What is MEV, and who is actually paying for it
MEV is not a tax on blockchains; it is a transfer from one set of users to another. Since the Merge, roughly $760m has moved through MEV-Boost. Here is who paid it and who collected.
Between 15 September 2022 — the day the Merge introduced proposer-builder separation in its current form — and the end of Q1 2026, validators on Ethereum have collected roughly $760m in priority fees and MEV through the MEV-Boost relay network. The figure comes from mevboost.pics‘s rolling dashboard, cross-checked against the Flashbots transparency reports. That is the easy number. The harder one — the one this article exists to walk through — is who actually paid it. Because MEV is not a tax on the network. It is a transfer, mostly from one set of users to another, with three or four intermediaries taking a cut on the way through.
Once you see MEV as a transfer rather than as friction, a lot of the policy debate around it makes more sense. Sandwich attacks are not free money out of the air; they are money taken from the swap user, routed through the searcher, the builder, the relay, and the proposer. The right question is not “is MEV good or bad” but “what fraction of that pipeline is rent, and what fraction is paying for a service the user actually wants?” The answer in 2026 turns out to be uncomfortable.
The MEV stack, defined precisely
“MEV” originally stood for miner extractable value, a term coined in the Flash Boys 2.0 paper in April 2019. After the Merge it was rebranded to maximal extractable value, which is more honest because the extractor is no longer always the block producer. The current pipeline has four roles, and the boundary between them matters enormously for who profits.
| Role | What they do | Where their revenue comes from | Typical share of an MEV block |
|---|---|---|---|
| Searcher | Spots an opportunity (arb, liquidation, sandwich), builds a bundle | Difference between opportunity value and bundle bid | 10-40% |
| Builder | Assembles a full block from searcher bundles + public mempool | Margin between block value and proposer bid | 0-5% (often razor-thin) |
| Relay | Forwards builder blocks to proposers, enforces data availability | Mostly free; some run by Flashbots, bloXroute, Eden, Ultra Sound, Aestus | 0% |
| Proposer (validator) | Signs the highest-bid header from the relay | Proposer bid (priority fees + MEV tip) | 55-90% |
Two things stand out in that table. The first is that builders, despite being where most of the technical sophistication lives, capture almost none of the value. The market has compressed builder margins to single-digit basis points in a brutal race-to-zero that is well-documented in Flashbots’ own research blog. The second is that the proposer — the validator who just holds the key — gets the lion’s share without doing any of the searching. This is the proposer-builder separation working as designed: validators stay decentralised by outsourcing the hardest part of block construction to a competitive market of specialists.
The four kinds of MEV, with examples
Not all MEV is the same, and conflating it is the most common mistake in the debate. There are roughly four categories, and they have very different ethical weight.
- Atomic arbitrage — A searcher buys ETH on Uniswap v3 at 3,200 USDC and sells it on Curve at 3,205 USDC in the same transaction. The price difference existed before the searcher arrived; the arb just closes it. This is unambiguously useful: it keeps prices consistent across venues. Roughly 40% of MEV revenue.
- Liquidations — When an Aave or Compound position falls below its health factor, anyone can call
liquidationCalland earn the liquidation bonus. Liquidators are doing the protocol a favour; without them, bad debt accumulates. Roughly 10-15%. - Sandwich attacks — A searcher sees a large swap in the mempool, front-runs it with a buy that pushes the price up, then back-runs with a sell. The victim pays the worse price; the searcher pockets the difference. Roughly 20-30% of MEV revenue, and the controversial part of the chart.
- Long-tail / time-bandit / oracle — JIT (just-in-time) liquidity on Uniswap v3, oracle frontrunning during NFT mints, NFT bidding wars. Maybe 15-20% combined.
The honest scoreboard is: arbitrage and liquidations make markets work. Sandwiches and JIT liquidity are user-funded. EigenPhi’s running attribution consistently shows that sandwich losses to ordinary swappers run between $1m and $3m per week on Ethereum mainnet, plus a similar amount on the major L2s combined. That is the line item that should bother policymakers — not the abstract noun “MEV.”
The MEV-Boost pipeline, slot by slot
It helps to walk through what actually happens between t-12 seconds and t-0 in a slot where the proposer is using MEV-Boost. At t-12, the previous block’s transactions are being mined into the public mempool; searchers are watching, simulating, and building bundles. By t-2 or t-3, searchers have submitted bundles to one or more builders, each bidding for inclusion. By t-1, builders have submitted full blocks plus header bids to the relays. At t-0, the proposer pulls the highest-bid header from each relay it trusts, signs it, and gets the body in exchange. The relay’s job is to ensure the proposer cannot see the full block contents before signing — otherwise the proposer could steal the searcher’s strategy.
This is why relays are the trust-minimised bit of the pipeline that nobody pays for. They exist to make MEV-Boost work at all, and they have repeatedly stepped in when something goes wrong — most famously during the April 2023 sandwich-stealing incident, when a malicious proposer exploited a missing signature check to steal $20m from a sandwich bundle. The patch was shipped within hours, the protocol added explicit equivocation slashing in subsequent upgrades, and the lesson — that relays need cryptoeconomic enforcement, not just code — drove a lot of the design of MEV-Boost v2 and later proposer commitments.
Who pays: a worked example
Consider a real-shaped scenario. A trader swaps 500,000 USDC for ETH on Uniswap v3 with a slippage tolerance of 0.5%. The pool price is 3,200 USDC/ETH. A searcher detects the pending transaction in the public mempool, front-runs with a 2,000 USDC buy that pushes the executed price to 3,210, lets the victim’s 500,000 USDC trade fill at the worse price, then back-runs with a sell that captures roughly 1,500 USDC of profit. The searcher offers 1,200 USDC of that as a tip to the builder; the builder keeps 50 USDC margin and bids 1,150 USDC into the proposer header. The proposer signs, collects 1,150 USDC worth of ETH on top of base reward and priority fee.
The accounting: the swapper lost 1,500 USDC. The searcher netted 300 USDC. The builder netted 50 USDC. The proposer netted 1,150 USDC. Nobody created any value; 1,500 USDC moved from the swapper to a chain of intermediaries. Repeat this a few thousand times per day across mainnet and the major L2s and you get the EigenPhi numbers. The mitigations are not theoretical: routing through Flashbots Protect or CowSwap’s batch auctions removes the public-mempool exposure that makes sandwiching possible. Our tools page tracks which DEX aggregators currently route through private orderflow by default.
L2s changed the geography but not the physics
Optimism, Arbitrum, Base and zkSync have all moved enormous volume off mainnet — combined L2 TVL passed $50bn in 2025 — and each runs a single sequencer. That sequencer is, today, the only entity that can extract MEV on that L2. Arbitrum and Optimism both publish sequencer revenue; Base, run by Coinbase, has been less transparent. L2Beat tracks the decentralisation roadmap for each, and the “shared sequencer” projects — Espresso, Astria, Radius — are explicitly trying to import the MEV-Boost separation into the L2 world before a single sequencer becomes a permanent rent-collector.
The interesting wrinkle on L2s is that the sequencer can offer guarantees that mainnet cannot: pre-confirmations, encrypted mempools, and FCFS (first-come-first-served) ordering. Each of these eliminates a category of MEV. Encrypted mempools — see Shutter and SUAVE — make it impossible for searchers to see the transaction until after ordering is committed. FCFS rules, as on Arbitrum’s classic sequencer, make sandwiches mechanically impossible because the searcher cannot insert a transaction after the victim’s. These are not minor tweaks; they reshape which categories of MEV are even possible.
What’s coming: encrypted mempools, MEV burn, SUAVE
Three structural changes are working through the protocol research community. The first is encrypted mempool proposals — Shutter Network and the various threshold-encryption schemes — which would make sandwiching cryptographically impossible without sacrificing the public ordering guarantee. The second is MEV burn, sometimes called execution-tickets or attester-proposer separation, which would route some or all of the proposer’s MEV revenue into the EIP-1559 burn instead of into validator pockets. EthResearch has running threads on both. The third is SUAVE, Flashbots’ attempt to build a chain-agnostic block-building network, which would let users express preferences (e.g. “execute only if not sandwiched”) and have searchers compete to satisfy them.
None of these are deployed in their final form yet, but together they suggest the MEV problem is being attacked at the right layer: the ordering protocol, not the user. The dashboards to keep an eye on are mevboost.pics for relay-level flow and EigenPhi for the user-loss attribution, both of which feed our market dashboard. The upcoming Ethereum upgrade calendar — Pectra completed, Fusaka next — is tracked on the events page; each upgrade has the potential to move the MEV pipeline materially.
The takeaway
MEV is not free money from the chain. It is a transfer from some users — mostly swappers, mostly via sandwich attacks and JIT liquidity — to a four-stage pipeline that funnels most of the value back to validators. About half of MEV is the kind of arbitrage and liquidation work that markets need; the other half is a tax on retail traders who route through the public mempool. The single best thing an ordinary user can do is use a private-orderflow RPC or a batch-auction DEX. The single most interesting protocol-level question is whether MEV burn will eventually route this revenue away from validators and into ETH-holders generally, which would change the calculus of staking and of every L2’s MEV strategy.
Either way, the next time you see a headline number for “MEV extracted this month,” ask the same question that applies to any tax: who paid it, and to whom did it go.