An exchange rented out its book and kept 14%

On August 24 a team out of Citadel Securities and Optiver opened perps on stocks and on stakes in private companies on Hyperliquid, including a contract on Anthropic. On August 25 LayerZero unveiled ATLAS and said plainly what it is for: a trading venue should not have to build on a competitor's rails. We took a census of every market already built on those rails and counted how much money is actually there.

The anomaly

The venue widely treated as the best place to trade perps in crypto now carries 269 markets it did not build. Outside operators built them: anyone who freezes 500,000 of the venue's tokens earns the right to open a market on its order book and keep half the fee.

A market here is one ticker: 269 markets means 269 tickers, from the S&P 500 to a stake in SpaceX. We went through them one by one on August 25 at 15:01 UTC. 125 of the 269 hold no open position and no dollar of volume in 24 hours. That is 46.5% of the storefront.

Of the ones that do work, 99.3% of the money sits with a single operator.

How we counted

The numbers come from the venue itself, through a public request anyone can make without keys or a subscription. The first request returns the list of operators. The second returns each operator's markets. The third returns, per market, the price, the money held in positions, the 24-hour volume, the position cap, the maximum leverage and the fee mode. Daily history was collected as candles over 120 days across all 269 markets and all 232 markets of the venue itself.

Money in positions is the sum of everything open right now. Volume is what passed through in a day. Volume divided by money in positions we call churn: it shows how many times a position turned over during the day.

We did not estimate the fee by eye. The venue publishes the formula: the rate follows from two market settings, the operator's share and growth mode. We took it as published and ran it across every market.

Figures for neighbouring venues, for backed stock tokens and for the venue's own till come from 3 public trackers: a perp summary across 44 venues, a dashboard for this venue, and an open token catalogue. We do not recompute them. We use them as a check: where our number and a third-party number disagree, we print both.

One caveat up front. The fee depends on how you trade. Whoever takes an existing order off the book pays the full rate. Whoever posts an order and waits pays three times less. The venue does not publish which share of volume does the former.

So we compute the till 3 times: as if all volume paid the full rate, at 60% and at 50%. Where a single figure appears below, it is the upper bound.

The storefront

The outside markets are mostly not crypto. They are the S&P 500, gold, oil, silver, shares of Nvidia, Google, Tesla, Micron memory, South Korea's SK Hynix, a stake in SpaceX. All of it as perps: contracts with no expiry, settled in stablecoins, trading around the clock including the weekend, when the exchange in New York is shut.

The storefront is large. It is filled unevenly.

The top 12 markets hold 68% of all the money. The top 20 hold 83%. After that comes a long tail of tickers somebody opened once. 56 markets have not seen a single trade in all 120 days of observation. Another 69 did not trade in the past 24 hours.

The count of live markets in a day is not growing. On June 8 it was 162, on July 6 it was 109, today it is 142.

One tenant

The system is designed to be open: anyone can open a market. Ten operators did. The money went to one.

Trade XYZ holds $3.73B in positions out of $3.76B in the whole system. The second operator holds $12.8M, 291 times less. This is not a one-off day: Trade XYZ's share of weekly volume has not dropped below 97.5% in any week since late April.

Inside Trade XYZ the money is bunched as well. Three markets, the S&P 500, gold and SK Hynix, hold a third of its positions.

The right to open a market really does belong to anyone. The money, meanwhile, sits with one operator out of 10, and has done so for all 4 months we can see.

The till

Here is what we were counting for.

In 24 hours the outside markets produced 20.9% of the venue's total volume and 28.2% of all open positions. Fees on them, at the upper bound, came to $358K. Of that the venue keeps $180K and the operator keeps $178K.

The same volume on the venue's own markets would have brought it $1.32M.

So for a fifth of its own volume the venue collects 14% of the money that fifth could have earned it. The proportion holds across all 3 splits of taker and maker, because it is set by configuration rather than by trader behaviour.

An outside check: a third-party fee tracker gives the venue $4.22M for the day. Our model for its own markets gives $4.99M if all volume pays the full rate and $3.33M if half of it does. The external number lands inside our range, which means the formula is applied correctly.

The switch

The gap comes from one setting. It is called growth mode and it cuts a market's fee tenfold: a 90% discount for the trader and for both sides that split the fee.

The intent is clear enough: nobody opens a new market if trading on it costs more than on an old one. The discount is the price of getting a market started at all.

Today growth mode is on for 190 of the 269 markets, and 96% of outside volume runs under it.

The discount was meant to be a starting price and works as the normal one: it has been on for a fourth month and covers nearly all of the storefront's volume.

From that follows something to keep in mind in any conversation about this venue's revenue: while the switch stays on growth, volume growth barely turns into money. The difference between "this system feeds the venue" and "this system feeds the operator" is not market force. It is one field in a market setting.

What the quarterly numbers show

Our census is one day. The quarterly figures say the same thing more slowly.

Gross venue revenue has fallen four quarters in a row: roughly $357M in the third quarter of 2025, then $295M, then $217M, then about $202M in the second quarter of 2026. That is 43% off the peak, booked while the trade count climbed.

The share of revenue that leaves the building was under 6% in the second quarter of 2025. A year later it is 18%. It leaves to 3 recipients: outside operators, the people who keep orders on the book, and the venue's own liquidity vault.

The buyback of the venue's own token, the machine the token is held for, shrank from $290M a quarter to $149M.

Positions on the venue are at records over the same stretch, and its share of world open interest in perps rose from 7% to 9%. More trades, less money from each.

How the share moved

One more cut we computed ourselves, because we could not find it anywhere: what part of venue volume the outside markets take, day by day.

The picture is not the one usually retold. Over the past 60 days the outside share is 45.1%, over 30 days also 45.1%, over 14 days already 31.3%, and over the past week 22.9%.

The share is not growing. It has been shrinking for a third week.

The peak fell on July 30: 75.2% of all venue volume in a single day went through operator markets. On August 18 it was still 59.5%. On August 22 it was 4.2%.

A cross-check: a public dashboard for this venue puts the RWA share, smoothed over 7 days, at 21.6%. Our independent count from candles is 22.9%. The difference is that the dashboard counts closed days while we include the current one.

The reason for the squeeze is visible in the same data, and it is not about stocks. Between August 19 and 24 the venue's own markets did $10B to $17B a day against a usual $4B to $5B: crypto was moving. Outside volume stayed where it was.

Hence the rule for reading any number about this system. The operator share swings between 4% and 75% depending on what bitcoin did that week. One date proves nothing, ours included, which is why we print 4 windows at once.

The new arrivals

On August 24 a new operator appeared on this storefront. The Entropy team raised $14M from Ribbit Capital and froze $40M in the venue's token to earn the right to its own markets. By its own account, these are researchers and traders out of Citadel Securities, Optiver, Polymarket and Millennium.

The first contract is a perp on SanDisk stock. The second is a perp on a stake in Anthropic, a private company. The contract runs to August 18, 2028, the price is held inside a band from 300 to 4,200, and leverage is capped at 3.

Six days after launch: $2.97M of positions on SanDisk against a $5M cap, and $2.92M on Anthropic against the same cap. That is 59% and 58% full.

Those are the two fullest caps in the entire system after gold.

The volume is out of proportion: the SanDisk contract turned over $51.7M in 24 hours on $2.97M of positions. Churn of 17.4, meaning the position flips 17 times a day. On the neighbouring contract with the same name at Trade XYZ, churn is 1.5.

We do not know who trades there. We do know that with a 90% fee discount such churn is cheap, and that a new market with a queue at the door looks exactly like this.

Same asset, different counter

Next door sits Ventuals, an operator whose markets have been open since spring. It carries contracts on exactly the same names: Anthropic, OpenAI, SpaceX.

Open positions on all three are zero. Volume is zero. Not one contract has passed through them in 120 days.

The difference is not the asset and not demand for it: demand is visible, the neighbour filled its cap in 6 days. The difference is who runs the market, who holds a price there and what it costs to get in.

That answers the question this story attracts most often: won't ten more of these appear? They did. Ten operators, 269 markets. Two of them hold more than $10M.

How much room is left

Every market has a cap: how much money it is allowed to hold in positions. Caps are published for 232 markets and add up to $17.32B.

They are 21.7% full.

Exactly three markets stand above half of their cap: gold at 54.6% and both markets of the new arrival at 59.4% and 58.4%.

This matters for the argument about the future. To a supporter of this system the caps say there is four times more room than is taken. To a critic they say the opposite: the room exists and there are no takers, since 46.5% of markets are not occupied at all.

The risk of a single price source

A tenant monopoly has a price, and the bill has already been presented once.

On Monday of last week a single trade on a thin Korean pre-market venue dropped Trade XYZ's SK Hynix contract by 19% and forced liquidations. The operator has agreed to reimburse them.

The SK Hynix contract is the third largest in the system by money: $377.5M of positions and $480M of volume in 24 hours.

The dependency is simple from there. The price used to settle a perp is not conjured from air: it is taken from an outside venue, and the operator picks that source. The same operator decides how much collateral to require and when to close a position by force.

So the venue's record numbers rest on the decisions of one operator. If it is wrong, it pays, and it will look like a failure of the whole venue.

Neighbours: an engine with no storefront

On August 25, the day after Entropy launched, LayerZero unveiled ATLAS. They call it a headless exchange: the insides are there, the face is not. Inside are matching, checking and settlement. Outside there is no site, no app and no customers of its own. The storefront is built by whoever rents the engine.

The wording in their own post lands squarely on the model we had been counting all day:

"Trading venues should not have to build on infrastructure that siphons their own users away from them. ATLAS makes this possible: instead of having to compete with the underlying exchange on price, trading venues can now focus on delivering the best possible product experience."

LayerZero, August 25, 2026

The money is split differently. A venue freezes the network's token and gets a fee discount for it. Of what is left after the discount, the market creator takes 25% and 75% goes to buying the network's token and burning it.

So in one model the market operator gets up to 50% of the fee and the venue lives on the rest. In the other the operator gets a quarter and the network takes three quarters.

The claimed speed: median under a millisecond, 1.418 ms at the 95th percentile, 200,000 trades per second at launch. The first partners are GTE, Bullish, Defined and TrueNorth. Launch is promised before the end of the year.

None of that can be checked today, because the engine is not running. All we can honestly do is record the claim and the date, and compare in a quarter.

Neighbours: your own land

The third way has been running for a third year and looks the dullest of all. Kraken is building its own network, Ink, and on it its own exchange, Nado. Spot, perps and one shared collateral pool across positions: up to 20x leverage, 77 pairs.

The numbers on August 25: $363M of volume in 24 hours, $73M of open positions, $50.6M in its vaults. Churn of 4.93, which puts Nado in the same group as the venues that hand out points.

It holds 51 times less in positions than the tenant storefront on Hyperliquid, and 132 times less than the venue itself.

On the other hand it is the only one of the 3 options where nobody can change your fee with one field in a market setting.

Three ways to hold a stock

A perp on a stock is not the only way. While some built synthetics, Binance issued backed tokens on the same names: bStocks on BNB Chain, one token per real share held by a custodian. On August 12 GameStop was added and accepted as margin collateral, and the Binance wallet opened a separate storefront for such paper.

The first five bStocks names are Nvidia, Tesla, Circle, Micron and SanDisk. The same five names stand as perps on the Hyperliquid storefront.

Counting from public data on August 25. Binance has 62 backed stock tokens worth $498.2M with $355.1M of volume in 24 hours. The neighbouring issue on Solana has 100 tokens worth $798.6M, but only $76.7M of volume.

Positions in perps on Hyperliquid's outside markets: $3.76B.

So there is almost three times more unbacked synthetic exposure on the storefront than all backed stock tokens on the two main platforms combined. With leverage that is exactly what you would expect: $3B of positions do not require $3B of shares at a custodian. But it is also why the phrase "billions of stocks trade onchain" needs a note about which billions.

The same SanDisk: $158.7M of positions in the Trade XYZ perp, $45.5M of issued backed token at Binance, $3.0M of positions at the new arrival. Volume is nearly the same on the first two: $234.5M in the perp against $49.6M in the token.

The difference between those three lines is not demand for SanDisk. It is that the first and third let you take leverage, and in the second the share actually sits in an account.

Who buys the volume

Now the neighbours outside this venue, because there the same disease goes by another name.

A summary across 44 perp venues on August 25: $218.24B of volume in 24 hours against $112.27B of open positions.

Churn, volume divided by positions: Hyperliquid 1.08, Trade XYZ 0.75. Lighter 3.11, EdgeX 3.08, Extended 3.19, Variational 2.94, Nado 4.93, Pacifica 12.48.

The whole top group shares one trait: points handed out for volume. What happens when the handout ends is covered below, in the Lighter section.

That is not an accusation, but it is a question. And it is not the desk asking it: the people farming the handout ask it themselves. On Reddit the same week, someone published their costs after pushing $1M of volume through Variational for points. They held equal positions on both sides so the price would not matter. The season runs to September 30 and the venue charges no fee.

Such volume is real in the sense that money moved. It is unreal in the sense people read volume numbers for: there is no opinion about price behind it.

Hence the rule we apply to Hyperliquid's outside markets and to any venue with points. Volume at a free or nearly free fee is not comparable to volume at the full rate. What can be compared is positions: they are held with money under any fee regime.

Two venues from the summary do not belong in the comparison: Coinbase International has churn of 40 and Kalshi 51.6. Different product, different contract lifetime, and putting them in one row means measuring different things with one ruler.

Lighter: what happens when the handout ends

This venue deserves its own section, because it is this year's best example of how volume differs from demand.

Lighter is a perp exchange with its own order book, built as a separate layer over Ethereum with a mathematical proof that matching is honest. There is no fee for retail at all. Behind it are a16z and Lightspeed.

The numbers that follow are of different quality, so we keep them apart.

The venue's own claim: in the 30 days before its token drop, $232.3B passed through it. That is more than Hyperliquid or Aster over the same period. After the drop, by the same write-ups, $39B remained over 30 days. We did not recompute those 2 figures and print them as a claim.

Our own measurement on August 25: $3.11B in 24 hours against $1.00B in positions. Churn of 3.11, so a position turns over 3 times a day.

If the claim holds, volume fell sixfold exactly when it stopped being paid for in points. That is the answer to how much demand such volume contains.

The token was distributed like this: a quarter of supply straight to the people who collected points, 26% to the team and 24% to investors, both with a one-year cliff and three years of vesting. Today it trades at $3.35 with a market cap of $836.6M.

One detail closer to home than the rest: a perp on that very token trades on Hyperliquid and holds $159.6M of positions. Bets on a competitor's success are taken by the venue it competes with, which earns on them.

The conclusion on Lighter is short. Technically it is strong work and it holds a top-five spot. But its own chart before and after the handout is the main argument for our rule: compare venues by money in positions, not by volume.

What the network token is worth

Three models, three tokens. These are effectively bets on who ends up collecting the fee.

The Hyperliquid token trades at $80.79 with a market cap of about $17.97B. Behind it is a buyback funded from 97% of venue fees, the machine that has shrunk by nearly half in a year.

The LayerZero token trades at $1.20 with a market cap of $422.6M. After ATLAS launches it is promised a second job: 75% of the fee left after venue discounts goes to buying and burning.

The gap in market cap between the two is 42 times. That is roughly the gap between a working till and a promised one. Our interest is not a price forecast but a checkable date: the engine is promised before year end, the partners are named, and in a quarter it will be clear how much volume they brought.

The licence as a weapon: the US, the CFTC and Robinhood

Everything above happens outside American regulation. Since May 2026 that is no longer the only option.

On May 29 the CFTC approved the first perpetual contract on a regulated US exchange: Kalshi's BTCPERP, cash settled, referencing the spot price of bitcoin. Three documents came out the same day: a policy statement on listing perpetual contracts, staff guidance on running an exchange and clearing around the clock, and a letter on Coinbase that allows Deribit perps to be treated as foreign futures.

In plain terms: perps came onshore in the US. Not all of them and not at once, since each new contract is reviewed case by case, but the door is open.

In our venue summary Kalshi shows $1.06B of volume in 24 hours against $21M in positions. Churn of 51.6, which is why we left it out of the desk ranking: different product, different contract lifetime. By volume, though, it is already comparable to a large perp DEX.

What Robinhood is doing

Robinhood is assembling the same winning shape we described above, only from the other side.

Its perps already run in Europe across 30 countries: crypto first, then commodities, ETFs and currencies with up to 10x leverage. The list includes gold, silver, Brent and WTI, the euro against the dollar, QQQ and even the Korean EWY fund, the same names that stand as perps on the Hyperliquid storefront.

It has its own network too: Robinhood Chain, a layer over Arbitrum built for tokenized assets and round-the-clock trading. It issues stock tokens for customers in the European Union.

And above all, a licence. Through a joint venture with Susquehanna it gains control of an exchange and a clearinghouse licensed by the CFTC for event markets. That is the third element none of our other characters has: the right to serve American retail legally, in its own jurisdiction.

Why this matters for our bet

Hyperliquid's advantage is one margin pool and free listing. Neither is granted or withdrawn by an American regulator: both are technical.

Access to the American customer, on the other hand, is granted by exactly that regulator. And there the picture reverses: Hyperliquid does not have it, Robinhood and Kalshi do.

So our forecast splits in two, and we print both halves.

Onchain derivative liquidity stays with Hyperliquid: one margin pool is not going anywhere, and no licence is needed to serve a global customer.

American retail will go elsewhere. It will go to whoever has a licence and a familiar app, and today that is Robinhood with Kalshi. These are different markets, and mixing them is a mistake we nearly made ourselves.

A checkable claim for 12 months: licensed American venues will not take position share from Hyperliquid, but they will pass it on customer count. The first is checked with a public request, the second with Robinhood's quarterly reports, which are open.

Three ways to become an exchange: the table

The whole August argument comes down to a choice between 3 rows. We put them in one table and scored each on 5 traits that matter to whoever is opening the exchange.

The gap between first and third is small, and it rests on one trait, money today. Renting the book wins because $3.76B already stands there, while the engine has zero, because it does not exist yet.

If in a quarter the engine has its first venues with real positions, the order in this table changes without a single edit to the other traits.

Venue ranking: where the money is and where only volume is

The second ranking we computed for ourselves, because the desk has to decide where to hold a position at all. It comes from 3 numbers, each taken from public data on August 25.

First, how much money actually stands in positions. Second, honesty of volume: the more often a position turns over in a day, the lower the score, because volume at a free fee does not mean what volume at a full fee means. Third, how many instruments are available.

The weights differ: money in positions 50%, honesty of volume 30%, choice of instruments 20%.

Two caveats, without which this table lies.

First: a low place does not mean a bad venue. Entropy is last because it is 6 days old and has 4 markets. In a month its row will look different.

Second: high churn is not a verdict by itself. It only means volume on that venue cannot be compared directly with volume where the fee is paid in full.

What the ranking tells the desk right now: the money stands in 2 places, Hyperliquid and its own tenant Trade XYZ. Together they hold $13.35B of the $17.59B in positions across the list, or 75.9%.

Who will lead, and why

What follows is the desk's bet, not a measurement. We separate it from everything above: numbers there, a conclusion here that may turn out wrong. The test conditions are written at the end of the section.

The TT desk is betting on Hyperliquid. Not on the token and not on the price: on this venue staying the place where the main onchain derivative liquidity sits when stocks arrive.

The rule we work from: the leader is whoever has one margin pool serving every market at once.

The reason is simple and not technical. A trader needs the money in the account to work for everything held. When gold, Nvidia, bitcoin and a stake in SpaceX sit under one collateral pool, free money is counted once. When the same positions are spread across 4 venues, each demands its own buffer and half the capital idles.

That is why the outside markets on Hyperliquid gathered $3.76B in 10 months, while backed tokens on the same names gathered $1.30B over the same period with a far cleaner legal structure.

Three conditions for winning

The winner is not the fastest venue but the one with 3 things at once.

First: one margin pool across all markets. Second: the right to open a new market without permission. Third: its own till, meaning the fee that stays with the venue.

Hyperliquid has the first and the second. It has no till on outside markets: 14% stays.

ATLAS from LayerZero has the second and the third. It has no liquidity at all, because the engine is not running.

Nado has the first and the third. Listing is closed: the exchange decides the list.

Binance bStocks has a till and real shares at a custodian, but neither free listing nor one margin pool: issuing a token means buying and storing an actual share.

Nobody has all 3. So leadership right now is decided not by who is built better, but by who closes their own gap first.

Our forecast on the gaps: the till is the easiest to fix, one field in a setting, and the venue has already announced plans to raise fees on outside markets. Liquidity from zero takes years, as Nado shows with a third year of work and $73M of positions. Opening a listing that lawyers closed is the hardest of all.

Hence the order: Hyperliquid fixes its gap in one release, ATLAS earns its own over years, Binance does not fix its own at all, because a closed listing is the product.

Why stocks will take the main share of liquidity

The first reason is size. All the world's stocks are worth $151.94T, all of crypto $2.68T. A gap of 56.7 times. Even 1% of that market arriving for round-the-clock trading is larger than the entire daily volume of crypto.

The second reason is that a perp is cheaper than a warehouse. Issuing a backed token means buying a share, placing it with a custodian and keeping the paperwork. Opening a perp needs a price source and stablecoin collateral. That is why synthetics already outweigh backed tokens threefold, even though backed tokens started earlier.

The third reason is time. The New York exchange runs 6.5 hours on weekdays, roughly a fifth of the week. For the other four fifths a person holding news can do nothing, and that window is where stock perps live.

The fourth reason has already happened and is not a forecast. On Hyperliquid, perps on stocks and commodities passed bitcoin in money held in positions and became the venue's largest market. Between July 13 and 19 they did $25B of volume, 52% of the week, outpacing crypto for the first time.

The fifth reason is that money is arriving from that side. Behind Zero stand Citadel Securities, DTCC and the owner of the New York Stock Exchange. The team of the newest operator on Hyperliquid came from Citadel Securities, Optiver and Millennium. These are not crypto funds. They are trading houses from a market 56.7 times larger.

What this bet means for the desk in practice

A bet is not an opinion, it is a decision about work. Four things follow from it, all of them done or queued.

First. Perps on stocks, commodities and indices enter the desk's instrument list as a separate group, not as an oddity next to crypto. They have their own market phase: the New York session and the hours after it behave differently, and measuring them with the same ruler as bitcoin does not work.

Second. Every such market is checked for churn before we even look at it. A contract with churn of 17 and a contract with churn of 1.5 under one ticker are different books.

Third. Operator risk is counted separately from venue risk. Our instrument passport gains a field for who runs this market: 99.3% of the money sits with one operator, and the SK Hynix incident has already shown the price of its mistake.

Fourth. We keep watching the competition. ATLAS launches before year end, Nado is in its third year. Betting on Hyperliquid does not mean we stopped counting the others: it means that until the neighbours have numbers, we work where the money already stands.

What we claim and how to check it

Three claims for 12 months. Each has a number and a way to check, so that in a year the argument is about a fact rather than a memory.

First. The share of stock, commodity and index markets in Hyperliquid's money in positions stays above 40%. Today it is 28.2% of the whole venue and almost all of it in outside markets. The check is the same public request, open to anyone.

Second. The discount gets cut, and after that outside volume falls by less than half. This is the central argument: the sceptic says all the volume rests on a cheap fee. The check is the share of volume under the discount, today 96%.

Third. No new engine gathers more than 10% of Hyperliquid's money in positions within a year. Today ATLAS has zero and Nado has $73M against $9.63B.

What would break our bet

One margin pool stops being an advantage if large venues learn to net collateral between themselves. Then liquidity stops sticking to one place and the winner is whoever holds the till.

One operator is one mistake. 99.3% of the money sits with Trade XYZ, and the SK Hynix case showed what such a mistake costs. A second case without reimbursement sends that money looking for another home.

Regulators hit synthetics harder than backed tokens. A perp on a stock carries no vote and no dividend, and that is its legal weak spot. If retail access is closed, a professional market remains, and it is smaller.

What the other side says

The August argument is not about whether onchain stock trading works. It is about who gets the money.

"We are releasing the first liquid way to trade Anthropic. For far too long, frontier assets have only been accessible to a select few. We are here to bring them to the masses."

Entropy, August 24, 2026

"Builders on Hyperliquid have made impressive progress towards this vision, but some problems remain unsolved. Novel assets like Pre-IPO markets have had short-dated requirements or near-zero liquidity. Instruments on public market assets can suffer from high funding rates, manipulation during non-trading hours, and thin liquidity."

Entropy, August 24, 2026

"HIP-3 perp markets split fees 50% between the deployer and the protocol, aligning incentives between participants, and importantly still growing the pie for Hyperliquid."

FalconX on the HIP-3 model

"If Growth Mode is enabled, taker fees drop by ~90%, deployer revenue collapses."

ether.fi Ventures on HIP-3 economics

"When growth mode is on, the protocol cuts its own side of the fee by 90%. HIP-3 is not funding the HYPE buyback today."

Max Michielsen on Trade XYZ fee rates

"The biggest threat to Hyperliquid isn't Polymarket. It's Hyperliquid. Everyone is celebrating the HYPE token. Almost nobody is asking whether Hyperliquid is actually capturing users, fees, revenue, and distribution."

Valentin Martynov, August 10, 2026

The bull case is stated plainly and it is measurable: the current discount is temporary, the venue has already announced plans to raise outside market fees by up to threefold, and then the same volume yields several times more revenue.

The sceptics have the same facts and the opposite conclusion. While the discount is on, outside markets do not feed the token buyback. And the volume used to measure success costs the venue part of its own till.

Our measurement does not pick a side. It puts a number on the argument: 14%. That is what the venue collects today from an outside market against what it would collect from its own.

What we cannot see

Who pays the full rate. We do not know what share of volume pays the full fee. The error runs one way: if the full rate is paid less often than we assumed, the real till is lower in all 3 columns. The 14% proportion does not change.

Private discounts. Large participants pay less than the published rate, and the size of those discounts is not published. So the real till is below our upper bound, again on both sides.

Who trades. A public request returns the market, not the participants. This method cannot separate genuine flow from churn farmed for points, which is why we never write the word manipulation: we compute churn and name it out loud.

One day. The census was taken on one date. Volume has history, positions do not: candles store trade volume, not how much money stood in the market that day. So "46.5% of markets are empty" is verified for today, while "56 markets never traded" is verified over 120 days.

Private agreements. What an operator really receives after arrangements with the venue is not visible in a public request. We count by the published formula.

What would have to be true for us to be wrong

If the discount switch is turned off and volume does not fall, our conclusion about a setting rather than a market is weak: it would mean traders pay for access to the asset, not for cheapness. Checked in one quarter, through the share of volume under growth mode.

If the second and third operators reach even 10% of the system's money, the single-tenant claim dies. Today all nine together hold 0.7%.

If empty markets gain positions with no new demand, then our 46.5% measured a storefront rather than demand.

If venue revenue turns up in the third quarter of 2026 with the discount unchanged, then the source of the decline is not the fee split, and we will have to look for it again.

Corrections made along the way

First. The initial fee model treated every outside market as "double rate, split in half". The documentation shows that 25 markets have a different operator share, which changes the formula for them. The numbers were recomputed by the published formula; the final proportion did not move, because those 25 markets make 0.1% of volume. We write it down anyway: a 0.1% discrepancy today is a discrepancy in the model, not in the data, and tomorrow it could cost more.

Second, and this is the largest correction. The first version of this text said the outside share of venue volume is growing: 33.1% at the start of the window against 45.1% over the past 30 days. That is true for a monthly window and false for today. Adding 14-day and 7-day windows showed the opposite move: 31.3% and 22.9%. A third-party dashboard on the same date gives 21.6%. The conclusion was rewritten: the share has been shrinking for a third week, and the monthly figure stays high because of the late-July peak.

Third. We came across a claim that outside markets make about 50% of venue volume. Our census on August 25 gives 20.9%, and that is not a refutation: the 30-day average is 45.1% and the daily range runs from 4% to 75%. We print both figures together, because either one alone misleads.

Notes from the TT desk

All year we treated stock perps as a curious storefront. The census changes one working thing for us. A market where large money stands and a market where large volume runs are different markets. Confusing them is expensive.

So one field is added to our instrument selection. Churn is now computed per market rather than per venue. The threshold below which we look at a market at all will be announced once we have that number across every outside market over 120 days. The collection is already running.

One number we watch from here: the share of volume running under the discount. Today it is 96%. While it stays above 80%, any conversation about this venue's revenue is a conversation about a future decision by its team, not about trader behaviour.

And one request. Check the 14% behind us. The request is public, the fee formula is published, and the census repeats in 5 minutes on any machine. If your number differs, we want to know before we put other people's money on ours.

The desk's thesis, briefly. An exchange now earns not only on its own markets but on letting other people's markets in. Volume grows for both sides. Money grows only for whoever controls the fee, and today it is turned in the tenant's favour: the venue collects 14% of what the same volume would have earned it. We do not call this a mistake. It is the price the venue pays for outside markets, and the price is temporary: one field in a setting raises its revenue tenfold. So the desk watches two numbers rather than volume. First, what share of volume runs under the discount, today 96%. Second, how much money stands in positions, today $3.76B, and 99.3% of it with one tenant. The first number shows when the venue decides to take its own back. The second shows whether there will still be anyone to pay by then. The TT desk is betting on Hyperliquid: we think liquidity stays where one margin pool pulls it.