The Solver Field in Intent-Based DEXes

TL;DR

Intent-based DEXes (CoW Swap, 1inch Fusion, UniswapX) do not execute trades directly against AMM pools. They route user orders to a competitive set of solvers, who submit bids to fill those orders; the best bid wins and settles. On Ethereum-family venues, solvers come in two flavors with very different economics: market-maker solvers filling from their own inventory using off-chain pricing (Wintermute, and others), and algorithmic solvers simulating AMM math offchain and routing through public liquidity (Barter, formerly Copium Capital, and a shrinking list of others). On Solana, a distinct third architecture has emerged: PropAMMs, on-chain inventory + execution with off-chain pricing pushed every ~100ms (BisonFi, HumidiFi, SolFi, Obric v2, ZeroFi).

On CoW Swap, the market-maker solvers are winning. That is structurally interesting beyond the solver market itself, because every order a market-maker solver fills from inventory is an order that never touches an AMM pool. If that pattern generalizes, the AMMs are left with what's left, more adversely-selected flow, lower-quality fee revenue, and worse LP economics. For tokens whose value-capture depends on AMM fee revenue (UNI's fee switch being the prime case), that's a slow-moving structural headwind that doesn't show up in a volume chart.

On Solana, PropAMMs are winning against a different baseline: per Jump Crypto's March 2026 analysis, aggregate PropAMM volume on SOL dollar pairs (~$19.87B) matched the combined volume of Binance + Coinbase + OKX + Bybit (~$19.22B) on the same pairs, and 91.9% of fills beat the best-tier CEX all-in cost of 2.57 bps. The historical "DeFi is worse execution" premise is no longer true on established pairs where a PropAMM competes.

What is a solver, exactly?

A solver is a hybrid market participant with three jobs layered on each other:

• Aggregator: watch all the public liquidity that exists (AMMs across chains, RFQ sources, wrappers) and find the optimal path for any given user order.
• Market maker: sometimes, instead of routing through public liquidity, fill from private inventory at a price the solver sets, earning the spread.
• MEV bot: be aware of block-level conditions (pending transactions, gas, inclusion economics) and optimize for winning the bid by being fastest, cheapest, and most likely to land.

Where a retail-facing aggregator like 0x is a UI, a solver is a competitor. On CoW Swap specifically, solvers bid in batch auctions: for each batch of user orders, whichever solver submits the settlement plan with the best execution quality wins the right to settle it. The solver earns a small fee; the user gets the best execution across everyone who bid.

That competitive structure is the reason "who solves" matters. If the set of competing solvers becomes homogeneous, all the same type, with the same advantages, the competition collapses and the market-clearing behavior changes.

The two types of solvers

The bifurcation is cleanest on CoW Swap, where both types compete directly. Both are trying to win the same batch auction; the how is radically different.

The key asymmetry is information. A market-maker solver knows what CEX order books look like right now. An algorithmic solver knows what on-chain AMM state looked like at the end of the previous block, with no visibility into what other transactions in the same block will do to those pools. On any order where off-chain price signal is better than on-chain (which is most orders, most of the time), the market maker wins.

A third category emerging on Solana: PropAMMs

The two-category model above is clean on Ethereum-family intent systems (CoW Swap, 1inch Fusion, UniswapX). On Solana, a third architecture has become dominant that fits neither bucket cleanly: the proprietary AMM, or PropAMM. BisonFi, HumidiFi, SolFi, Obric v2, and ZeroFi all operate in this category.

The PropAMM trick is to collapse the "off-chain pricing advantage" of a market-maker solver with the "on-chain inventory and execution" of an AMM. It keeps the inventory and execution logic on-chain (so every fill is independently verifiable and settles in-protocol) while pricing that inventory from an off-chain pricing engine that continuously pushes signed oracle updates into the onchain program. Jump Crypto, the operator of BisonFi, calls this "colocation, but the market-making logic is inside the exchange execution environment itself."

Why this matters. PropAMMs break the "either off-chain advantage or on-chain transparency, pick one" framing that the first two categories forced. Per Jump's March 2026 analysis of 20 million fills: the median SOL/USDC fill on a PropAMM was 0.72 bps from the best midpoint across Binance, Coinbase, OKX, and Bybit. 91.9% of fills beat the CEXes' best-tier all-in cost of 2.57 bps (the tier requiring billions per month in volume to qualify). 99.3% beat retail-tier CEX pricing (~10 bps). And the aggregate PropAMM volume on SOL dollar pairs in March 2026 (~$19.87B) matched the combined volume of the top four CEXes (~$19.22B) on the same pairs.

How PropAMMs relate to toxic flow (see next section). PropAMMs explicitly price against toxic flow inside the program: the onchain logic tracks historical quality by routing path and widens spreads or withholds depth when an incoming swap arrives through a pathway that has previously delivered adverse selection. This is the same dynamic the next section describes on the Ethereum side, but on Solana the defense is built into the venue, not into a separate solver. The "who is winning" story that follows is the CoW Swap story (market-maker solvers winning at CoW). On Solana, PropAMMs are doing their own winning against a different baseline (centralized exchanges), which is what the execution-quality data above measures.

Is it Solana-specific? In principle no, a PropAMM could run on any chain with fast enough block times to make sub-second oracle repricing economically viable. In practice yes for now. Ethereum's 12-second block window is structurally too slow for the 100ms oracle cadence the model depends on. L2s with <1s block times are the natural next venues if the model exports.

Who is winning on Ethereum-family intents: the CoW Swap ranking trend

CoW Swap publishes solver rankings, which makes it one of the cleanest venues to watch this play out. The trend is unambiguous.

Source: public CoW Swap solver leaderboard; commentary per Barter leadership interview, April 2026. Rankings reflect volume / settled-transaction share on CoW Swap specifically, not total solver-market share across all intent systems.

The trend is not just about Wintermute out-competing. It's about algorithmic solvers exiting. Copium Capital stopped solving; Barter acquired its codebase in September 2024. New intent systems launching in 2025 and 2026 are explicitly onboarding market makers rather than algorithmic solvers. The supply of algorithmic solver competition is shrinking independently of market-maker performance.

Why this matters beyond the solver market: toxic flow

Here is where the solver story becomes a token-economics story.

When a market-maker solver fills a user order from inventory, the AMM pool that would otherwise have facilitated that trade never sees the trade. That is a clean loss of fee revenue to the AMM. But the deeper issue is which orders flow where:

A market maker with good information rejects orders it expects to lose money on. Those rejected orders are precisely the ones an informed trader is submitting, orders where the trader's view of price is more right than the market maker's. Those orders still need to execute somewhere, and if an algorithmic solver picks them up and routes them through AMMs, the AMM LP is the one taking the losing side of the trade.

In aggregate: the AMM pool becomes a disposal layer for the orders that sophisticated off-chain pricing identified as adverse. Volume can look fine; fee-revenue quality quietly collapses. LPs in the pool see unexplained negative returns even during "healthy" volume periods. This is the mechanism academic market-microstructure literature refers to as adverse selection, now playing out in DEX order flow.

Counterforces worth watching

This is not a certain outcome. Three paths could neutralize or reverse the pattern.

1. AMMs fight back via custom-pool logic

Uniswap v4 hooks allow per-pool customization that can rebate LPs for toxic flow, integrate MEV-protection primitives, or differentiate informed vs. uninformed order handling. If the hook ecosystem matures fast, AMM pools can structurally approximate some of the inventory-management advantages market makers have, without needing off-chain pricing.

2. Multi-primitive liquidity convergence

Newer venues (Fluid, 1inch Aqua once it launches) use a single shared pool of deposits to power both AMM-style swap liquidity and lending simultaneously. Capital throughput is structurally higher than a pure AMM, because the same dollar earns fees from swaps and interest from borrowers. That economic gap is the kind of thing that could pull LP capital back from pure-AMM positions toward multi-primitive venues, at which point Uniswap either matches via v4 hooks or loses the LP base permanently.

3. The monopoly-reversal scenario

If algorithmic solvers get squeezed out entirely and market-maker solvers become effectively monopolistic, competitive pressure to keep spreads tight disappears. Market-maker spreads widen. AMMs become competitive again on rate for common flow, and the dynamic unwinds. This is the argument algorithmic solvers make about why their survival matters for the broader DeFi ecosystem, and it's structurally plausible, though a monopoly-equilibrium takes years to form and years to reverse.

An aside: solving is logistics

The mathematical problem an algorithmic solver is trying to solve, optimally split a quantity of an input asset across many possible routes to raise output, subject to constraints on gas, slippage, and settlement certainty, is formally the same problem as optimizing a shipment across a supply-chain network. Barter's algorithmic lead reportedly spent nine years building shipment-tuning software used by IKEA, Philip Morris, and Pepsi before building the DEX solver. Engineers at Fidelity independently arrived at the same parallel when they saw it described. The cross-domain applicability is part of what makes the problem tractable to more teams than crypto-natives, and part of why convex-tuning approaches from operations research have become a mainstream solver technique.

The practical implication: the barrier to entry in algorithmic solving is not crypto-native knowledge. It's access to high-quality tuning infrastructure and the willingness to compete at thin margins. That's a worse competitive moat than market-maker solvers have (which requires inventory + institutional-grade pricing relationships), which is a second reason algorithmic solvers are losing.

How TokenIntel uses this framework

This framework shows up in a specific place in our coverage:

• Uniswap (UNI) has a dedicated "Toxic Flow Risk from Intent-Based Routing" subsection in its research page, scored as medium risk, structural, uncertain magnitude. The UNI thesis depends on fee-switch revenue quality, and fee-switch revenue quality depends on what intent-based solvers are pulling out of the AMM order flow.
• Aave (AAVE) shows the same pattern at the application layer, but in reverse, and with a structural response. In December 2025 a dispute erupted because CoWSwap deployment fees on Aave.com were flowing to an external recipient instead of the DAO treasury. The Aave Will Win (AWW) proposal, passed April 2026, explicitly addresses this by redirecting 100% of revenue from all Aave-branded products back to the DAO and requiring service providers to build exclusively for Aave. That is precisely the "don't let solvers or aggregators leak your flow's value" answer at a governance level, protocol-side language against the same toxic-flow dynamic this page describes.
• Aerodrome (AERO) faces a weaker version of the same risk. AERO's ve(3,3) model pays veAERO lockers from fees, so the same fee-quality concern applies, but AERO volume is more retail and less routed through CoW/1inch-style intent systems today. Worth monitoring as intent systems expand.
• Raydium (RAY) and Jupiter (JUP) are Solana-native and face the PropAMM dynamic described above: BisonFi, HumidiFi, SolFi, Obric v2, and ZeroFi (PropAMMs, not passive AMMs) capture the clean routed flow on SOL/USDC, leaving Raydium and Orca as the residual public AMM layer. The Solana research page's Private DEX Dynamics section has the Jump Crypto data (~$19.87B PropAMM volume in March 2026, matching the top-4-CEX aggregate on SOL dollar pairs) and the fill-quality numbers in full.

The underlying question across all of them is the same: when the flow that feeds your fee revenue gets re-routed around you, does your token's value-capture story survive? That's the lens this page is for.