What Is DeFi? A Guide to Decentralized Finance

MakerDAO launched its DAI stablecoin system in 2017, letting users borrow against locked-up crypto collateral entirely through code, no bank or broker involved, and it became one of the founding building blocks of what’s now called decentralized finance, or DeFi. DeFi is a set of financial applications built on blockchains like Ethereum that aim to recreate services such as lending, borrowing, and trading without a middleman. Instead of a company approving a loan or matching a trade, self-executing pieces of code called smart contracts handle those functions automatically according to rules anyone can inspect ahead of time.

One of the most common DeFi activities is lending and borrowing through protocols like Aave or Compound. Users deposit crypto into a shared pool and earn interest paid by borrowers, who put up their own crypto as collateral to take out a loan. Because there's no credit check or human underwriter, these systems rely on overcollateralization and automatic liquidation to protect lenders if a borrower's collateral value drops too far.

Trading in DeFi typically happens through decentralized exchanges, or DEXs, which use automated market makers instead of a traditional order book matching buyers and sellers. Liquidity providers deposit pairs of tokens into a shared pool, and prices adjust automatically based on the ratio of tokens in that pool as people trade against it. In exchange for supplying liquidity, providers earn a share of trading fees, though they also take on the risk of a pool's tokens losing value relative to each other.

The appeal of DeFi is that it's open to anyone with an internet connection and a wallet, with no application, approval, or geographic restriction in the way most services impose. That openness is also its biggest risk. Smart contracts can contain bugs that get exploited for millions of dollars, and because transactions are irreversible, there's usually no customer service line to call and no way to reverse a mistake or a hack once it happens.

DeFi has grown from a niche experiment into a sector that at times has held tens of billions of dollars in deposited assets, known as total value locked. Whether that growth continues depends heavily on regulatory clarity, smart contract security improving over time, and whether ordinary users decide the tradeoffs, more control paired with more responsibility, are worth it compared to sticking with familiar centralized services.

What DeFi actually replaced

It helps to be precise about the claim. DeFi did not remove finance; it removed the intermediary’s discretion. In a bank, a human or a policy decides whether you can borrow, at what rate, and whether your account stays open. In a lending protocol, those decisions are made in advance and written into a contract that executes identically for everyone, including its authors. That is the whole substance of the word decentralised in practice, and it is genuinely new. Everything else — the yields, the tokens, the governance forums — is downstream of it.

What it did not replace is credit judgement. Because a contract cannot assess whether you will repay, it demands collateral worth more than the loan and sells it automatically if the value slips. That single constraint explains almost everything odd about the sector. It is why there is no meaningful unsecured lending, why borrowing is used mostly for leverage rather than for consumption, and why the system is procyclical: it lends most freely when prices are high and forces selling when they fall.

The five things DeFi is made of

Automated market makers replace an order book with a pool of two assets and a formula. Overcollateralised lending markets pool deposits and lend them against posted collateral at an algorithmic rate. Stablecoins supply the unit of account that makes the rest usable, and are covered in our stablecoin reference. Perpetual futures and options add leverage and hedging. Staking and liquid staking turn network security into a yield-bearing position. Nearly every protocol you will encounter is one of those five, a wrapper around one of those five, or a router that combines several.

How an automated market maker prices a trade

The classic design holds two assets in a pool and keeps the product of their quantities constant. Buying one asset removes it from the pool and adds the other, which moves the ratio and therefore the price. Nobody quotes; the formula does. The consequence is that price impact rises with trade size relative to pool depth, so the same swap can cost four basis points in a deep pool and four per cent in a shallow one. Concentrated-liquidity designs improve this by letting providers place capital in a price band, which makes depth much better inside the band and much worse outside it.

Liquidity provision is frequently sold as passive income and is better understood as writing options. If the pool’s two assets diverge in price, the pool automatically sells the winner and buys the loser, so a provider ends up with less of the asset that rose than if they had simply held both. The industry calls this impermanent loss, which is a marketing term: the loss is entirely permanent once you withdraw, and it is only offset if fee income exceeds it. Our view is that the majority of retail liquidity provision in volatile pairs has been unprofitable against simply holding, and that anyone entering it should model the fee income against the divergence rather than reading an advertised yield.

Lending markets, liquidation and the cascade

A lending market sets an interest rate from utilisation: the more of the pool that is borrowed, the higher the rate, which pulls in deposits and pushes out borrowers until it balances. Collateral is valued continuously, and if the loan-to-value ratio breaches a threshold, anyone can repay part of the debt and take the collateral at a discount. That liquidator incentive is what keeps the pool solvent without a credit department.

It also means the protocol is a forced seller at exactly the wrong moment. A sharp fall in a collateral asset triggers liquidations, which sell that asset, which pushes the price lower, which triggers more. In thin markets this ends with bad debt: collateral sold for less than the loan it backed, absorbed by the protocol’s reserve or by depositors. Every serious DeFi failure of the last five years has run through some version of that sentence. When you assess a lending market, the questions that matter are which assets are accepted as collateral, at what loan-to-value, and how deep the market for that collateral actually is.

Oracles: the seam where it usually tears

A contract cannot see a price. It is told one, by an oracle. That makes the oracle the most security-critical and least discussed component in the stack. If a protocol reads a price from a single on-chain pool, an attacker with enough capital can move that pool for one block, borrow against the false valuation and leave. The mitigations are well understood — time-weighted averages, multiple independent feeds, circuit breakers on implausible moves — and they are frequently skipped on newly listed assets, which is exactly where thin liquidity makes manipulation cheapest. If you read one section of a protocol’s documentation, read the oracle section.

Where the money actually is

On figures pulled on 3 August 2026, DefiLlama counted $74.6 billion of total value locked across 461 chains. Ethereum held $40.76 billion of it, or 54.7%. Bitcoin, worth several times Ethereum as an asset, held $3.62 billion, or 4.9%. We set out what that gap means in the Bitcoin reference. Two things follow. DeFi is overwhelmingly an Ethereum-and-its-Layer-2s phenomenon, whatever any individual chain’s marketing says, and the absolute numbers are smaller than the sector’s share of attention implies.

Total value locked is also a worse metric than it looks, and we use it with our eyes open. It double counts: a token deposited in one protocol, wrapped, and deposited in a second is counted twice. It moves with token prices rather than with usage, so a chart can rise thirty per cent in a week without a single new user. And it says nothing about whether the capital is sticky or is renting itself out for an emissions programme that ends next quarter. Read it as a rough measure of scale, never as a measure of health.

Where the yield comes from, honestly

There are exactly three sources. Fees paid by users who wanted the service. Interest paid by borrowers who wanted leverage. And tokens printed by the protocol itself. Only the first two are income. The third is dilution dressed as income, and it is the dominant component of most headline rates you will see advertised. A pool paying forty per cent in its own governance token is not generating forty per cent; it is distributing ownership to whoever shows up, and the rate falls to something ordinary the moment the emission schedule tapers or the token price stops cooperating.

The test we would apply is simple and unforgiving. Ask what the yield would be if the protocol printed nothing. If the answer is close to the risk-free rate available elsewhere, you are being paid in tokens to take smart-contract risk, and you should price that trade accordingly. This is not a reason to avoid the sector. It is a reason to stop reading advertised percentages as if they were interest.

The risk stack, in the order it has actually failed

Key and access control. In DefiLlama’s theft register, 62.7% of the $16.90 billion recorded across 610 incidents went out through a compromised key, a phished multisig approval or an access-control failure, on figures pulled on 3 August 2026. Most of what gets reported as a hack is an administrator problem, not a cryptography problem.

Bridges. Moving an asset between chains means locking it in one place and issuing a claim in another, which concentrates enormous value behind a single contract and a single validator set. Bridges have produced several of the largest individual losses in the register.

Smart contract logic. Reentrancy, rounding, and unchecked assumptions about token behaviour. Audits reduce this and do not remove it; an audit is a snapshot of one version by one firm under a time budget, and protocols are upgraded afterwards.

Governance and admin keys. A protocol with an upgradeable contract and no timelock is a protocol whose rules can change while your money is inside it. Check whether upgrades are behind a delay long enough for you to exit.

Economic design. Depegs, bad debt, and reflexive collateral where the token backing a system is the token issued by it. This category needs no exploit at all; it fails on its own terms when the price moves.

Regulatory. Front ends get blocked, developers get charged, and a protocol that is decentralised in operation may still have a company somewhere with a jurisdiction. The current picture is in our regulation pillar.

MEV, and why your trade costs more than the quote

Transactions sit in a public queue before they are included, and whoever orders them can profit from that ordering. A large swap visible in the mempool can be front-run and back-run so that the trader receives a worse price and the difference is captured by a searcher and shared with a block builder. This is extracted value, it is measured in the billions cumulatively, and it is paid by ordinary users who never see a line item for it. Slippage limits, private transaction relays and batch auctions all reduce it. Setting a generous slippage tolerance to make a trade go through is the single most expensive habit in on-chain trading, and it is the default advice in most tutorials.

On-chain versus centralised, in numbers rather than slogans

On-chain venues handled 16.4% of spot volume in the period we examined, which is a real share and not a majority. We published the method and the caveats in that measurement, alongside our failed attempt to measure centralised exchange concentration — a failure that is itself the argument for on-chain settlement, since the on-chain number can be counted by anyone and the centralised one cannot. The practical trade-offs between the two are set out in the exchanges reference.

How to check a protocol before you use it

Read the documentation for the oracle, the liquidation parameters and the upgrade path, in that order. Look up the admin address on a block explorer and confirm whether it is a multisig, how many signers it needs, and whether upgrades pass through a timelock. Confirm the contracts are verified so the code you are reading is the code that runs. Check the protocol’s own history on DefiLlama for the shape of its deposits over time rather than its current figure. Look for a Dune dashboard that shows usage, since fee revenue is harder to fake than total value locked. And treat the existence of an audit as a hygiene check rather than a safety certificate: what matters is who audited it, when, against which version, and what they excluded from scope.

Finally, use a wallet you are willing to lose and revoke token approvals when you are finished. Almost every retail loss in this sector arrives through a signature the owner did not read, which is covered in detail in the self-custody reference. Nothing on this page is investment advice, and we do not recommend protocols or accept payment for coverage.

Where this sits in the rest of our coverage

DeFi is one of seven subjects we maintain as standing reference pages. The adjacent ones are Bitcoin, stablecoins, Ethereum and Layer 2 scaling, self-custody and wallets, exchanges and tax and regulation.

Revision history

• 6 August 2026. Expanded from a short explainer into the standing DeFi reference. Added the five primitives, market-maker pricing, lending and liquidation mechanics, oracles, total value locked with its caveats, the real sources of yield, the risk stack, MEV, on-chain versus centralised share, and a verification checklist. Value-locked and theft figures are the 3 August 2026 pull.
• Original publication. Introductory guide to decentralised finance.