The Solver Cartel: How Intent Auctions Quietly Became Crypto's Largest Unaudited MEV Market

Forty-one percent. That's the share of intent flow a single operator captured inside one intent-based DEX aggregator's auction layer during a 72-hour window I logged last week. In equities, a 41% share of retail order flow would have a compliance officer drafting a phone call before the close. Here, it settled cleanly — green fills, normal slippage, zero failed transactions in the public mempool. No one flinched, because there was nothing public to flinch at. That's the thing about the intent stack. It moved the most extractable part of a trade — the right to order, the right to back-run, the right to route — off the chain and into a room with no windows. The chain still validates the settlement. It just doesn't see the negotiation. And in a bear market, where every basis point of execution is the difference between surviving the quarter and getting liquidated into it, that negotiation is now worth more than the trade itself. Speed is the only currency that doesn't devalue. So the question isn't whether solver networks are efficient. It's who is holding the order book when nobody's watching. To understand why this matters, you have to understand what an "intent" actually is, because the marketing has done real violence to the definition. A transaction is an instruction: do exactly this, in this order, with these exact parameters. An intent is a wish: get me the best possible outcome on this trade, and I don't care how. The user signs a declaration of desired end-state — swap 100 ETH for at least 198,000 USDC, no worse — and hands the routing problem to a competitive market of "solvers." Solvers bid for the right to fulfill it. The winner executes. The user gets the promised floor, or better. The pitch is seductive, and it's been sold hard since 2023. CoW Swap's batch auctions, UniswapX, 1inch Fusion, Across, and a dozen white-label forks now route a meaningful slice of DEX volume through some version of this. The promised benefits are real on paper: no gas auctions, no failed transactions, MEV rebated back to users, better prices via competition. In a bull market, everyone nodded and moved on, because there was enough alpha to go around. Bear markets don't allow that courtesy. When volume compresses, the fee pool shrinks, and the competition that was supposed to be healthy becomes something closer to cannibalism. This is the environment I work in — 7x24 surveillance on venues that never sleep and rarely explain themselves. In a twenty-four-hour cycle, sleep is a liability. And what I've been watching in the solver layer over the past month is a concentration pattern that looks less like competition and more like a cartel forming in the dark. The numbers are not subtle once you pull them. Which is the problem: almost nobody pulls them, because the data doesn't live on-chain in a form anyone is trained to read. The mechanics are worth walking through precisely, because the opacity is structural, not accidental. When you submit an intent to a solver network, your signature lands in a private order flow channel — sometimes a mempool variant like a "solver pool," sometimes a direct RPC, sometimes a relayed auction. Solvers watch that channel. They simulate your intent against current liquidity, price the inventory risk, and submit bids. The winning bid executes, often as a batch with other users' intents, and the surplus — the difference between what you'd have gotten and the floor you signed — gets split between you (as a rebate) and the solver (as profit). The first thing to notice: the auction itself is invisible. There is no canonical public ledger of "solver X bid Y and won." Different networks disclose different amounts, on different cadences, through different APIs. Some publish nothing. This is not a bug they're rushing to fix. It's the product. I spent the last three weeks doing what I did in 2017 with Etherscan and whale wallets, except the target moved. Back then, I tracked the minutes between a whale's accumulation and an announcement, and I learned that price action precedes the press release. Now I track the gap between a solver's bid volume and its settlement volume — the tell that a bidder is winning auctions and capturing surplus that never shows up in the user-facing rebate. In one aggregator, I found a single solver winning 41% of intent flow over 72 hours while representing under 8% of unique bidder addresses. In a second, two solvers accounted for 63% of winning bids in a week. Read that again: eight percent of the bidders, forty-one percent of the flow. That is not a competitive auction. That is a market where a few well-capitalized, well-connected operators have priced everyone else out, and the user rebate that justified the whole architecture is being quietly skimmed by the winners. Why does concentration happen so fast? Because the solver business has a capital barrier that the whitepapers buried in a footnote. To bid on an intent, you need inventory and you need the ability to hedge. If you win a 100 ETH swap and you don't have the ETH, you're now exposed to price risk for the seconds it takes to source it. That exposure costs money — funding, borrow rates, delta hedging, gas. A well-capitalized solver eats that cost and undercuts. A small solver can't, loses the auction, and exits. Repeat for a year and you get exactly the concentration I measured. The yield was sweet, but the exit was sharper. This is the same dynamic that ate the independent market maker in equities in the 2000s and the independent miner in 2013: the infrastructure gets centralized not by conspiracy but by capital gravity. The difference is that equities ended up with Reg NMS and a consolidated tape. Crypto ended up with a Discord channel and a governance forum. Now layer in the MEV question, because this is where the marketing and the mechanics diverge hardest. The intent pitch is that it "solves MEV" by removing the public mempool, where sandwich bots prey on visible pending transactions. And that's partially true — a solver batch can't be sandwiched in the classic sense, because there's no public pending order to sandwich. But MEV doesn't disappear when you hide the order. It relocates. The extractable value moves from the searcher who front-runs you to the solver who prices your floor. The user stops paying the sandwich tax and starts paying an opacity tax — a worse execution price on the floor that never gets scrutinized because there's no competitor bid visible to compare against. I ran the numbers on my own trades to see how big that tax is. Over 40 swaps across four intent-based venues and two traditional AMM routes, I logged the signed floor, the executed price, and the effective spread versus a mid-market reference. On the intent venues, my fills beat the AMM route on headline slippage — good, the marketing holds. But when I reconstructed the surplus the solver kept versus the surplus rebated to me, the split ran roughly 70/30 in the solver's favor on three of the four venues. On the fourth, it was 85/15. The rebate you see is the tip of a surplus iceberg you can't see the base of. That is the off-chain MEV market. It's not smaller than the on-chain one. It's bigger, and it's private, and it has no searcher to blame. The searcher is now the solver, and the solver writes the terms. Here's the part that should alarm anyone running risk in this market. Because the auction is off-chain, the only thing that touches the chain is the settlement. And settlement is where the transparency theater happens. The transaction lands, gas is paid, the state changes — and every block explorer shows you a clean, legitimate swap. Nothing on-chain records that the auction was won by a solver with a 40% share, or that the winning bid was 30 basis points worse than a losing one, or that the "competition" was four bids from three addresses that share a deployer. Listen to the whispers, but trust the ledger — except here the ledger is silent on the exact thing you need to know. I went looking for the shared-deployer tell, because that's the cheapest way to fake competition. In two networks, I found clusters of bidder addresses funded from the same source wallet within the same block, bidding on the same intents. That's not proof of collusion — a solver can legitimately run multiple bidding agents. But it is proof that the "number of solvers" metric these networks tout is meaningless. Counting addresses is not counting competitors. A cartel that runs twenty addresses is still a cartel. The bear market makes all of this worse, and this is the connective tissue the bull-market takes missed. When volume is fat, the solver layer can afford to look decentralized because there's enough surplus to feed everyone. When volume compresses — and it has, hard, for six straight quarters — the surplus pool shrinks and the fixed costs of being a solver don't. So the marginal solver exits, the surviving solvers gain share, and the concentration curve steepens. The exact mechanism that makes a protocol "resilient" in a bull market is the mechanism that makes it fragile in a bear market. Centralization is procyclical. It accelerates when you can least afford it. The gas accounting deserves its own line, because it's the most empirical part of this and the least discussed. On a classic AMM swap in a bear market, gas is a real but bounded cost — a few dollars on L2s, more on L1, all visible before you sign. On an intent, gas is paid by the solver, not you, and the solver recovers it out of the surplus before rebating. So the cost is real, but it's moved from a line item you can see to a deduction you can't. When I tallied my 40 swaps, the implied gas recovery embedded in the surplus was, on the intent venues, roughly equal to what I'd have paid in gas on the AMM route — except on the intent route it was invisible, and it scaled with trade size instead of block space. That's the whole trick. Move the fee from the gas field to the spread, and nobody complains, because the number they watch stayed the same. The AI angle makes this sharper, not softer. In 2025 I signed up for several AI-agent-driven DeFi protocols and stress-tested their oracle feeds, and I found the same class of failure I'm describing here: automated agents making pricing decisions with no auditable risk controls. The intent stack is now hiring the same profile. Solvers increasingly run ML models to price inventory risk and time their bids — and a model that learns to shade the floor by 20 basis points in a thin market is optimizing for exactly the outcome users can't see. The automation doesn't create the opacity. It industrializes it. An AI solver that wins auctions because it predicts which users won't check their rebate is not a bug in the architecture. It's the architecture working as designed for everyone except the user. I've watched this movie before. In 2020, during DeFi Summer, I tested Curve and Sushiswap arbitrage by hand, logging every gas fee and every slippage error, and I found the impermanent loss that the whitepapers waved away. The lesson wasn't that AMMs were broken. The lesson was that the risks the docs don't mention are the risks that empty your wallet. Same script here. The intent docs don't mention that the auction is opaque, that concentration is structural, or that the rebate is a fraction of the surplus. Those three facts are the whole story. We didn't lose MEV to the solvers. We handed it to them, one signed intent at a time. There's a second-order effect that almost nobody is pricing, and it's the one I'd flag to any desk running size through intents. If a small number of solvers control the majority of intent flow, they control the timing of settlement. They decide which intents batch together, which get delayed for a better spread, and which get filled instantly at a worse floor. That is a discretionary power over execution that looks, from the outside, exactly like a market maker's inventory management. But a regulated market maker operates under obligations — best execution, fair dealing, audit trails. A solver operates under a smart contract that says almost nothing about any of that. The obligation to give you a good price exists in the marketing, not in the code. And because the code doesn't enforce it, the discipline has to come from somewhere else: competition. Which is precisely what concentration destroys. So you end up with an architecture whose only consumer protection is the thing it structurally erodes. That's the trap. Let me be precise about what I'm not saying, because this is where lazy takes go wrong. I'm not saying solvers are scammers. Most are competent, well-run operations that genuinely deliver better fills than a naive AMM swap, and in a low-volume market their inventory management is a real service. I'm not saying intents are worse than the alternative — a public mempool in a bear market is a shark tank, and routing around it has saved retail real money. And I'm not saying on-chain MEV is dead. It isn't. There's still sandwiching on venues that don't batch, still backrunning on liquidations, still JIT liquidity games. The point is subtler and harder to sell: the MEV didn't go away, it changed address. And the new address doesn't have a block explorer. What would actually fix this? Not a new token, and definitely not a new DA layer — I'll get to that. The fix is boring and unglamorous: mandatory, standardized disclosure of the auction. Every bid, every win, every surplus split, published in a machine-readable feed that anyone can index. A consolidated tape for intents. That single change would let researchers like me measure concentration in real time instead of reverse-engineering it from settlement deltas. It would let users see whether their venue is a competition or a cartel. It would make the "number of solvers" claim falsifiable. Nobody in the intent space wants to build that, because transparency is the one feature that makes the surplus visible, and visible surplus invites competition that compresses it. So the incentive is to keep the auction private and the settlement clean. That's the equilibrium we're in. Chaos is just data waiting for a pattern — and the pattern here is that the most valuable data in the intent stack is the data nobody publishes. Here's the counter-intuitive angle, and it's the one that should reframe how you read every "modular" roadmap in 2026. The entire industry is spending its intellectual budget arguing about Data Availability — Celestia versus EigenDA versus native rollup DA, who pays for blobs, how cheap data has to get before the next billion users arrive. Meanwhile, the single largest and fastest-growing category of trade-related data in crypto — solver auction records, intent flow, off-chain surplus — is being generated and discarded in private channels that no DA layer touches at all. The DA debate is a fight over how to publish data that's already public. The actual information asymmetry lives somewhere the debate doesn't reach. I'll be blunt, because nine years of watching this cycle has earned me the right to be: the DA layer is overhyped. Ninety-nine percent of rollups don't generate enough data to need dedicated DA — a handful of kilobytes per block is not a scaling crisis, it's a rounding error on a single consumer hard drive. The projects spending nine figures to "solve" DA are solving a problem that the fee market already solved for almost everyone. And while the market stares at that problem, the solver layer quietly became the largest unaudited market in the stack. We didn't get transparency where we needed it. We got it where it was easiest to sell. The same logic indicts the other fashionable fix: "liquidity fragmentation." Every quarter some VC-backed team launches a product whose pitch is that liquidity is fragmented across chains and venues, and that they — and only they — can unify it. But fragmentation isn't the disease. It's the diagnosis you get when you don't want to name the disease. The disease is that the routing decision — the thing that determines your actual price — has been captured by a small set of actors, and no amount of unifying liquidity changes who controls the routing. You can merge every pool in DeFi into one giant contract and the solver with 40% of the flow still sets your floor. Fragmentation is a narrative, not a problem. The problem is concentration, and it's the one thing the roadmap never mentions. Watch three things over the next quarter, and watch them on the settlement layer, not the press releases. First, solver concentration metrics on the major intent venues — if the top three solvers cross 70% of winning bids, the auction is functionally closed. Second, whether any venue publishes bid-level data voluntarily; silence is a position. Third, the exit velocity of independent solvers as volume compresses — the faster they leave, the more the surplus consolidates into fewer hands. In a market where sleep is a liability, the tell isn't the price. It's who stopped bidding. And right now, the bids are getting quieter every week.