# After the Flippening: Auditing the Perp DEX Wars

> Onchain perps reached 26% of futures volume, but Hyperliquid's 70% share claim fails basic arithmetic. An audit of architectures and moats.

- Source: https://www.steadyrain.xyz/en/articles/perp-dex-wars-2026-hyperliquid-challengers/
- Publisher: Steady Rain
- Author: 0xYatha
- Published: 2026-09-28T04:03:20.687Z
- Categories: DeFi
- Language: en

## Introduction

In October 2025, perpetual futures DEXes settled $1.049 trillion in a single month, the first time the category passed a trillion. The first week of November added $340 billion. By late 2025, monthly volume had moved past $1.2 trillion, and Coinbase Institutional's 2026 Market Outlook, as cited by BlockEden, placed onchain venues at roughly 26% of global crypto futures volume, up from 2.7% in 2023. For scale, every perp DEX combined produced $64.76 billion across the whole of 2023.

The growth figures are repeated almost everywhere. The claim usually attached to them is that Hyperliquid holds about 70% of onchain perpetual volume, making the category a one-venue market with a fringe of challengers. That claim does not survive division. Using figures drawn from the same two sources that assert it, Hyperliquid's reported 30-day volume implies a share somewhere in the twenties once Lighter, Aster, and edgeX are included in the denominator. One of the numbers is measuring a venue set that no longer describes the market.

This matters beyond scorekeeping. If Hyperliquid holds seventy percent of a trillion-dollar monthly market while charging fees and booking over $650 million in revenue by the end of 2025, the reasonable conclusion is that order book quality compounds and the race is effectively over. If it holds a quarter against three venues running zero-fee or emissions-funded books, the reasonable conclusion is that onchain perp liquidity is far more contestable than the dominance narrative suggests, and that a meaningful slice of the $1.2 trillion is rented rather than owned.

Both available sources are secondary. Neither publishes a methodology for how volume is counted, whether self-trading under points programs is filtered, or how figures are deduplicated across aggregation layers. Everything below treats their numbers as claims to be cross-checked against each other, not as measurements.

## The dominance number does not divide

*[Figure: img1]*

Start with the one cross-check that works. BlockEden reports Aster at roughly 20% of global perp DEX market share in September 2025, on $187.9 billion of monthly volume. That implies a total onchain market of about $940 billion for the month, which sits just under the $1.049 trillion recorded in October. Two independently stated figures produce a consistent denominator, which is weak evidence that the aggregate market-size numbers are roughly right.

Now apply the same arithmetic to Hyperliquid. Yellow, citing DefiLlama, reports Hyperliquid's 30-day perpetual volume at more than $180 billion as of April 2026, or roughly $6 billion per day. BlockEden's separate framing of weekly volume above $40 billion lands in the same neighbourhood. Against a market running at $1.2 trillion per month, $180 billion is about 15%. Against the $940 billion implied by Aster's September share, it is about 19%.

Build the denominator bottom-up and the result is similar. Hyperliquid at \~$6 billion daily, Lighter at $3.75 to $4.58 billion in January 2026, Aster above $6 billion, and edgeX at $6.6 billion sum to roughly $23 billion per day before any of the long tail is counted. Hyperliquid's share of that partial sum is about a quarter.

For the 70% figure to hold alongside $180 billion of monthly volume, the entire onchain perp market would have to be running at about $257 billion per month, a decline of roughly 80% from the $1.2 trillion reported for late 2025. Neither source reports a contraction of that magnitude, and BlockEden's January 2026 challenger figures are inconsistent with one.

The likely explanation is visible in Yellow's own comparison set. Its ranking places dYdX second at roughly 10 to 12% of Hyperliquid's volume, with GMX and Synthetix smaller still. Lighter, Aster, edgeX, and Paradex do not appear. Measured against dYdX, GMX, and Synthetix alone, 70% is entirely plausible. That denominator describes the 2024 perp DEX market, not the 2026 one. BlockEden's own phrasing hedges in the same direction, attributing 70% to Hyperliquid "at its peak" rather than currently, while separately describing Hyperliquid as around 32% of decentralised perpetual trading. The two BlockEden figures are hard to reconcile with each other, and the lower one is closer to what the daily volumes produce.

The headline figure on Yellow's piece, that Hyperliquid owns 13% of all perp volume, fails a different test. BlockEden reports that perp DEXes collectively capture only about 10% of CEX perpetual volume, with Binance at $69 billion daily and Bybit at $26 billion. If onchain volume is roughly a tenth of centralised volume, then holding 70% of the onchain market means holding around 6% of the combined total, not 13%. To reach 13% of all perp volume under that ratio, Hyperliquid would need to trade more than the entire onchain market. The headline number is arithmetically impossible as labelled, though it is suspiciously close to the onchain share the daily volumes actually imply.

None of this establishes what Hyperliquid's share is. It establishes that no figure in circulation has been computed against a stated, current denominator, and that the gap between the plausible answer and the repeated one is a factor of roughly three. Anyone sizing the category should recompute from venue-level daily volumes rather than inherit a share number.

## Three ways to build a matching engine, and what each one gives up

*[Figure: img2]*

The venues taking share are not running variations of one design. They resolve the same tension differently: an order book needs microsecond-class matching and deterministic settlement, and general-purpose blockchains supply neither cheaply.

| Venue | Execution architecture | Stated performance | Fee posture | Principal structural exposure |
|---|---|---|---|---|
| Hyperliquid | Custom L1 (HyperCore) with Tendermint BFT-derived consensus, separate HyperEVM for contracts | \~0.2s median block, up to 100k orders/sec, sub-1s finality | Charges fees; >$650M revenue through end-2025 | Security and liveness rest on its own validator set |
| Lighter | zk-rollup on Arbitrum with proven order book | Not stated in available sourcing | Zero fees pre- and post-TGE | Fee revenue absent; operating cost funded from treasury |
| edgeX | StarkEx-based | 200,000 orders/sec claimed, >$10M depth within 1bp | Institutional targeting, terms not disclosed | Proving-system and operator dependency |
| Paradex / Extended | StarkEx-based; grouped at 16% of perp DEX volume with edgeX | Not disclosed separately | Not disclosed | Same; per-venue data unavailable |
| Aster | Multi-chain aggregation across BNB Chain, Ethereum, Solana, Arbitrum | Not stated | "Trade & Earn" incentives, up to 1001x leverage | Cross-chain custody and collateral solvency |

Hyperliquid's design is the most deliberate about the tradeoff. HyperCore runs the matching engine as the chain itself, with a Tendermint BFT-derived consensus targeting roughly 0.2-second median blocks and sub-second finality, per Yellow. It does not execute general-purpose smart contracts. HyperEVM does, and shares state with HyperCore, so an EVM contract can read live order book, margin, and position data without a bridge or an oracle relay. That is the substantive achievement: composability with the book rather than with a wrapper around it. The cost is that Hyperliquid's censorship resistance and liveness are properties of its own validator set, not of Ethereum's. A trader holding collateral there is trusting a specific, relatively new consensus set to keep producing blocks during the exact volatility windows when block production matters most.

Lighter takes the opposite position on that question and pays for it elsewhere. Running as a zk-rollup on Arbitrum, matching correctness can be proven rather than assumed, which removes the operator's ability to misreport fills. Validity proofs constrain what the operator can compute; they do not by themselves constrain what the operator chooses to sequence, or when. Proof generation also imposes a latency and cost floor that a purpose-built L1 does not carry. Lighter absorbs that cost while charging zero fees, which is a financing question rather than an engineering one and is addressed below.

The StarkEx cohort, which BlockEden credits collectively with 16% of perp DEX volume across Paradex, Extended, and edgeX, sits between those poles, outsourcing the proving stack to shared infrastructure. edgeX claims 200,000 orders per second and more than $10 million of liquidity within a one-basis-point spread, which is a market-maker-facing specification rather than a retail one. It reports $515 billion cumulative volume against $6.6 billion daily, a ratio implying the venue accumulated its entire history in roughly eleven weeks. Either it is very young or the daily figure reflects a peak rather than a run rate, and the sourcing does not distinguish. Paradex appears in the available material only as part of that 16% aggregate, with no venue-level architecture, fee, or volume detail, so any claim about its competitive position would be invention.

Aster is the outlier because it is not primarily a matching-engine design. It aggregates liquidity across four chains, which addresses fragmentation from the user's side while reintroducing the bridging and custody surface that single-chain venues avoid by construction. Its 1001x maximum leverage is the more consequential parameter. At that ratio, initial margin is roughly one tenth of one percent of notional, so a ten-basis-point adverse move exhausts the position. Offering that leverage makes the liquidation engine the actual product, because at those margins the venue is not intermediating a trade so much as operating a very fast stop-loss with socialised residual risk.

## Who is paying for the volume

*[Figure: img3]*

Lighter processed $232.3 billion in the 30 days before its December 30, 2025 token generation event, an implied run rate near $7.7 billion per day. By January 2026, BlockEden reports daily volume of $3.75 to $4.58 billion. That is a decline on the order of 40 to 50% within weeks of the emissions event.

The inference is not clean and should not be presented as one. The two figures come from different measurement windows, market-wide conditions changed between them, and no source in this material performs a before-and-after comparison or controls for anything. But the direction is what a points-farming cohort rotating out would produce, and the magnitude is consistent with a large share of pre-TGE flow having been priced off expected token allocation rather than off the trade.

The mechanism contaminates every volume figure in current circulation. When a venue charges zero taker and zero maker fees and accrues points proportional to notional traded, the marginal cost of round-tripping a position against oneself collapses to the spread plus funding. If the participant is also the maker on both sides, the spread cost approaches zero as well. Volume under those conditions is not a demand signal; it is an accounting artifact of the emissions schedule. None of the aggregators cited here disclose whether they attempt to filter it.

That framing cuts against Aster too. Its "Trade & Earn" programme and yield-bearing collateral options are both designed to raise the return on parking capital at the venue rather than to improve execution. A trader posting asBNB or USDF as margin earns a yield on collateral that would otherwise sit idle, which genuinely improves capital efficiency and also embeds a second protocol's solvency directly into the margin calculation. If the yield-bearing asset trades below its assumed value during a stress window, margin requirements tighten precisely when the underlying market is least liquid, and the venue's risk engine is marking collateral it does not control. That dependency is not visible in a volume chart.

Lighter's zero-fee model raises a separate question about who funds the matching engine, proof generation, and market-maker incentives in the absence of fee revenue. The LIT token allocation, split 50% ecosystem and 50% team plus investors per BlockEden, is the answer by elimination: the operating subsidy is capitalised in the token. Yellow cites Galaxy Digital research suggesting that tokens with high institutional pre-sale concentration tend to underperform after listing because early holders supply persistent sell pressure. A 50% insider tranche is a materially different structure from Hyperliquid's, and the valuation projections circulating for Lighter, spanning $1.5 billion to $12.5 billion by end-2026, are wide enough to be uninformative about anything.

Against all of that, Hyperliquid's revenue figure is the most load-bearing data point in this entire landscape. Booking more than $650 million by end-2025 and ranking fourth across the crypto ecosystem by revenue requires flow willing to pay for execution. Subsidised volume can match a fee-charging venue on notional. It cannot match it on revenue, because the subsidy is the reason the volume exists. Revenue per unit of volume is the metric that separates rented liquidity from owned liquidity, and it is the one almost never reported in share comparisons.

## Where a moat could actually form

*[Figure: img4]*

Four candidate moats appear in this market. They are not equally durable.

### Latency

Sub-second finality and six-figure order throughput are the most-cited differentiators and probably the least defensible. For directional retail flow, the difference between 200 milliseconds and 50 milliseconds is not a binding constraint. It binds on the market maker, whose adverse selection cost scales with how long a stale quote sits exposed. Lower latency permits tighter quotes, tighter quotes produce depth, and depth attracts the size that generates fees. The chain runs through market-maker economics, not through trader experience.

Latency is therefore a moat only as long as it is scarce. HyperCore's \~0.2-second blocks, edgeX's claimed 200,000 orders per second, and StarkEx's proving throughput are converging on a range where the residual differences are smaller than the spread improvements a venue can buy outright with rebates. A competitor cannot easily replicate a liquidity network. It can replicate a fast matching engine, and by 2026 several have.

### Listings velocity

HIP-3, launched in Q1 2026, lets third parties deploy perpetual markets on Hyperliquid by staking HYPE, without core-team approval. Within roughly a month, HIP-3 markets reached $1.4 to $1.5 billion in daily volume with open interest growing from $260 million to $790 million. As a distribution mechanism this is strong, because the venue that can list an asset first captures the price discovery window and the funding-rate flow that follows.

The risk filter is the staked bond. A bond disciplines the deployer's incentives but does not validate an oracle specification or the liquidity of the underlying. A thinly traded reference asset with a manipulable price feed produces liquidation cascades whose losses land somewhere. Whether HIP-3 markets are risk-isolated from the main book, or whether they share an insurance backstop and auto-deleveraging queue, is the single most important unanswered question about the framework, and the available sourcing does not address it. Anyone underwriting counterparty exposure to Hyperliquid should resolve it before the volume grows further.

### Collateral flexibility

Yield-bearing collateral and cross-margining raise the return on deposited capital, which is a real switching incentive in a market where most collateral sits idle between trades. Lighter's reported TVL above $1.4 billion against $1.53 billion of open interest in January 2026 implies aggregate leverage close to one, meaning most deposited capital was not carrying risk. That is what a book looks like when deposits are chasing an allocation rather than a position, though open interest and TVL are measured differently enough that the comparison is indicative at best.

### Liquidation-engine trust

This is where a genuine moat can form, because it is the one property a competitor cannot buy with emissions. When a liquidated position cannot be closed at its bankruptcy price, the venue must source the loss somewhere, and auto-deleveraging takes it from profitable traders on the opposite side by force-closing their winning positions. ADL converts a correct trade into an involuntary exit at the worst possible moment. A venue's ADL ranking rules, its insurance fund depth, and its liquidation price bands determine who absorbs that loss, and professional desks size their exposure against those parameters rather than against fee schedules.

The brief that prompted this analysis treats October 2025 as the moment traders began pricing those differences across venues. The available sourcing does not describe such an event, which venues were affected, or what post-mortem conclusions were drawn. What the sources do establish is that October 2025 was the first trillion-dollar month for perp DEXes, which is consistent with a large volatility episode but does not confirm one. We are not asserting an incident we cannot verify.

## Two load-bearing gaps: October 2025 and the CFTC perimeter

*[Figure: img5]*

The strongest available reading of this market rests on two things our sourcing cannot supply.

The first is the mechanics of whatever happened in October 2025. If a major venue's liquidation engine or ADL queue behaved unexpectedly during a stress window, that event would do more to sort competitive positions than any fee schedule, because it would give desks a realised loss distribution to underwrite against rather than a documented policy. Until the specifics are established, the claim that liquidation-engine design has become a differentiator is a hypothesis about trader behaviour, not an observation.

The second is the regulatory perimeter. BlockEden's framing, that perp DEXes capture only about 10% of CEX perpetual volume against Binance's $69 billion and Bybit's $26 billion in daily flow, is presented as headroom. It reads equally well as a ceiling. The capital that has not moved onchain is largely capital that cannot custody collateral at a venue with no legal entity, no segregated client assets, and no recognised resolution process. Whether CME's regulated derivatives complex and the CFTC's treatment of offshore perpetuals pull that flow toward compliant venues or leave the onshore gap open is the variable that determines whether 26% is a waypoint or a plateau. Nothing in our sources addresses it, and we will not speculate about a rulemaking posture we have not read.

## What would have to change for onchain share to move again

The 26% figure was produced by conditions that are individually reversible. A large share of 2025 volume was purchased with emissions, and Lighter's post-TGE decline suggests what that costs when the purchasing stops. Hyperliquid's revenue is the clearest evidence that some meaningful portion of the flow pays for itself, but no source reports revenue per dollar of volume for any challenger, which is the comparison that would settle the question.

Three conditions are observable without waiting for a new report. First, whether Lighter's daily volume stabilises near the $4 billion range or continues falling through 2026, which distinguishes a retained user base from a migration. Second, whether HIP-3 markets are risk-isolated, because permissionless listing velocity is either a distribution advantage or a shared-liability structure depending on an answer the framework documentation should already contain. Third, whether any venue begins publishing insurance fund balances and ADL trigger statistics, which would indicate that professional flow has started demanding the disclosures that make counterparty risk underwritable.

The share figures will keep circulating regardless, because they are easy to quote and nobody recomputes denominators. The arithmetic here took venue-level daily volumes from the same sources that assert 70% and produced roughly a quarter. That gap is the market's actual state of information.

## References

- [The Perp DEX Wars of 2026, BlockEden](https://blockeden.xyz/blog/2026/01/29/perp-dex-wars-2026-hyperliquid-lighter-aster-edgex-paradex-decentralized-derivatives/)
- [Hyperliquid perp volume dominance research, Yellow](https://yellow.com/research/hyperliquid-perp-volume-dominance-how-2026)