loan, all atomically, converting a multi-step, risk-exposed process into one indivisible action.

The third is self-liquidation and refinancing. A borrower approaching liquidation can flash-borrow the repayment amount, close their own position, recover their collateral, sell a portion to repay the flash loan, and keep the remainder, escaping the liquidation penalty that a third-party liquidator would otherwise have taken. Similarly, debt can be moved from a protocol charging 8% to one charging 5% in a single transaction.

The fourth is liquidations themselves. The keeper bots that enforce solvency across DeFi lending, repaying underwater borrowers’ debts in exchange for their collateral at a discount, routinely fund those repayments with flash loans, which means the entire liquidation apparatus of decentralized lending runs substantially on borrowed-for-twelve-seconds money. Without flash loans, liquidations would depend on the working capital of a handful of well-funded firms; with them, anyone with a good bot can compete, which makes the system faster and more decentralized at once.

The dark side: how a flash-loan attack actually works

Flash loans became famous for the other thing they do, which is supply attack capital. The crucial point, the one every serious post-mortem makes and every headline obscures, is that the flash loan is almost never the vulnerability. It is the funding. The vulnerability is elsewhere, usually in how a protocol measures prices or counts votes, and the flash loan merely lets an attacker with pocket change exploit a flaw that would otherwise require a whale’s balance sheet.

Walk through the classic price-manipulation pattern. A lending protocol decides how much you can borrow by valuing your collateral, and suppose it values a token by checking the token’s spot price in one decentralized exchange pool. An attacker flash-borrows an enormous amount of one asset, dumps it into that pool, and momentarily distorts the price the protocol reads, making the attacker’s collateral appear far more valuable than it is. Against that inflated collateral, the attacker borrows heavily from the lending protocol, then reverses the pool trade, repays the flash loan, and walks away with the borrowed funds, all in one transaction. The protocol is left holding collateral worth a fraction of the debt against it. The flaw was the naive price oracle; the flash loan just made the flaw affordable to exploit.

The second classic pattern is governance. In 2022, an attacker used a flash loan to momentarily acquire a supermajority of Beanstalk protocol’s governance tokens, voted through a proposal sending the treasury to themselves, and repaid the loan, netting around $76 million from a protocol that allowed same-block voting. The defense, requiring tokens to be held or locked before voting power activates, is now standard, and the episode sits alongside slower-motion treasury raids in the genre of governance attacks this publication has anatomized, where the lesson is identical: if momentary token possession equals power, someone will rent the tokens for one block.

Cumulative losses to flash-loan-assisted exploits run well into the billions across DeFi’s history, with individual incidents ranging from six figures to the hundreds of millions. The defenses that matter are all oracle and logic hygiene: time-weighted average prices instead of spot readings, multiple independent price sources, borrowing caps, same-block action restrictions, and vote-locking. Protocols that do these things are not meaningfully threatened by flash loans; protocols that do not are exploitable by anyone patient enough to read their code, flash loans merely lower the capital requirement from millions to gas money.

A worked example, dollar by dollar

Abstractions hide the arithmetic, so walk through one realistic arbitrage from end to end.

Suppose a mid-cap token trades at $2.00 in a large Uniswap pool and $2.03 in a smaller pool on another venue, a 1.5% gap opened by a burst of one-sided buying. A searcher’s bot spots the spread and assembles a flash-loan transaction. Step one, it borrows 1,000,000 USDC from Aave, incurring a 0.05% fee obligation of $500. Step two, it buys the token in the cheap pool; a purchase that size moves the pool’s price, so the average fill is not $2.00 but roughly $2.006, price impact eating a slice of the edge, and the bot receives about 498,500 tokens. Step three, it sells those tokens into the expensive pool, where its own selling pushes the price down from $2.03, filling at an average around $2.022 for proceeds of roughly $1,008,000. Step four, it repays Aave 1,000,500 USDC. What remains, about $7,500, minus the gas fee, call it $150 for a complex transaction, and minus any priority fee paid to win the block position, perhaps a few thousand in a competitive auction, is profit, banked in twelve seconds with zero capital at risk beyond gas.

Now run the counterfactual that shows why atomicity matters. If, between the bot’s simulation and its execution, another searcher closed the gap first, step three returns less than the repayment requires, the final check fails, and the entire transaction reverts: no tokens bought, no loan outstanding, nothing lost but gas. The bot did not take a risk and lose; it took no risk and paid a small fee to discover that. Multiply this loop by thousands of bots and every venue pair in DeFi, and you have the invisible janitorial service that keeps a fragmented market’s prices coherent, funded entirely by the gaps it closes.

The same arithmetic explains why casual users rarely profit. The gross edge in the example was 1.5%; price impact took roughly a third of it, fees a fixed slice, and the priority auction, where competing bots bid away their expected profit to validators for first execution, takes most of the rest. Flash-loan arbitrage is a real business with real margins, and those margins have been competed down to the point where infrastructure, latency, and auction strategy decide who earns them.

Where flash loans came from

The idea is younger than it feels. The first named implementation arrived in 2018 with the Marble protocol, whose creator described it as a bank for the flash-loan age; Aave popularized the primitive at scale in early 2020, and within weeks the bZx incidents, a pair of exploits in February 2020 that used flash loans to manipulate thin markets for six-figure profits, introduced the wider world to the attack pattern and set the template for hundreds of imitations. The years since have run an arms race in miniature: attackers found protocols reading spot prices or counting same-block votes, protocols adopted time-weighted oracles and vote-locking, attack sizes peaked with nine-figure incidents, and the survivor population grew steadily harder to rob. Along the way the legitimate uses quietly institutionalized, arbitrage desks, liquidation keepers, and treasury managers folding flash loans into ordinary operations, until the primitive that headlines still call a hacking tool became, by volume, mostly plumbing.

Costs, venues, and limits

Flash loans live wherever there are large idle pools and smart contracts to guard them. Aave is the canonical venue, charging 0.05% on flash-loan principal; Uniswap V3 offers the sibling concept of flash swaps from its trading pools; Balancer, dYdX historically, and a range of smaller protocols offer variants with fees from zero to a few basis points. The size limit is simply pool liquidity: you can flash-borrow whatever the pool holds, which on major assets means nine-figure loans are routinely available to anyone. Fees flow to the pools’ liquidity providers, making flash loans a small but real income stream layered on top of the fee-and-risk economics providers already juggle.

The practical constraints are computational and adversarial rather than financial. Everything must fit within a block’s gas limit, complex multi-protocol payloads are expensive to execute and brutal to debug, and a failed transaction still burns its gas. Above all, profitable flash-loan opportunities are the most contested territory in crypto: arbitrage and liquidation payloads compete in the public mempool where sophisticated MEV searchers observe, copy, and front-run them, which is why serious operators submit through private relays and why the naive dream of easy flash-loan riches dies, for most people, on first contact with the competition.

One boundary worth drawing precisely: flash loans are native to single-chain atomicity. The guarantee dissolves across chains, because moving assets between blockchains takes minutes and trust, not one atomic breath, which is why there is no such thing as a cross-chain flash loan and why claims to the contrary should be read as marketing.

A note on nomenclature before the verdict, because two neighbors are often confused with flash loans. A flash swap is the exchange-native sibling: a trading pool sends you tokens first and lets you pay, in either asset, by the end of the transaction, which enables the same atomic patterns from a different venue. A flash mint goes further still, protocols that will create unlimited amounts of their own token for the duration of one transaction, since tokens that must be destroyed before the block closes cost the issuer nothing. All three run on the same atomicity guarantee, differ only in where the temporary liquidity comes from, and collapse into the same rule: whatever exists only inside one transaction is free to create, because it is impossible to steal.

The honest verdict

It is tempting to sort flash loans into good and evil and demand the evil half be banned, and the temptation misunderstands what they are. A flash loan is capital with the time dimension removed, and removing time removes the advantages that capital size normally confers. Whales could always manipulate thin oracles and buy governance outcomes; flash loans democratized the attack, which sounds terrible until you notice it also democratized the defense-testing. Every protocol now ships knowing that any flaw exploitable with money will be exploited by someone with almost none, a brutal but effective form of natural selection that has made surviving DeFi protocols measurably harder targets.

The equilibrium the ecosystem has reached is roughly this: flash loans as infrastructure are permanent, their legitimate uses quietly dominate their volume, their role in attacks is that of an amplifier for flaws that were always fatal, and the security burden sits, correctly, on protocol design rather than on restricting the primitive. For a user, the practical takeaways are smaller: flash loans explain how attackers with no visible wealth drain nine-figure protocols, they are part of why DEX prices track each other so tightly, and if a yield product ever advertises returns powered by flash-loan magic with no legible strategy, the magic is being performed on you.

A final perspective worth keeping: flash loans are the purest expression of what smart contracts changed about finance. Every other DeFi primitive, lending, trading, stablecoins, has a traditional ancestor it improves on; the flash loan has none, because its enabling condition, provable atomicity, does not exist in a world of couriers, clearing days, and courts. That is why traditional finance cannot copy it, why its risks are novel instead of imported, and why understanding it repays the effort even for readers who will never deploy one: it is the clearest available demonstration that programmable settlement does not just speed up old finance but permits transactions that were previously not merely impractical but conceptually impossible. Whatever else DeFi becomes, the flash loan is its signature invention, and knowing how it works is knowing what the technology is actually for.

Disclaimer: This article is for educational purposes only and does not constitute investment advice. Interacting with DeFi protocols carries significant risk, including total loss of funds. Details are current as of July 8, 2026. Always do your own research.

Frequently asked questions