AMM Design in 2026: One Formula, a Dozen Machines

In 2017, one equation described almost every automated market maker worth naming: x times y equals k. A pool held two tokens, their product stayed constant, and the price moved along a single curve as traders bought and sold. Nine years later that formula still runs underneath a large share of on-chain trading, but the machines built on top of it have drifted so far apart that they barely look related. A memecoin pool on Solana, a stablecoin pool on Ethereum, and a batch-auction pool on Balancer share an ancestor and almost nothing else. This is the story of that divergence: the small set of design choices behind every AMM, and how rival teams and chains have pushed those choices to opposite extremes.

The stakes are no longer academic. Decentralized exchanges captured a record 24% of centralized-exchange spot volume in July 2026, the highest reading since The Block began tracking the ratio in 2019, with Solana alone clearing roughly $49.86 billion in a single month. The design of the pool your order touches now decides how much slippage you eat, how exposed you are to being sandwiched, and whether the person who supplied the liquidity is quietly making money or quietly losing it. Understanding AMM design in 2026 means understanding why there is no longer a single answer.

What an automated market maker actually does

An automated market maker is a smart contract that quotes prices from a formula instead of an order book. There is no bid, no ask, and no counterparty waiting on the other side of your trade. Liquidity providers (LPs) deposit tokens into a shared pool; traders swap against that pool; the formula recalculates the price after every swap; and a fee skimmed from each trade accrues to the LPs as their reward for showing up. The whole thing runs without anyone posting or cancelling orders, which is exactly why it works on a blockchain where every action costs gas and a resting order would sit on-chain doing nothing.

Take the classic constant-product design. Imagine a pool with 100 ETH and 250,000 USDC, so the implied price is 2,500 USDC per ETH and the invariant k is 25 million. When a trader buys ETH, they add USDC and remove ETH; because the product must stay at 25 million, the pool quotes a steadily worse price as the ETH reserve shrinks. Buy a little and you pay close to 2,500; try to buy 10 ETH at once and the average price climbs well above it. That price impact is not a bug, it is the curve doing its job: the formula guarantees the pool can never be fully drained, because the marginal price approaches infinity as the last unit leaves.

The idea has a clear paper trail. Vitalik Buterin formalized the constant-product market maker in a March 2018 ethresear.ch post about making x times y equals k markets more resistant to front-running, crediting the formulation to the team at Gnosis. Hayden Adams turned it into a shipping product, launching Uniswap V1 in November 2018 with ETH-paired pools and a flat 0.30% fee, then V2 in May 2020 with arbitrary ERC-20 pairs and an on-chain time-weighted average price oracle. The appeal was immediate: liquidity was passive, listing a market was permissionless, and any contract could plug into any pool. Everything that followed is a variation on those first choices.

The four levers behind every AMM design

Strip away the branding and every AMM is a set of answers to four questions. Get comfortable with these four levers and the entire 2026 landscape stops looking like a zoo of unrelated protocols and starts looking like one design space with occupants scattered across it.

• The invariant. Which mathematical curve maps reserves to a price? Constant product is the default, but it is far from the only option.
• Liquidity distribution. Where on that curve does the capital actually sit, spread thin across all prices or packed tightly around the current one?
• The fee. How are LPs paid, how does the fee respond to conditions, and who else takes a slice before the LP does?
• The execution model. How does a trade actually clear, one at a time against the live curve or batched and auctioned to third parties who compete to fill it?

A fifth, higher-order lever sits on top: programmability and incentives, meaning whether a pool can run custom code and how its liquidity is recruited and steered. The great divergence of the last few years is simply different teams turning these dials to opposite settings, because a curve tuned for two stablecoins that never leave a penny of each other is the wrong curve for a memecoin that can 10x or collapse in an afternoon.

Lever one: the invariant that sets the price

The invariant is the heart of an AMM, and the choice among a few families explains most of the behavior a trader feels. Constant product (x times y equals k) is the generalist: it works for any pair, needs no assumption about the assets, and stays solvent no matter how lopsided trading gets, at the cost of meaningful slippage on large orders. Constant sum (x plus y equals k) is the opposite, offering zero slippage but happily letting arbitrageurs drain one side to nothing the moment the market price drifts, which makes it useless on its own.

The interesting designs live in between. Curve’s StableSwap invariant, introduced in Michael Egorov’s 2019 whitepaper, blends constant sum and constant product and tunes the blend with an amplification coefficient A. A high A keeps the curve almost flat near the peg, so a USDC/USDT swap barely moves the price, while the curve bends back toward constant product at the edges to avoid depletion if the peg breaks. Curve sets a very high coefficient for its rock-solid three-pool of USDC, DAI and USDT and a much lower one for more volatile assets. Balancer generalized in a different direction with weighted-pool math, allowing more than two tokens at arbitrary weights (an 80/20 pool, say), which turns a liquidity pool into something closer to a self-rebalancing index fund.

Lever two: where the liquidity sits

Picking the curve is only half the decision; the other half is where LPs place their money on it. In Uniswap V2 and its clones, liquidity is spread uniformly from a price of zero to infinity. That is beautifully simple and almost entirely wasteful, because an ETH/USDC pool trading near 2,500 will never touch the liquidity parked at 10 or at 100,000.

Uniswap V3, launched in May 2021, attacked that waste with concentrated liquidity. LPs choose a price range, and the protocol advertises up to 4,000x more capital efficiency for capital concentrated tightly around the market price, with each position represented as an NFT and a choice of fee tiers at 0.05%, 0.30% and 1.00%. The catch is that concentrated liquidity turns a passive LP into an active portfolio manager: when price leaves your range, your position sits entirely in the losing asset and earns nothing until you move it, which sharply amplifies the losses discussed later.

Solana’s favorite design takes concentration a step further into discrete bins. In the Liquidity Book model pioneered by Trader Joe and now most visible in Meteora’s Dynamic Liquidity Market Maker (DLMM), the price line is chopped into bins, and liquidity inside a single bin follows a constant-sum rule, so a trade that stays within one bin has zero slippage. Put simply, ticks aggregate liquidity horizontally while bins aggregate it vertically, and bins can carry their own dynamic fee that rises with volatility. The result is a market maker built for exactly the fast, jumpy assets that dominate Solana.

Concentration also has a darker cousin: just-in-time (JIT) liquidity, where a sophisticated LP spots a large pending swap, wraps it with a huge single-block position to capture almost all of its fee, then withdraws in the same block. It is legal, it is efficient, and it quietly transfers fee income from ordinary LPs to bots, which is one more reason fee design has become a battleground.

Lever three: the fee, from flat tiers to live auctions

The fee is how LPs get paid, and its design has evolved from a blunt instrument into a live control system. Uniswap V1 and V2 charged a single 0.30% on every trade. V3 introduced tiers so that stable pairs could run at 0.05% and exotic pairs at 1.00%. Neither adapts to conditions, which is a problem, because the ideal fee during a calm hour and during a violent one are not the same number.

Dynamic fees close that gap by scaling with volatility. Meteora’s DLMM raises fees automatically when its volatility accumulator spikes, and Uniswap V4 lets a pool implement the same behavior through a hook, with the Arrakis Pro hook among the first whitelisted dynamic-fee modules for token issuers. A more radical idea is the auction-managed AMM (am-AMM), proposed by Austin Adams, Ciamac Moallemi, Sara Reynolds and Dan Robinson in a 2024 paper: instead of hard-coding a fee, the protocol runs a continuous on-chain auction for the right to be the pool’s manager, and whoever wins sets the swap fee and keeps it, aligning the fee with whatever the market will bear. Bunni is among the projects putting am-AMM and dynamic fees into production on V4.

Then there is the newest slice out of the LP’s paycheck: the protocol itself. In late 2025, Uniswap governance approved the UNIfication proposal, which switched on protocol fees for the first time. For V2 pools the LP share drops from 0.30% to 0.25% while a 0.05% cut is routed to burning UNI, and the vote came with a one-time burn of 100 million UNI from the treasury. CoinDesk reported the measure passed with roughly 99.9% support. The point for design is structural: LPs are now squeezed from three sides at once, by impermanent loss, by arbitrage, and now by the protocol’s own take.

Lever four: execution, from continuous swaps to intents

The last lever is the one traders notice least and pay for most: how a trade actually clears. The default is a continuous swap. Your order hits the live curve the instant it lands on-chain, at whatever price the pool quotes in that block. That immediacy is also the single biggest surface for maximal extractable value (MEV), because bots can see your pending swap in the mempool and sandwich it, front-run it, or arbitrage the price it creates. It is worth reading our full 2026 guide to how on-chain value is extracted and reclaimed to appreciate how much of an AMM’s economics is decided at this step. The transaction you sign in your wallet is the exact moment that risk crystallizes, which is also why the same wallet layer is where drainers try to strike.

The alternative is to stop clearing trades one at a time. CoW Protocol batches orders and settles them at a uniform clearing price, with independent solvers competing to find the best execution across every available source of liquidity. Because the batch clears at one price, the ordering games that make sandwiching profitable simply stop working. Push that idea into the pool itself and you get the CoW AMM, covered in its own section below. The lesson of this lever is that the same reserves, behind the same curve, can offer a completely different trade depending on the wrapper around them.

Intents push that logic one step further. Instead of signing a specific swap against a specific pool, a user signs a desired outcome (sell this, receive at least that) and lets a competitive network of solvers or fillers work out the route, which may split across several AMMs, cross a bridge, or never touch a public pool at all. UniswapX uses signed off-chain orders and a Dutch auction among fillers to do exactly this, and emerging cross-chain intent standards aim to make the same pattern work across networks. For AMM design the implication is subtle but large: as intents mature, the pool stops being something the user picks and becomes raw inventory that solvers route through, which shifts competitive pressure from the front-end onto the curve, the fee, and the depth.

The bill every liquidity provider pays: IL and LVR

No AMM design survives contact with reality without answering one uncomfortable question: does the liquidity provider actually make money? Two costs stand between LPs and profit, and nearly every modern design decision is an attempt to shrink one of them.

The first is impermanent loss, the gap between holding two tokens in a pool and simply holding them in a wallet. When the price of the pooled assets diverges, arbitrageurs rebalance the pool at stale prices, leaving the LP with more of the falling asset and less of the rising one. The loss is called impermanent because it reverses if prices return to where they started, but it becomes permanent the moment you withdraw. The magnitude is fixed by math, not luck, as Binance Academy’s breakdown tabulates.

The second cost is subtler and, for volatile pairs, larger. Loss-versus-rebalancing (LVR, often pronounced “lever”) was formalized by Jason Milionis, Ciamac Moallemi, Tim Roughgarden and Anthony Lee Zhang. It measures what an LP loses to arbitrageurs specifically because the pool always quotes a slightly stale price that informed traders pick off. In the researchers’ worked example, an ETH/USDC constant-product pool running at 5% daily volatility bleeds roughly 3.125 basis points a day to this effect, on the order of 11% a year, as the a16z crypto research writeup lays out. Roughgarden’s framing is blunt: an AMM only works if its LPs come out ahead, so fee income has to beat LVR or the liquidity eventually leaves.

That single sentence explains most of the 2026 design race. Dynamic fees, am-AMM auctions, concentrated liquidity and batch auctions are all, at bottom, attempts to make LP revenue exceed LVR. CoW Protocol and Balancer estimate that LVR costs liquidity providers on the order of $500 million a year across DeFi, per CoW Protocol’s documentation, which is why the mechanism has moved from an academic curiosity to the number every serious pool designer now optimizes against. LPs chasing sustainable, fee-driven returns rather than token emissions are part of the same hunt for real yield that reshaped DeFi’s credit markets.

The Ethereum school: Uniswap V4 and programmable liquidity

With the levers in hand, the divergence comes into focus, and the clearest way to see it is to walk the major schools. The first is Ethereum’s, and its flagship is Uniswap V4, launched on 31 January 2025. V4 keeps concentrated liquidity but rebuilds the plumbing around three ideas: a singleton contract that holds every pool in one place, flash accounting that settles only net balances at the end of a transaction instead of moving tokens on every hop, and hooks.

Hooks are the reason V4 matters for design. A hook is a contract that runs custom code at defined moments in a pool’s life, before or after a swap, when liquidity is added or removed, and so on. That turns a pool from a fixed product into a programmable platform: the same base curve can now carry dynamic fees, on-chain limit orders, MEV-capture logic, compliance gating for institutional pools, or an entire on-chain game living inside the hook. By mid-2026, V4 pools were handling close to a third of Uniswap swap volume, with hundreds of custom hooks and thousands of hook-enabled pools deployed, according to DEXTools’ V4 guide. The Ethereum school’s bet is maximal programmability and composability, even at the cost of complexity and gas.

Uniswap’s founder is unbothered by the periodic claims that AMMs are a dead end. Hayden Adams has argued that “AMMs are only just getting started,” pointing to cheaper capital and deep composability as advantages order books cannot easily match, in comments rejecting the unsustainability thesis. V4’s hook explosion is that argument made concrete.

All that programmability carries a cost that traders rarely see: risk. A hook is arbitrary code sitting in the swap path, so a buggy or malicious hook can break a pool in ways a plain constant-product contract never could, which is why hook security review has grown into its own cottage industry. The same tension runs through the whole design space. Every lever pulled toward more capital efficiency, more customization, or more yield tends to add a new failure mode, and the maturity of AMM design in 2026 is really the growing discipline of pricing those risks rather than pretending they are free.

The Solana school: DLMM bins and raw speed

Solana starts from a different constraint and lands on a different design. Where Ethereum gas makes frequent rebalancing expensive, Solana’s low fees and high throughput make active, constantly adjusted liquidity practical, so the chain’s AMMs lean into it. Raydium is the largest venue by volume, Orca runs a concentrated-liquidity Whirlpools model, and Meteora’s DLMM has become one of the chain’s biggest venues on the strength of its bin design.

The numbers explain the emphasis. Solana led every chain with roughly $49.86 billion in 30-day DEX volume in the summer of 2026, ahead of BNB Chain, Ethereum and Base, on the reading reported by The Block. A great deal of that is fast-moving memecoin flow, and the DLMM’s bins plus volatility-scaled fees are close to purpose-built for it: LPs can shape a distribution around the active price, capture zero-slippage fills inside a bin, and get paid more precisely when volatility spikes. The Solana school’s bet is speed and capital efficiency first, deep composability second.

Solana’s design also leans hard on aggregation. Most retail orders on the chain do not go straight to a single pool; they flow through Jupiter, the dominant router that splits a trade across Raydium, Orca, Meteora and dozens of smaller venues to find the best price. That matters for AMM design, because it means individual pools compete less on brand and more on whether the aggregator’s algorithm picks them, which rewards tight spreads and deep bins over marketing. The upshot is a design culture where the pool is a component and the router is the storefront, the mirror image of Ethereum, where the pool contract itself is increasingly the platform.

The specialists: Curve, Balancer, and purpose-built curves

Not every team wants to be a generalist. The specialist school picks one asset class and shapes the entire design around it. Curve is the archetype: by choosing the StableSwap invariant and tuning the amplification coefficient per pool, it became the default venue for swapping assets that are supposed to trade near each other, from fiat stablecoins to liquid-staking tokens. Its DEX still anchors the stablecoin corner of DeFi and remains one of the deepest venues in the market for swapping pegged assets. The project extended the same thinking into lending with crvUSD and its LLAMMA engine, a lending-liquidating AMM that softens liquidations by selling collateral gradually along a curve instead of dumping it in one block.

Balancer represents the other specialist instinct: make the pool itself do more work. Balancer V3 pairs a streamlined vault architecture with a hooks framework of its own and, most notably, 100% Boosted Pools built with Aave, which automatically route idle pool liquidity into external lending markets so LPs earn swap fees and lending yield at the same time. That blurs the line between an AMM and a money market, and it is one of the clearest signs that the AMM is no longer a single-purpose object but a composable base layer other protocols build on.

A quieter branch of the specialist tree ties the curve to an external price. Proactive market makers, such as the one DODO pioneered, use an oracle to keep the pool’s quote close to the wider market price and flatten the curve right around it, which concentrates liquidity where trading actually happens without asking LPs to manage ranges by hand. The tradeoff is a new dependency: an oracle-anchored pool is only as safe as its oracle, and manipulation of that feed becomes an attack surface the pure constant-product design never had. It is another reminder that every design choice buys an advantage by taking on a matching risk.

The incentive machines: Aerodrome and ve(3,3)

Some of the most consequential divergence has nothing to do with the curve at all. It is about how liquidity is recruited. The ve(3,3) model, descended from Andre Cronje’s Solidly design, treats liquidity as something to be rented through governance. Users lock the protocol’s token to receive a vote-escrowed NFT, then vote each week to direct token emissions toward specific pools, and projects that want deep liquidity bid for those votes with incentives, still often called bribes. Lockers capture the fees and the incentives, which turns the token into a claim on real cash flows.

Aerodrome is the model’s showcase, serving as the primary liquidity hub on Base, where emissions attract LPs, deeper liquidity draws volume, and volume produces the fees and incentives that reward lockers, with the protocol directing effectively all of its trading fees to vote-escrowed holders. In this school the AMM curve is almost a commodity; the product is the incentive machine wrapped around it, and the competitive question is not which curve is best but who can rent the deepest liquidity most cheaply.

The MEV-internalizing school: CoW AMM and batch auctions

The final school changes the clearing mechanism rather than the curve, and it exists almost entirely to kill LVR. The CoW AMM, live on Balancer, is what its designers call a function-maximizing AMM. Instead of quoting a continuous price that arbitrageurs pick off, it lets solvers compete in a batch auction for the right to rebalance the pool, and the winner is whoever moves the pool’s price in the direction most favorable to its LPs. The arbitrage profit that a normal pool leaks to searchers is captured for the LPs instead, and because everything clears in a batch at a uniform price, sandwich attacks lose their footing too.

Balancer frames the pitch bluntly, citing LVR as a cost that runs into the hundreds of millions of dollars a year and positioning CoW AMM pools as the fix, per CoW Protocol’s documentation. It is the purest example of treating execution as the design frontier: the reserves and the underlying invariant can stay ordinary, and yet the LP outcome changes completely because the trade clears through an auction instead of a live curve.

The 2026 map: who trades where

Put the schools on a map and the shape of the market in 2026 is clear. Decentralized exchanges are taking a record share of spot trading, Solana leads raw volume, Uniswap remains the largest single protocol, Curve owns stablecoins, Aerodrome owns Base, and PancakeSwap anchors BNB Chain. No single design won; each dominates the niche its choices fit.

The token markets tell a more sober story about who captures the value. ETH traded near $2,500 and UNI around $6.50 to $7 in early September 2026 (CoinGecko), a reminder that leading on volume and leading on token value are two different scoreboards, even after a fee switch that burns nine figures of supply. That gap is one we explored in our look at what the market has already priced for 2026. For LPs and traders alike, the practical takeaway is that which AMM you use is now a real decision with real consequences, not a formality.

Regulation and where AMM design goes next

Design does not happen in a legal vacuum, and the most-watched question in the United States is whether an AMM with a company, a front-end, a treasury and a DAO counts as sufficiently decentralized to sit outside securities law. That question got a partial answer when the Securities and Exchange Commission (SEC) issued Uniswap Labs a Wells notice in 2024, then closed the investigation in February 2025 without an enforcement action, under Acting Chair Mark Uyeda and Commissioner Hester Peirce. The retreat did not settle the underlying doctrine, but it dramatically lowered the temperature, and it is one data point in a broader shift we track in our coverage of SEC crypto enforcement in 2026. Readers outside the US should note this is US framing; the European Union regulates the service providers and issuers around a protocol under MiCA rather than the protocol code itself.

As for where the design goes next, the trajectory is set by LP economics. Expect dynamic and auction-set fees to become the default rather than the exception, hooks and their equivalents to keep absorbing features that used to require separate protocols, intents and cross-chain solvers to make which AMM increasingly invisible to end users, and CoW-style MEV internalization to spread as LVR becomes the number every pool is judged against. The one safe prediction is that the divergence continues. There is no universal curve, only a design space, and the winning point in it keeps depending on the asset, the chain, and who is paying the bill.

Frequently Asked Questions

What is an AMM in simple terms?

An automated market maker is a smart contract that lets people trade one token for another using a math formula and a shared pool of funds, instead of matching buyers and sellers through an order book. Liquidity providers deposit tokens, traders swap against the pool, and a small fee on each trade pays the providers.

What is the difference between concentrated liquidity and liquidity bins?

Both pack capital near the current price for efficiency. Concentrated liquidity, used by Uniswap V3 and V4, relies on continuous ticks and a constant-product curve within a chosen range, while the bin model used by Trader Joe’s Liquidity Book and Meteora’s DLMM splits prices into discrete bins that each follow a constant-sum rule, giving zero slippage for trades that stay inside one bin.

What is impermanent loss and can you avoid it?

Impermanent loss is the shortfall between providing two tokens to a pool and simply holding them, caused by arbitrageurs rebalancing the pool as prices move. You cannot fully avoid it in a volatile pair, but you can reduce it by choosing correlated or pegged assets, using stablecoin pools, or providing to designs that fight the related cost of LVR.

What is LVR and why do AMM designers care?

LVR, or loss-versus-rebalancing, measures the money liquidity providers lose to arbitrageurs because the pool always quotes a slightly stale price. For a volatile pair it can run to double-digit annual percentages, so most modern AMM features such as dynamic fees, auctions and batch clearing exist to make fee income beat LVR.

Is providing liquidity to an AMM still profitable in 2026?

It can be, but it is no longer passive. Fee income has to overcome impermanent loss, LVR and, on Uniswap, a new protocol fee, so profitable providers increasingly choose the right pool design for the asset, actively manage concentrated positions, or use pools that internalize MEV. Correlated-asset and stablecoin pools remain the most reliable.

By Yuki Tanaka, senior DeFi correspondent at HOGE Wire.