Isometric illustration of separate platforms each holding stacked crates, linked by cables, with one small vessel moving between them.

After Bridges: Who Fronts the Capital for Intents

Intents won cross-chain UX. ERC-7683 standardizes the order format, not the balance sheet: every 1.2-second fill is a solver lending its own inventory.

17 min read
Share X LinkedIn

cover

Introduction

Across states on its homepage that it has processed more than $40 billion in cumulative volume for 5 million users at an average fill speed of 1.2 seconds. Those are vendor-reported figures we have not seen reproduced by an independent analytics source, but the mechanism they describe is not in dispute: every one of those fills was a solver sending its own tokens to a user on the destination chain before being paid, then waiting for a settlement proof to release escrowed funds on the origin chain. The cross-chain experience that displaced lock-and-mint bridging runs on short-duration inventory lending by whoever already holds assets where the user is going.

A cross-chain intent is a signed statement of the outcome a user wants, with constraints on it, rather than the sequence of transactions that produces that outcome. The user specifies what they are giving up, what they want to receive, where, by when, and the worst price they will accept. Someone else figures out the execution and gets paid for it. ERC-7683 standardizes how that statement is expressed and how settlement contracts are called, while leaving the settlement process itself to each implementation. The standard’s own site claims 70 protocols support ERC-7683 and cross-chain intents, though the page carries a “last updated” notice roughly a year old and does not define whether “support” means a live implementation or an endorsement.

The marketing case for this architecture, stated plainly on the ERC-7683 site, is that standardization “reduces failure rates and speeds up transaction times by fostering competition among fillers.” That claim has two parts. Faster fills are observable and plausibly true. More competition among fillers is an assertion about market structure, and no public source we could find measures it. There is no third-party dataset on how fill volume is distributed across solvers on Across or Relay, how many distinct fillers are active, how often new ones enter, or what fraction of solver revenue comes from quoted spread as opposed to incentive rebates or destination-chain execution advantages.

That gap matters because the binding constraint on becoming a filler was never the order format. It is holding capital on the destination chain, carrying it through the reimbursement window, and getting the order in the first place. Standardizing the message does not touch any of the three.

The fill is a loan with a settlement proof attached

img1

The intent lifecycle separates the moment a user is made whole from the moment anyone verifies that it happened, and the gap between those two moments is where solver economics live.

The sequence has four steps. The user signs an order and deposits the input asset into an escrow contract on the origin chain, with the order specifying the output asset, amount, destination chain, recipient, deadline, and the compensation the filler will receive. A solver observes the open order, prices it, and decides whether to fill. If it fills, it sends the output asset from its own balance on the destination chain directly to the user. The settlement system then establishes that the fill occurred and releases the origin-chain escrow to the solver.

From the user’s side the transaction is complete at step three, typically within seconds. From the solver’s side nothing is complete at step three. It has parted with destination-chain inventory and holds a claim against an escrow contract on a different chain, contingent on a proof system behaving as specified. Steps three and four can be separated by minutes or by considerably longer, depending on the settlement design and the finality assumptions of the origin chain.

ERC-7683 does not dictate how step four works, and the designs in use differ materially in trust assumptions. Optimistic settlement posts an assertion that a fill occurred and releases escrow after a challenge window elapses without dispute, which makes reimbursement latency a deliberate parameter rather than a technical limit. Verification through a light client or validity proof removes the challenge window at the cost of implementation complexity and per-chain support. Attestation by an external messaging network substitutes the honesty of that network’s signers for either. A solver operating across many routes is underwriting a different counterparty risk on each one, and it is doing so with capital that is already out the door.

Several failure modes sit in this gap, and they are solver losses rather than user losses by construction. If two solvers fill the same order, one of them has made an unreimbursed transfer unless the protocol enforces an exclusivity window or a first-claim rule at the escrow. If the origin chain reorganizes after the fill, the deposit backing the claim may no longer exist, which is why solvers price finality waits into their quotes rather than treating all chains alike. If the settlement proof stalls, the solver’s capital is immobilized without default having occurred, which is a liquidity event rather than a credit event but constrains the same balance sheet. And if the input asset the solver eventually receives has moved in price during the window, the realized spread differs from the quoted one.

The arithmetic consequence is the part that shapes the market. Required inventory per chain is roughly throughput multiplied by reimbursement latency. As an illustration, a solver filling a steady million dollars an hour into a given chain, reimbursed on a one-hour cycle, needs about a million dollars sitting idle on that chain at all times to sustain it; cut the cycle to six minutes and the same throughput needs a tenth of that. The advertised 1.2-second fill speed describes the user’s leg. The solver’s leg is set by the slow one, and public sources do not report the distribution of settlement latencies on any major implementation.

What ERC-7683 standardizes, and what it does not

img2

ERC-7683 standardizes the demand side of this market and explicitly declines to standardize the supply side. The standard’s own documentation concedes the point: “While ERC-7683 standardizes how intents are created, it offers flexibility in how they’re settled.”

That flexibility is defensible as engineering. Settlement proof requirements differ between a rollup with a canonical bridge, an alt-L1 reachable only through third-party messaging, and a chain whose finality is probabilistic. Forcing one proof system on all of them would have narrowed adoption. But the consequence for market structure is that a solver integrating a second ERC-7683 implementation reuses the order parsing and almost none of the risk work. It must still underwrite a new settlement system, model a new reimbursement latency, and fund a new set of chain balances.

LayerStandardized by ERC-7683Where a new entrant’s cost actually sits
Order expression and encodingYes, via a shared order formatLow; integration work is one-time and now portable
Settlement contract interfaceYes, interface levelLow to moderate
Settlement proof and trust assumptionsNo, left to each implementationHigh; per-implementation risk underwriting
Reimbursement latencyNoHigh; determines idle capital per unit of throughput
Destination-chain inventoryNoHighest; scales with throughput and chain count
Access to order flowNoOpaque; governed by integrations, not by the standard

The historical analogue is message standardization in traditional markets. A common wire format for orders lowered the cost of connecting to a venue and did not make market making a commodity business, because the capital, the latency infrastructure, and the relationships with order originators remained privately held. ERC-7683 occupies the same position. It makes the request legible to everyone and leaves the ability to answer it unevenly distributed.

The counterclaim deserves a fair statement. Lower integration cost does expand the set of participants who can attempt to solve at small size, and a long tail of small fillers taking small-notional orders on cheap routes is a real outcome, not a hypothetical one. The question is whether that tail captures a meaningful share of volume or only the residual that larger balance sheets decline to quote. Without per-solver fill data, neither answer is established.

Who can afford to front the capital

img3

Filling concentrates capital because the inventory requirement scales with the number of supported chains, not with the volume on any one of them, and because flow netting rewards size in a way that linear cost models do not capture.

Consider the position a solver must hold to quote competitively across a broad route set. It needs working balances of each commonly requested output asset on each destination chain it serves, sized for peak rather than average demand, because a solver that runs dry on a chain stops winning orders there and surrenders its position to whoever is still liquid. Adding a chain adds a full set of idle balances before it adds any revenue. A solver covering ten chains carries roughly ten times the idle inventory of one covering a single corridor at comparable per-chain throughput, and earns per unit of flow, not per unit of inventory.

Scale changes this arithmetic in two ways that favor incumbents. First, flow netting: a solver seeing orders in both directions on a corridor consumes inventory on one side and replenishes it on the other, so its rebalancing cost depends on net imbalance rather than gross volume. A small solver that only sees one-directional flow pays the full round-trip cost on every unit. Second, fixed costs: node infrastructure, pricing models, finality monitoring, and settlement-system risk review are largely independent of volume, so unit costs fall with throughput.

Against those advantages sits adverse selection, which cuts the other way and is underappreciated. An open fill market is an auction in which the winner is the solver that quoted most aggressively, and the orders a solver wins disproportionately include the ones it mispriced. A filler with a worse model of finality risk, destination-chain gas spikes, or short-horizon price movement wins more of the orders that lose money. Competition among fillers therefore does not reliably compress the spread toward the true cost of capital; it can instead select for whoever is most willing to misprice, until that participant exits. Concentration among a few survivors is a plausible equilibrium of a competitive auction, not evidence that the auction was rigged.

Where solver revenue actually comes from is not publicly documented for any major implementation, and we decline to estimate it. The candidate sources are the spread between the quoted output and the solver’s all-in cost, protocol or ecosystem incentive programs that subsidize fills on specific routes, execution advantages captured on the destination chain while filling, and privately negotiated arrangements with order originators. The user pays for all of them through a single number: the output amount in the quote. An intent quote bundles the bridging fee, the inventory cost, the risk premium, and any rebate offset into one figure the user either accepts or does not. That is better UX than an itemized bridge fee plus slippage plus gas, and it is strictly less legible. Nothing in the architecture reveals how much of the spread is cost and how much is rent.

Rebalancing, the cost nobody quotes

img4

Rebalancing is the recurring cost that determines which routes are cheap, and it is a function of flow imbalance rather than of volume.

After filling, a solver has less inventory on the destination chain and a claim, later cash, on the origin chain. If user demand ran equally in both directions, the positions would offset and no capital would need to move. Demand does not run equally in both directions. Flows concentrate toward chains running incentive programs, toward wherever yields or new markets have opened, and away from chains in decline. On a structurally one-way corridor, every fill moves the solver’s inventory in the same direction, and it must pay to move it back.

Moving it back is not free and not fast. The canonical bridge of an optimistic rollup imposes an exit delay measured in days, which is an eternity against a reimbursement cycle measured in minutes and makes canonical exits useless for working capital. The alternative is to pay a third-party route, which means paying someone else’s spread, or to cycle through a centralized venue that supports deposits and withdrawals on both chains, which introduces custodial exposure and operational constraints. Each option trades cost against latency against counterparty risk, and the solver’s choice is invisible to the user whose quote absorbs it.

No public source documents how solvers on Across or Relay actually rebalance, what it costs them, or how that cost scales. This is the single largest empirical gap in the intents thesis, and it should not be filled with plausible-sounding estimates. What the mechanism does yield is a testable prediction: if rebalancing cost is a meaningful component of spread, quoted spreads should be systematically wider on routes with persistent one-way flow than on balanced corridors of similar size and finality characteristics, and should widen further when a destination chain’s inbound incentive program is running. Route-level spread dispersion is observable from quote data. Anyone with access to a quote feed can test this without needing a solver’s books.

Everclear’s clearing design attacked this cost at the protocol layer by netting offsetting flows before anyone has to move capital, which is the right place to attack it. Our earlier post-mortem covered what happened to that attempt. The cost it targeted did not disappear when intents won the interface argument; it moved onto solver balance sheets and into quoted spreads, where it is harder to see.

Order flow decides which solvers get to compete

img5

A permissionless filler network determines who is allowed to compete, not who gets the chance to. Order flow reaches solvers through front ends, and front ends are a concentrated set.

Across names Uniswap, MetaMask, Coinbase, and PancakeSwap among its integrations and markets itself to users with a “Why Go Direct” argument against aggregators. Relay has positioned differently, describing itself in its own documentation as an enterprise API offering “instant global access to the onchain economy” through “a single onchain balance,” with use cases spanning onchain payment acceptance, bridging, same-chain and cross-chain swaps, and cross-chain calls. Relay’s counterparty is the application, not the end user. In both cases the entity choosing the routing layer is a wallet, exchange, or application with its own commercial considerations, and the end user never sees a solver set, let alone chooses from it.

AcrossRelay
Self-descriptionIntent-based bridge with a permissionless relayer networkDeveloper platform for multichain payments, bridging, swaps, and cross-chain calls via API
Reported scale$40B+ cumulative volume, 5M users, 1.2s average fill (vendor-reported, homepage as of Nov 2026)No volume, chain-count, or fee figures disclosed in its public documentation
Stated filler model”Anyone can compete to fill transfers”Not described in the reviewed documentation
Named distributionUniswap, MetaMask, Coinbase, PancakeSwapApplications integrating the API; no named partners in the reviewed pages

Chain-count and team-count figures circulating around this sector, including a commonly repeated count of 85+ chains for Relay and 30+ teams in the Ethereum Foundation’s Open Intents Framework, do not appear in the primary sources we reviewed, and we treat them as unverified.

Large transfers appear to route differently from retail flow. Across describes an integration with Paxos Labs’ Transit enabling stablecoin transfers of up to $50 million to Robinhood Chain at a locked rate. A locked rate on a nine-figure-capable channel is not a competitive fill auction among relayers; it is a bilateral arrangement with a specialized capital provider, surfaced through the same interface. That is a reasonable product decision, since no general relayer pool holds $50 million of idle inventory waiting on a single order, and it illustrates the point: at the sizes where intermediation economics matter most, the system resolves to a named counterparty with a balance sheet.

Resource-lock designs aim at a different part of the problem, reducing or eliminating the window during which the solver’s capital is exposed by locking the user’s funds in a way the solver can rely on before it fills. If such designs work as described, the inventory-to-throughput ratio falls and the capital bar falls with it. We were unable to verify the specific claims and funding details of the most-cited project in this category; the source we attempted to use returned a 404, and we do not cite it as evidence.

The case that concentration is transitional

The strongest argument against treating solver concentration as structural is that nothing observed so far distinguishes an entrenched market from a young one, and the mechanisms that would lower the capital bar are actively being built.

Four versions of that argument hold up. Capital requirements fall mechanically when reimbursement latency falls, so faster settlement proofs and resource locks directly shrink the moat rather than merely redistributing it. Shared liquidity layers and bonded solver designs let participants fill against capital they do not own, converting a balance-sheet requirement into an underwriting one. Multiple competing standards and frameworks fragment infrastructure investment, which makes any single network’s incumbency harder to defend. And concentration produced by performance selection, where solvers with better latency, uptime, and pricing win more flow, is what a functioning competitive market looks like rather than evidence of capture.

There is also a fair challenge to the premise. Concentration is only a user problem if it shows up in prices or in reliability. If fills clear in seconds at spreads competitive with the alternatives, a user has no stake in whether three firms or three hundred provided the inventory. The place where the distinction becomes material is stress: a concentrated filler market withdraws together. When one chain’s finality is in question, when a stablecoin is depegging, or when gas on a destination chain spikes, a handful of solvers running similar models will widen or stop quoting at the same moment, and the intent interface will simply fail to fill rather than filling expensively. The resilience claim and the concentration claim are the same claim, tested on a day nobody schedules.

What would settle the question

The intents thesis rests on an empirical claim about market structure that no public dataset currently supports or refutes. Four measurements would resolve most of it.

The first is per-route fill concentration: the share of fill volume captured by the top one, three, and ten addresses on each major corridor, tracked over time, together with the count of distinct addresses filling at all and the rate at which new ones appear and disappear. Fill transactions are on-chain and attributable. This is a dashboard, not a research program, and its absence from public analytics is itself notable given how central filler competition is to the standard’s stated value proposition.

The second is settlement latency distribution, not the headline fill speed. Median and tail reimbursement times per implementation determine the inventory-to-throughput ratio that sets the capital bar. A protocol that fills in 1.2 seconds and reimburses in four hours is running a very different market from one that reimburses in four minutes, and both can advertise the same number.

The third is spread dispersion by route, correlated with flow imbalance. If rebalancing is the cost it appears to be, one-way corridors should carry visibly wider spreads, and those spreads should compress when flow turns two-way. This is the cleanest available proxy for a cost structure no solver discloses.

The fourth is behavior under stress. Fill rates, quote widths, and the number of distinct active fillers during a destination-chain congestion event or a sharp price move tell more about the market’s structure than any quiet-period volume figure.

Until those exist, the honest position is that ERC-7683 standardized the order and left the balance sheet alone, that the capital-fronting requirement gives well-funded participants advantages that compound through netting and fixed-cost amortization, and that distribution partnerships determine which of them sees the flow. Whether that produces a durable oligopoly or a competitive market that merely looks concentrated early is unresolved. The next time a major chain’s finality comes into question and intent fills either continue or quietly stop, the answer will be visible to anyone watching the fill rate.

References

Read next

Follow the next market structure breakdown

New Steadyrain research is published several times a week across DeFi risk, BTCFi, stablecoins, and RWA.