DeFi Explained: How It Works, How the Money Is Counted, and Why the Headline Number Is Not a Sum

On 6 August 2026, DefiLlama published a total value locked in decentralized finance of $75.19 billion.¹ Add up every protocol listed on the same site on the same day, excluding the centralized exchanges it also tracks, and you get $242.93 billion.³ Both figures come from one organization, from public endpoints anyone can call, within the same minute. The gap is a factor of 3.23. It is not an error, and it is not a scandal. It is what happens when the same dollar can legitimately appear in four places at once and somebody has to decide how many times to write it down. Understanding which decisions were made, and by whom, is most of what it takes to read DeFi honestly. That is what this page is for.

What DeFi actually is

Decentralized finance is the set of financial services that run as public software on a blockchain rather than inside a company. A lending market, an exchange, a derivatives venue and a savings product can all be expressed as code that holds assets, applies rules, and executes without anyone approving the transaction. The code is visible. The balances are visible. There is no account to open, no counterparty to be approved by, and—this is the part that matters when things break—no one obliged to make you whole.

The useful mental model is not “a bank without a bank”. It is a vending machine with the front panel removed. You can see the mechanism, you can verify what it will do before you use it, and if it eats your money because the mechanism was badly built, the mechanism is what you are left arguing with.

Two properties follow from that, and almost everything else in DeFi is downstream of them. The first is composability: because every protocol is a public contract, any protocol can call any other. That is why yields stack, and also why failures cascade. The second is that custody is a spectrum rather than a state. Depositing into a lending pool is not self-custody, even though no company holds your keys; you have swapped a company’s promise for a contract’s behavior. Both can fail. They fail differently.

The building blocks, in order of how much money sits in them

On 6 August 2026, the categories carrying the most value, measured across the 5,669 protocols with a non-zero balance once centralized exchanges are removed, were these.³

• Bridges — $45.15bn, 18.6%. Contracts that lock an asset on one chain so a representation of it can exist on another. This is the largest single category in DeFi and the one most people forget is a category at all.
• Lending — $41.09bn, 16.9%. Overcollateralised loan markets. You deposit an asset, borrow a smaller value of another, and are liquidated automatically if the ratio breaks.
• Liquid staking — $35.37bn, 14.6%. Deposit a proof-of-stake asset, receive a token representing the staked position, keep using the token elsewhere.
• Real-world assets — $26.99bn, 11.1%. Tokenized claims on things that exist off-chain, mostly short-dated government debt. The fastest-growing category and the one whose risk is least on-chain.
• Decentralized exchanges — $11.06bn, 4.6%. Automated market makers and order books. Note how small this is relative to its share of attention.
• Staking pools — $10.51bn, 4.3%, and restaking — $7.97bn, 3.3%, which reuse staked capital to secure additional systems.

The ranking is worth sitting with. The public image of DeFi is trading, and trading venues hold under five per cent of the money. What DeFi mostly is, by balance, is plumbing between chains and a collateralized loan book.

Total value locked, and what it is measuring

Total value locked is the headline metric and it is badly named. It does not measure value, and nothing is locked. What it measures is the current market price of the assets sitting in a set of smart contracts that somebody decided to include, converted to dollars at the moment of the query.

Three consequences follow immediately. TVL moves when prices move, even if not one token has entered or left; a chain whose native asset doubles will show its DeFi “growing” without a single new deposit. TVL says nothing about whether the capital is doing anything, so a pool that has been idle for a year and a pool turning over daily contribute identically. And TVL is entirely a function of the inclusion list, which is maintained by the aggregator, not by the protocols.

That third point is where the number stops being a measurement in the sense we use on this site. Bitcoin’s block count can be recounted from the chain by anyone who disagrees with us. TVL can only be recounted if you also reproduce the editorial decisions about what counts.

The measurement: the published total is not the sum of its parts

We pulled three DefiLlama endpoints on 6 August 2026 within the same minute: the historical chain TVL series, the per-chain snapshot, and the full protocol list.⁵ Here is what they say.

The published headline for total DeFi value locked was $75.19 billion.¹ The per-chain snapshot, summed across all 362 chains carrying a non-zero balance, came to $75.39 billion—a 0.3% difference explained by the headline being an end-of-day figure and the snapshot being live.² Those two agree.

The protocol list does not. It contains 7,996 records, of which 5,747 carry a non-zero balance. Seventy-eight of those records are centralized exchanges, which DefiLlama tracks separately and which are not DeFi by any definition; they alone account for $244.88 billion, with Binance’s custody balance at $138.29 billion. Strip every one of them out and the remaining 5,669 protocols still sum to $242.93 billion—3.23 times the published total.³

Most of that gap has a legitimate explanation, and the explanation is double counting. A dollar of ether deposited with a liquid staking protocol appears once there. The receipt token it produces is then deposited into a lending market, where it appears again. Bridge a wrapped version to another chain and it appears a third time. The chain-level aggregate nets these out. The protocol pages do not, because each protocol genuinely does hold what it says it holds.

So we tried to close the gap. Removing every category where the double counting is structural and obvious—liquid staking, staking pools, restaking, and both flavours of bridge—takes $111.45 billion out and leaves $131.48 billion. That is still 1.75 times the published figure. We could not close the remaining gap from the public endpoints, because the endpoint that would let us—a per-protocol flag marking which balances are excluded from the chain aggregate and why—is not exposed in the response.

We are not alleging anything. DefiLlama is the most transparent aggregator in this industry by a distance, it is open-source, and its methodology is published per protocol. The finding is narrower and it is about the reader, not the aggregator: the headline DeFi number is a curated aggregate, not an addition. Anyone who quotes it as “the amount of money in DeFi” is quoting an editorial judgment they have not read, and anyone who tries to verify it by adding up the parts will get a different answer and conclude, wrongly, that someone is lying.

How concentrated it is

Using the same protocol snapshot with centralized exchanges removed, the ten largest protocols hold 36.3% of the total, the top twenty hold 51.5%, the top fifty hold 73.5%, and the top hundred hold 85.4%. The Herfindahl–Hirschman index across all 5,669 comes out at 202, which on the conventional reading describes an unconcentrated market.³

That index is misleading here and we would rather say so than print it without comment. It is dragged down by a very long tail: 4,429 of those protocols hold under a million dollars each, and thousands hold effectively nothing. A market where half the money sits in twenty places out of nearly six thousand is not what most people picture when they hear “unconcentrated”. The top-twenty share is the number to carry. The index is the number that will get quoted, which is a problem we have written about at length in The Weekly Take.

The ten largest, on 6 August 2026: Lido at $17.98bn (7.40%), Aave V3 at $14.14bn (5.82%), SSV Network at $9.45bn (3.89%), Morpho Blue at $7.81bn (3.22%), WBTC at $7.35bn (3.02%), Binance staked ETH at $7.08bn (2.91%), LayerZero V2 at $6.62bn (2.73%), Coinbase Bridge at $6.21bn (2.56%), Hyperliquid Bridge at $5.93bn (2.44%) and Sky Lending at $5.66bn (2.33%).³ Four of those ten are bridges. Two are operated by centralized exchanges. That is a truer picture of what the sector is than any category chart.

Four hundred and sixty-one chains, nine that matter

DefiLlama listed 461 chains on 6 August 2026. Of those, 362 carried any balance at all, 146 held more than a million dollars, 23 held more than a hundred million, and nine held more than a billion. Those nine account for 90.5% of all chain-level TVL. Ethereum alone holds 55.1%, and the top three—Ethereum, BNB Chain and Tron—hold 68.0%.²

The multi-chain future arrived and almost nobody moved in. Four hundred and thirty-eight of those 461 chains hold less than a hundred million dollars between the lot of them relative to the total, and most hold nothing. When you read that a new chain has launched, the base rate is that it will join the 438.

Volume is a different number, and a better one

Value locked is a stock. Volume is a flow, and flows are harder to inflate on-chain because every swap is a transaction someone paid for. On 6 August 2026, DefiLlama’s DEX endpoint reported $7.34 billion of 24-hour volume across 1,295 tracked venues, $45.50 billion over seven days and $196.87 billion over thirty.⁴

This is the number we would rather you used, and it comes with a real caveat: on-chain volume can be manufactured by wash trading, which costs gas but is not impossible, particularly where a protocol is distributing a token to traders. It is still an order of magnitude harder to fake than the self-reported figure a centralized venue publishes about itself, which is the argument we made when we tried to measure exchange concentration and published the attempt as a failure.

What it actually costs to use

The advertised cost of a DeFi transaction is the swap fee, usually a few basis points to thirty. The advertised cost is rarely the cost. Four other charges apply and only one of them appears on a screen before you sign.

Gas is the network fee, it is denominated in the chain’s native asset, and it does not scale with your trade size. That single fact determines who DeFi is usable by: a fixed fee of a few dollars is trivial on a $50,000 position and prohibitive on a $200 one. It is the main reason activity migrated to cheaper chains and rollups, and the main reason the balances stayed on Ethereum.

Slippage is the difference between the price you were quoted and the price you got, and it is a function of how much liquidity sits in the specific pair you are trading, not how large the venue is. This is the single most common way a retail user loses money in DeFi without anything going wrong. Set a slippage tolerance and understand that a high tolerance is an instruction to fill your order at almost any price.

MEV—the value extractable by whoever orders transactions in a block—is the charge nobody quotes. If your trade is large enough to move a price, that movement can be captured by placing transactions immediately before and after yours. You will not see it as a fee. You will see it as a worse execution price than you expected, and you will probably blame slippage.

Impermanent loss applies if you are supplying liquidity rather than trading. It is the gap between holding two assets and pooling them, and it is only impermanent if the prices come back. They frequently do not. A pool advertising an attractive yield can still leave you behind simply holding, and the advertised yield will not tell you that.

How to read a protocol before you put money in it

This is the practical section and it is deliberately unglamorous. In rough order of how often each one turns out to matter:

• Who can change the contract, and how fast. Look for the admin key or upgrade proxy, whether it is a multisig, how many signers, and whether there is a timelock. A protocol whose owner can upgrade the contract instantly is a custodial product with extra steps.
• What the oracle is. Most large DeFi failures are oracle failures. Find out where the protocol gets its prices, how many independent sources feed it, and what happens if the feed stalls.
• Whether the audit is real. An audit is a named firm, a dated report, a defined scope and a list of findings with resolutions. “Audited” on a landing page with no link is marketing.
• What the collateral actually is. Overcollateralised by a liquid blue-chip asset is a different product from overcollateralised by the protocol’s own governance token, even though both read as “overcollateralised”.
• Where the yield comes from. There are only a few honest sources: borrower interest, trading fees, staking rewards, and off-chain income in the case of tokenized treasuries. If the return cannot be traced to one of those, it is being paid out of token emissions, which means it is being paid by the next person in.
• How old it is and what it survived. A protocol that held through a major liquidation cascade has demonstrated something no audit can.
• Whether you can leave. Check the withdrawal path, any lock-up, any exit queue, and whether the exit depends on liquidity that would not exist in a panic.

Where DeFi actually fails

The failure modes are well documented and repetitive. Oracle manipulation, where an attacker moves a thin market to make a contract believe a false price. Economic exploits that break no code but combine functions in a way nobody modeled, usually assembled inside a single flash-loaned transaction. Governance capture, where enough voting power is acquired to authorise a withdrawal. Bridge compromise, which is where the largest individual losses in the sector’s history have come from—unsurprising, given bridges are the largest category by balance. And plain key compromise of an admin multisig, which is not a DeFi failure at all but a corporate security failure wearing a DeFi label.

What is largely absent from that list is the thing most newcomers worry about: someone guessing your seed phrase. The realistic risks are structural, and the mitigation is reading the structure. We treat the custody side of this in more depth in our reference on exchanges and what proof of reserves does not prove.

What the data does not show

It does not show ownership. TVL is denominated in dollars and says nothing about how many distinct people are behind it; a single treasury can be a meaningful fraction of a mid-sized protocol.

It does not show whether the capital is productive. Idle collateral and actively rehypothecated collateral read identically.

It does not show risk. A billion dollars in tokenized treasuries and a billion dollars in a leveraged farm on a three-week-old chain both appear as a billion dollars.

It does not reconcile. As set out above, we could not make the parts sum to the published whole, and we are not able to tell you how much of the residual $56 billion is double counting we failed to identify and how much is classification we would disagree with.

And it does not show duration. A snapshot cannot distinguish capital that has been in place for two years from capital that arrived this morning chasing an incentive program and will leave when it ends.

What would change these conclusions

A per-protocol exclusion flag in the public API, stating which balances are omitted from the chain aggregate and on what basis, would let anyone reconcile the headline to the parts. That single field would convert TVL from a curated aggregate into a measurement, and we would say so.

A widely adopted net-of-double-counting standard—published, versioned, and applied identically across aggregators—would do the same at industry level.

On concentration, a sustained move in the top-twenty share of more than five percentage points across two quarterly re-runs would be a structural change rather than noise, and we would write it up as one.

On chains, the number to watch is not how many launch. It is whether the count holding more than a billion dollars moves off nine.

Where this sits in the rest of our coverage

This page is the standing reference for decentralized finance. It is re-run quarterly at the same URL, with the revision date and a changelog line rather than a new post, because publishing a near-duplicate every quarter would split the authority this one earns.

• Stablecoins explained — the settlement layer most DeFi activity is denominated in.
• Crypto exchanges explained — the custodial side, and what proof of reserves does not prove.
• Layer 1 versus Layer 2 — where the 461 chains come from.
• On-chain venues handled 16.4% of spot volume — the DEX-to-CEX ratio, published as a floor rather than an estimate.
• Bitcoin explained — the asset that is 3.5 billion dollars of this total and none of its ideology.

Next measurement from this page: the share of DeFi TVL held in the top ten protocols, re-run on 6 November 2026, with the reconciliation attempt repeated and the residual reported either way.

Revision history

6 August 2026 — first publication. All figures pulled from DefiLlama public endpoints on 6 August 2026. Next scheduled re-run 6 November 2026.