MEV Strategies in 2026: Off the Mempool, Onto the Sequencer

Six years after a group of researchers gave it a name, maximal extractable value is neither a secret nor a rounding error. Cumulative MEV on Ethereum runs well past a billion dollars, and the bots that collect it are now among the most sophisticated trading operations in the world (DEXTools). What changed in 2026 is not the size of the prize. It is the address. The public mempool, the open waiting room where pending transactions once sat exposed to anyone willing to reorder them, has largely emptied out. The value is still being extracted. It has simply moved to places most traders cannot see.

That migration is the story this guide tells. The strategies themselves (arbitrage, liquidations, the notorious sandwich) are mostly unchanged, and we will recap them. But the arena has shifted to three new venues: the single sequencer that orders every transaction on a Layer 2, the solver networks that now execute a large share of DeFi swaps off-chain, and the applications that have started keeping their own MEV rather than leaking it. Each venue promised to tame extraction. Each, so far, has reproduced the same pattern: a handful of professional players capturing most of the value, and ordinary users paying a toll they never see on the receipt.

MEV in one paragraph, for anyone who skipped the first eight explainers

MEV is the profit a party can earn by choosing which transactions go into a block and in what order. The term dates to Flash Boys 2.0, the 2020 paper by Phil Daian and co-authors that documented front-running on decentralized exchanges and warned that the race to capture ordering profits could threaten consensus stability (par.nsf.gov). It originally stood for miner extractable value; after Ethereum moved to proof of stake, the M was redefined as maximal, because proposers, builders, and searchers now all share the spoils.

Modern extraction runs through an assembly line. A searcher spots an opportunity (a mispriced pool, a loan about to go underwater) and bundles the transactions that capture it. A builder assembles those bundles, along with ordinary user transactions, into a full block ordered to maximize value. A relay passes the block to the validator whose turn it is to propose, and that proposer signs whichever block pays the most. Ethereum’s own documentation walks through the canonical case, a decentralized-exchange arbitrage that turns 1,000 ETH into 1,045 ETH in a single atomic transaction (ethereum.org). The mechanics of that supply chain are the subject of several of our earlier pieces; here we care about where the searchers now hunt.

The classic strategy menu

The strategies sort into roughly seven buckets, running from benign and arguably necessary to openly predatory. Keeping them straight matters, because the public tends to treat all MEV as theft, while builders tend to treat all of it as a free market. Most of it sits in between.

Arbitrage and liquidations are the load-bearing kind: without bots racing to arbitrage pools and close bad loans, decentralized exchanges would drift off-peg and lending markets would accumulate bad debt. The sandwich is the strategy the public rightly hates. A bot sees a large pending swap, buys the same asset just ahead of it to push the price up, lets the victim fill at that worse price, then sells immediately after into the bump. A specialized cousin, oracle extractable value, turns the price feed itself into the trigger rather than a pending trade; we gave that its own treatment in our piece on the exploit DeFi made legal. Everything below is about where these strategies are now executed, because the answer has changed more than the strategies have.

Why the public mempool stopped being the arena

For most of Ethereum’s history this all played out in the open mempool. Dan Robinson and Georgios Konstantopoulos of Paradigm gave the dynamic its enduring name in 2020: Ethereum is a dark forest, a place where revealing a profitable transaction invites an instant predatory response from bots that detect, copy, and front-run it within milliseconds. For years, defending yourself meant not broadcasting anything worth stealing.

Two shifts emptied the forest. The first is private order flow. More than 90% of Ethereum blocks are now assembled through MEV-Boost, and a growing share of user transactions reach builders through private routes (Flashbots Protect, MEV Blocker, and similar relays) rather than the public mempool. If a predator cannot see your transaction until it is already in a block, it cannot pounce on it there. The second shift follows from the first: once the open mempool thinned out, the game stopped being see a victim and pounce and became control the privileged point where ordering is decided. That point is no longer one place. It is three.

The rest of this guide walks through each venue, then returns to the one thing they all share: underneath every clever ordering scheme, blocks still settle on Ethereum, where a few builders decide the final sequence.

Venue one: the Layer 2 sequencer

The large majority of DeFi activity now happens on Layer 2 rollups, not on Ethereum mainnet. That single fact relocated most MEV, because almost every major rollup runs exactly one sequencer, usually operated by the core team, that receives transactions and decides their order before posting them to Ethereum. A single operator choosing the order of every trade on a chain is, by definition, an ordering monopoly. The only question is what rule it uses.

Two policies dominate. First-come-first-served, which Arbitrum used for years, turns MEV into a latency race: whoever gets their transaction to the sequencer first wins, so professional searchers co-locate their servers next to it and spam it with attempts, which is wasteful but keeps ordinary fees low. Priority-fee ordering, used by Optimism and Base, turns it into an auction: transactions are ranked by the tip they pay, so the sequencer, not a searcher, captures much of the priority revenue. Either way, the sequencer earns real money on top of the spread between the Layer 2 fees it collects and the Ethereum costs it pays. A centralized sequencer could in principle front-run its own users, which is why rollup teams publish commitments not to, and why the question of who gets to order L2 transactions has become one of the most contested in the ecosystem. Arbitrum’s response was to stop pretending the ordering right was free and to sell it outright.

Timeboost: an auction for the front of the line

Arbitrum launched Timeboost in April 2025. Rather than leave ordering to a latency race, it auctions an express lane. A sealed-bid, second-price auction runs every minute and sells one party the right to a short head start, about 200 milliseconds, ahead of everyone else for that round; transactions outside the express lane still flow first-come-first-served. The pitch had three parts: internalize MEV revenue for the Arbitrum DAO instead of leaking it to latency bots, remove the incentive to spam the sequencer, and keep ordering predictable.

The early revenue looked healthy. Timeboost generated $2,491 for the DAO on its first day (Blockworks) and roughly $2 million in fees within three months of launch (The Block). It is now one line in an Arbitrum DAO that reported about $6.19 million in first-half 2026 income across fees, Timeboost, and treasury returns (BigGo Finance).

Then the audit arrived. An empirical study of Timeboost’s record, covering 48.5 million express-lane transactions across roughly 494,000 auctions, found the lane had reproduced exactly the concentration it was meant to dissolve: three entities won 99.74% of all auctions (arXiv). About 30% of express-lane transactions reverted, meaning the spam did not disappear, it moved into the fast lane. And the dominant use was not fairer retail access but CEX-DEX arbitrage, the same latency-sensitive strategy that already dominates mainnet extraction. The authors concluded that Timeboost fails on its stated goals of fairness, decentralization, and durable long-term revenue.

The lesson rhymes with Ethereum mainnet: when you sell priority, the parties who can pay most for it are the professionals who already run the fastest infrastructure. Moving the ordering right onto the L2 and auctioning it did not democratize MEV. It built the front of the line an oligopoly of its own.

Venue two: solvers and the intent economy

The opposite approach keeps users out of the ordering game entirely. Intent-based trading flips the model. Instead of signing a specific transaction (swap exactly this for that, on this pool, at this gas price), the user signs an intent: I want at least this much of that for my this, you work out how. A competitive network of solvers (also called fillers or resolvers) races to satisfy the intent, and the order never sits in a public mempool where it could be sandwiched.

Three systems dominate. UniswapX uses a Dutch auction over off-chain signed orders; an open network of fillers competes to fill them, covers the gas, eats the cost of failed transactions, and, crucially, cannot sandwich orders it fills from its own inventory, while arbitrage that would otherwise be extracted is returned to the swapper as a better price (Uniswap). CoW Protocol runs batch auctions that settle many orders at a single uniform clearing price, so there is no intra-batch ordering left to exploit; where two users want opposite sides of the same pair, they are matched directly in a coincidence of wants without ever touching an automated market maker, and bonded solvers compete to produce the batch solution with the most surplus (CoW docs). 1inch Fusion blends request-for-quote with a Dutch auction and settles through resolvers off the public mempool.

The volume is no longer marginal. CoW Protocol alone settled about $23.3 billion across 2.34 million settlements in 2026, with cumulative volume past $80 billion (Bitquery), and two to three dozen solvers were active on Ethereum by late 2026. Here is the catch that keeps this an MEV story rather than an MEV cure: the solver is the new searcher. Solvers still capture MEV; they just do it on the user’s behalf and, in the best case, hand most of the surplus back. The unresolved question is how much they keep, and whether solver competition stays open or collapses into the same few professional desks that already dominate block building.

Venue three: the app keeps its own MEV

The third answer is for the application to capture the MEV its own users generate and recycle it, rather than letting external searchers skim it. Uniswap v4’s hooks let a pool run custom logic at swap time, which opens the door to application-specific ordering. Angstrom, built by Sorella Labs on top of those hooks, runs an app-specific sequencer that captures the arbitrage value a pool would otherwise leak and returns it to the pool’s liquidity providers (Sei research). Unichain, Uniswap’s own Layer 2, uses trusted-hardware priority ordering with a similar goal. Uniswap’s UNIfication proposal goes a step further, selling temporary fee discounts and burning the proceeds (Uniswap).

The common thread is a change of posture: MEV is treated as a resource the application produces, not a leak it suffers. Whether that is fairer or simply relocates the rent from anonymous searchers to the app’s own token holders is the live debate. It is genuinely better for liquidity providers, who have spent years subsidizing arbitrageurs. It does nothing, by itself, to stop a sandwich on a user who routes around the protected path.

Who actually assembles the block

All three venues eventually settle to Ethereum mainnet, and there the old proposer-builder-separation supply chain still rules. Around 90% of Ethereum blocks are assembled through MEV-Boost, and building is dangerously concentrated. Live data from relayscan.io in October 2026 shows one builder, Titan, producing roughly half of all blocks, with Quasar and BuilderNet accounting for most of the rest and the top three together close to 90% of supply. The market’s concentration index sits in territory a competition regulator would call moderately concentrated. BuilderNet, a multi-operator builder running inside trusted hardware (Flashbots, Beaverbuild, and Nethermind), was launched specifically to dilute that dominance by refunding value in proportion to the order flow each party contributes; it has gained share without breaking the pattern. We traced the economics of that concentration, and who profits from it, in our companion piece on the MEV oligopoly.

At the very end of the line sit the validators, who increasingly run through professional node operators and distributed-validator setups rather than solo machines; we profiled that operator layer and its economics in our look at the people who run SSV restaking. The point for this guide is blunt: however clever the L2 or solver layer becomes, the final ordering of value still settles through a chokepoint controlled by a few firms. That is why so much hope now rests on the protocol-level fixes at the end of this piece, and why some rollups want out of the sequencer business altogether.

Based rollups: give ordering back to Ethereum

Not every rollup wants its own sequencer. A based rollup hands sequencing back to Ethereum’s own validators: the next mainnet proposer, working with Ethereum builders and searchers, permissionlessly includes the next rollup block as part of its mainnet block. Taiko is the largest live example (Four Pillars). The tradeoff is explicit. A based rollup inherits Ethereum’s liveness and censorship resistance and needs no trusted sequencer to go down or misbehave, but the MEV its users generate flows up to mainnet, to Ethereum builders and proposers, rather than staying on the Layer 2. You trade sequencer risk for lost revenue.

Shared sequencers aim for a middle path: a separate decentralized network that orders transactions for many rollups at once, so no single rollup team holds the ordering monopoly. Espresso’s network, which pairs its HotShot consensus with a data-availability layer and finalizes blocks in seconds, went live and began publishing sequencing confirmations for rollups across the Arbitrum and Celo ecosystems during 2026, with a fully permissionless version targeted for later in the year (Espresso). Astria pursues a similar design. The promise is credible neutrality, where no single team decides your order. The open question, familiar by now, is whether a shared sequencer simply becomes the next concentrated chokepoint with better branding.

Cross-domain MEV: the arbitrage that spans chains

As liquidity fragments across dozens of Layer 2s, the richest opportunities increasingly span domains. The same asset trades at slightly different prices on Arbitrum, on Base, and on a centralized exchange in the same instant, and whoever can act across all three fastest captures the spread. This is cross-domain MEV, and it is harder to observe and harder to regulate than anything confined to one chain, because no single block tells the whole story.

The plumbing that makes cross-chain trading smooth for users makes cross-domain extraction smoother too. ERC-7683, a cross-chain intents standard introduced by Uniswap Labs and Across (announcement), lets a single solver fill an order that originates on one chain and settles on another; by 2026 it was supported across Across, UniswapX, Eco, and dozens of other projects. The same solver that shields a user from a sandwich on one chain may be running a cross-domain arbitrage on the back of that user’s flow. Non-atomic arbitrage between decentralized and centralized exchanges remains the single largest and least visible category of extraction, which is precisely why Timeboost’s express lane filled up with it.

What it actually costs you

For a retail trader, MEV is an implicit tax: a few basis points of worse execution on every swap, invisible because it is priced into the trade rather than itemized on a receipt. It is the crypto analogue of payment for order flow in equities, value skimmed from the gap between the price you could have gotten and the price you did. On a bad trade, a large market swap pushed through a thin pool with no protection, a sandwich can cost far more than the visible fee.

The most-screenshotted illustration of 2026 was comic. JaredfromSubway, Ethereum’s best-known sandwich bot, wrapped a roughly $4 token swap by Vitalik Buterin inside more than a million dollars of its own volume and walked away with a few cents (CoinDesk). The same bot was later drained of about $7.5 million in an ironic exploit. The takeaway for everyone who is not Vitalik: you are usually not worth sandwiching, right up until you are.

Protecting yourself is now genuinely easy, which was not true two years ago.

• Route transactions through a private RPC such as Flashbots Protect or MEV Blocker, so your trade never sits in the public mempool.
• Trade through an intent-based venue (CoW Swap, UniswapX, or 1inch Fusion) and let a bonded solver absorb the ordering risk.
• Use a wallet that defaults to protected routing; the better consumer wallets now do this without the user asking.

Those defaults matter more than any single tool, because most users will never change a setting. We mapped where the wallet funnel still leaks in our onboarding breakdown, and what it means when wallets start routing and even transacting on your behalf in our look at the wallet that spends without you. For small swaps through deep pools, the MEV risk is minor. For large swaps through thin liquidity, it is not, and the protection is a single dropdown away.

Solana’s parallel universe

Solana never had a public mempool to empty, so its MEV story took a different shape. Transactions flow straight to the current leader, and extraction runs through Jito, whose client runs on the large majority of stake and operates an off-chain auction where searchers submit bundles with tips for priority inclusion. JitoSOL, a liquid staking token, recycles a share of those tips back to stakers, which turns MEV into a visible component of yield rather than an invisible tax, one of the clearer examples of the redistribution instinct now spreading across chains.

Solana’s hard line on the toxic end of the spectrum is also instructive. In 2024 the Solana Foundation removed more than thirty validators from its delegation program for running private mempools that enabled sandwich attacks on retail users; Tim Garcia, who leads validator relations at the Foundation, described the enforcement as final for repeat offenders (CryptoSlate). SOL trades around $111.66 as of this writing, with a market capitalization near $65.8 billion (CoinGecko), and the Jito-centered MEV layer is a meaningful part of why staking it pays what it does.

The law has not decided whether this is a crime

For English-speaking readers the regulatory picture is mostly a blank. The SEC’s authority reaches intermediaries (exchanges and brokers) and the issuers of securities, not the anonymous ordering of transactions by bots on permissionless networks. There is no US rule that says front-running a stranger’s swap on a decentralized exchange is market manipulation, because the securities laws were written for venues with identifiable operators, and a sandwich bot has none.

The one place the question is being forced is a criminal courtroom. Anton and James Peraire-Bueno, two MIT-educated brothers, were charged with wire fraud, money laundering, and conspiracy over a 2023 scheme that used the MEV supply chain itself to extract about $25 million from searchers in a single block. Their first trial ended in a hung jury and a mistrial in late 2025, and prosecutors moved to retry them in early 2026 (Cointelegraph). The case matters well beyond the defendants: a conviction would set a precedent for when exploiting the ordering process crosses from aggressive trading into federal fraud, and the defense has argued the brothers simply beat other bots at their own game. Until it is resolved, the honest answer to whether MEV is legal is mostly untested.

The roadmap: ePBS, FOCIL, and encrypted mempools

The protocol-level answer to builder and relay concentration is finally shipping. Glamsterdam, Ethereum’s next upgrade, forked on the Sepolia testnet on 6 October 2026 and is expected on mainnet later in the year (ethereum.org). Its headline change, EIP-7732, enshrines proposer-builder separation into the protocol itself, removing the trusted relay that today sits between builders and proposers and closing the gap that enabled a roughly $20 million relay exploit in 2023 (Flashbots post-mortem).

But ePBS alone will not decentralize block building. Vitalik Buterin laid out the rest of the plan in March 2026: pair ePBS with FOCIL (a random committee that can force transactions into blocks, limiting censorship), encrypted mempools, and anonymized routing (CoinDesk). Encrypted mempools bear most directly on the strategies in this guide. If builders cannot read a transaction’s contents until after they commit to an order, they cannot front-run or sandwich it. The dark forest does not get smaller; the lights get switched off for the predators instead of the prey.

Ethereum trades around $2,543 as of this writing, with a market capitalization near $311 billion, down about 6.5% on the week (CoinGecko). None of the roadmap changes the core fact this guide has traced. Value flows to whoever controls ordering, and the fight of the next few years is over how many hands that power sits in.

Frequently Asked Questions

What is the difference between MEV and a sandwich attack?

MEV is the broad category: any profit from choosing which transactions go into a block and in what order. A sandwich attack is one specific, toxic MEV strategy, where a bot buys just before a victim’s large swap to push the price up, lets the victim fill at the worse price, then sells immediately after. Most MEV, such as arbitrage and liquidations, is not a sandwich and is arguably necessary; sandwiches are the predatory subset that hurts ordinary traders directly.

Did MEV go away in 2026?

No. Cumulative extraction keeps rising; what changed is location. The public Ethereum mempool emptied out as transactions moved to private order flow, Layer 2 sequencers, and solver networks, so MEV is harder to observe than it used to be, not smaller. Much of it is now non-atomic arbitrage between decentralized and centralized exchanges, the least visible category of all.

How do I protect my trades from MEV?

Three practical steps: route transactions through a private RPC such as Flashbots Protect or MEV Blocker so they skip the public mempool; trade through an intent-based venue like CoW Swap, UniswapX, or 1inch Fusion, where a solver absorbs the ordering risk; and use a wallet that defaults to protected routing. For small swaps through deep pools the risk is minor; for large swaps through thin liquidity it is not.

Is MEV illegal in the United States?

There is no specific US law against it. The SEC regulates intermediaries and securities issuers, not the anonymous ordering of transactions on permissionless networks, so most MEV sits outside its perimeter. The main legal test is the criminal case against the Peraire-Bueno brothers over a 2023 exploit; its outcome will shape when exploiting the ordering process counts as fraud.

What is Arbitrum Timeboost and did it reduce MEV?

Timeboost is an auction Arbitrum launched in 2025 that sells a short priority head start, an express lane, each minute, meant to internalize MEV and stop latency races. It raised real revenue, but an empirical study found three entities won 99.74% of the auctions and the lane filled mostly with arbitrage between centralized and decentralized exchanges, so it concentrated extraction rather than democratizing it.

By Yuki Tanaka, senior DeFi correspondent at HOGE Wire, covering market structure and the plumbing beneath it.