The oldest argument in market design is venue structure, and DeFi has now run both experiments in public — an order book that a professional trader would recognise, and a formula that quotes a price without a counterparty. What is actually deployed today is neither, but a set of hybrids that exist because each pure design failed at something specific.
A constant-product pool quotes a price from its reserves, so liquidity exists at every price level without anyone posting an order. Listings are permissionless and the market never closes, which is why AMMs won the long tail of assets.
The costs are measurable. Liquidity providers take impermanent loss against the price path. Traders pay price impact that grows with size, because the formula must move along the curve to fill the trade. And the mechanism has no concept of order flow — it sees only the trade that arrives, which is the origin of most MEV on DEXs: a sandwich is possible precisely because the pool cannot distinguish a trade from the reaction to it.
Order books: liquidity from resting orders
A central limit order book quotes the true best price and lets makers post inside the spread. It wins on price for professional flow, and on information, because the book reveals intent — you can see where liquidity is and how it changes.
On-chain it is expensive, because every state transition is a transaction, so updating a quote costs gas in a way that a CEX maker would find absurd. Three workarounds have emerged. Off-chain matching with on-chain settlement, where the book lives in a sequencer and only fills hit the chain. GPU-accelerated on-chain books, where the cost of the update is engineered down rather than moved. And dedicated app chains — dYdX's v4 runs its own chain, and Hyperliquid runs its own L1 — which is the clearest statement that a serious order book needs its own blockspace to exist.
The hybrids that actually dominate
The deployed winners are neither.
Oracle-priced pools, the GMX model, quote against an external price feed rather than a curve or a book. The counterparty is a pool of LPs rather than another trader, which means the protocol must cap open interest per side to bound what the pool can lose — and the LP's return is the aggregate of everyone else's losses minus everyone else's wins.
Batch auctions, as in CoW Protocol, collect orders over an interval and match them against each other before touching an AMM, returning the surplus to traders. This removes the sandwich entirely, because there is no ordering to exploit: the batch clears at one price.
RFQ systems quote a firm price from a market maker who hedges into an AMM or a CEX, which suits size that would otherwise move a pool.
Hybrid venues expose a book for the top of book and route the remainder to pools, which is what most "order book DEXs" are when you look at the fill.
What changes when the product is perpetual
Perpetuals add constraints that pure spot design does not have.
Funding has to be computed and charged continuously, which means the venue needs a reliable index price and a settlement schedule that cannot be manipulated by the book itself. Liquidations need an engine and a backstop: on an AMM-style venue the LP pool absorbs the position, on an order-book venue an insurance fund and auto-deleveraging do. Oracle design becomes a risk parameter rather than plumbing, because liquidations fire on the mark price and the mark can dislocate from spot on a thin book.
That last point is the one that decides which venue survives a volatile day. A venue whose mark price can be pushed on thin liquidity will liquidate positions at prices that never existed on spot, and the resulting distrust is much more expensive than the fees it collected.
The decision rule
For a trader, the question is which structure is cheaper after fees, price impact, and the MEV tax — and the MEV tax is the term most often ignored, because it does not appear on a fee schedule.
For a market maker, the question is whether quotes can be updated cheaply enough to be worth posting at all. A venue where every revision costs full gas will attract only passive liquidity, and passive liquidity is what an AMM already provides.
Neither question has a permanent answer, which is why the venue debate keeps recurring: as gas costs, blockspace, and proving costs change, the boundary between what a formula does well and what a book does well moves with them.
References
Next, read Derivatives Trading 101. More in DeFi Markets.