Automated market makers versus order books

An order book matches buyers to sellers by holding a queue of resting offers and executing incoming orders against the best available prices. An automated market maker holds an inventory of two assets in a contract and quotes a price derived from the ratio between them, trading with anyone who arrives. The first mechanism discovers a price from what participants are willing to do; the second imports a price from elsewhere and relies on arbitrage to keep it current.

How a book sets a price

Participants post limit orders specifying a size and a price. The highest bid and the lowest offer define the spread, and the quantities resting at each successive level define market depth. An incoming market order walks that ladder, consuming the best level first and moving to worse prices as it exhausts each one, which is where slippage comes from on a book. Venues use price-time priority, so posting earlier at the same price earns precedence, and they charge maker and taker fees asymmetrically to reward the participant who supplies the resting quote.

The important property is that a quote is a revocable offer. Cancelling costs nothing, and in mature electronic markets the great majority of order messages are cancellations rather than trades. Displayed depth is therefore a statement of current willingness, not a stock of committed capital, and it can be withdrawn in the interval between a trader deciding to trade and the order arriving.

How a constant-product pool sets a price

The simplest liquidity pool enforces the constant product formula: the quantity of one asset multiplied by the quantity of the other must not fall after a trade, once fees are added. Suppose a pool holds 100 units of an asset and 200,000 units of a stablecoin. The marginal price is 2,000, but buying a single unit costs more than 2,000, because removing inventory from one side requires adding proportionally more to the other to preserve the product. The larger the trade relative to the pool, the worse the average execution, and the relationship is convex rather than linear. That is price impact expressed as arithmetic instead of as a queue.

A pool has no opinion and cannot decline to trade. If the wider market price moves, the pool keeps quoting the old one until someone profits by correcting it, and that correction is paid for by the liquidity providers who own the inventory. The industry name for this cost, impermanent loss, is a poor description of the economics. It measures the gap between the value of the pool position and the value of simply holding the two assets, it is not impermanent unless the price ratio returns to where it started, and its underlying cause is adverse selection: the pool systematically trades with participants who know something it does not. Fee income is the compensation, and whether it exceeds the cost is a question about flow composition, not about the formula.

Concentrated liquidity narrows the gap

Later designs let a provider allocate capital to a chosen price range rather than across every conceivable price. Concentrated liquidity makes each unit of capital far more effective inside that range, and a position with a narrow range behaves much like a stack of limit orders: as price moves through it the provider is progressively converted into the falling asset, exactly as a resting bid ladder would be. It also makes headline total value locked a weaker guide to executable size, because two pools with identical capital can offer very different depth at the current price depending on how that capital is placed. Utilization ratios such as DEX volume to TVL and volume to TVL capture some of this, though they are averages over a day rather than a measurement at the moment of a trade.

What each mechanism is genuinely better at

• Price formation. Books lead. A pool's price is a lagging function of the last arbitrage trade, which is why on-chain venues rarely set the reference price for a large asset.
• Availability. A pool quotes continuously without anyone choosing to be present. A book can empty, and it tends to empty when conditions are worst.
• Access. Anyone can create a pool for any pair on a decentralized exchange, with no listing decision. That removes a gatekeeper and also removes a filter, so a pool exists for tokens that no venue would list.
• Custody. On-chain trading is non-custodial; the asset moves from wallet to wallet. The exposure shifts to smart contract risk and to the correctness of the router being used.
• Pre-trade privacy. A pending on-chain trade is usually visible before it executes, which enables front-running and the sandwich attack, and makes MEV a running cost of trading in public. Books hide the intention until execution, at the price of trusting the operator.
• Cost structure. A book charges basis points on notional. A pool charges a pool fee plus network fees that do not scale with trade size, so small trades on a congested chain can be dominated by gas.

Who ends up holding the inventory

Both designs require someone to hold the asset nobody currently wants, and they differ in who that is and how quickly they can refuse. A book-based market maker can skew quotes, reduce size or cancel entirely within a fraction of a second, and it prices the risk of holding inventory into the spread. A pool provider cannot skew, cannot decline a trade and cannot reprice; the only available actions are to withdraw the position or, with a concentrated range, to move it, and both require a transaction to land. In calm conditions this passivity is what makes the pool a reliable counterparty. In fast conditions it is what makes the pool the last participant still quoting a stale price, and the accumulated cost of that is paid by providers rather than by traders.

The two designs keep borrowing from each other

Request-for-quote systems settle on-chain but source prices from professional firms quoting off-chain, which is a book wearing a pool's clothing. An aggregator splits a single order across many pools and books at once, so the venue a trader thinks they used is often not where the fill came from. Meanwhile centralized venues have adopted the pool idea for long-tail listings. Reading DEX volume next to total 24-hour volume shows how much of an asset's trading happens under each mechanism, which is a more useful split than the venue names themselves.

The decentralized exchange pages carry per-chain and per-protocol activity, total value locked gives the capital side of the same picture with the caveats above, and fees shows what traders actually pay to use each mechanism.