IOSG says HIP-3 lowers the gate to launching perpetual markets, but not the moat

HIP-3 may have opened the door to launching perpetual markets on Hyperliquid, but IOSG Ventures says the harder part was never access. It was building something other teams cannot cheaply copy.

In a report by Mario Chow, IOSG said 10 teams have registered their own perpetual venues on Hyperliquid and most of them locked roughly $40 million in HYPE to do so. One venue still controls 97.8% of HIP-3 trading volume, though its own monthly turnover has just fallen 44%. The report asks what the other nine teams actually bought, and bases the answer on onchain data rather than project announcements.

Scope, data sources and measurement windows

Every figure in the report was pulled from Hyperliquid’s public API endpoints, including perpDexs, metaAndAssetCtxs, daily candleSnapshot data for all 519 registered assets, delegatorSummary, userNonFundingLedgerUpdates and clearinghouseState{dex}. Routing data in the final section came from Flowscan because builder-code volume cannot be aggregated from the public API alone.

In the report, “30 days” refers to the full UTC natural-day window from Aug. 15, 2026 to Sept. 13, 2026. “Previous 30 days” refers to July 16 through Aug. 14. “7 days” refers to Sept. 7 through Sept. 13. HYPE was priced at $79.73 for the calculations.

Headline findings: HIP-3 has become a minority again

• Builder-deployed markets accounted for 25.8% of Hyperliquid perpetual volume over the last 30 days, down from 57.1% in the previous 30-day window. IOSG says that shift mostly came from the denominator: core-book volume more than doubled while HIP-3 itself was shrinking.
• Trade[XYZ] posted $64.60 billion in volume over the last 30 days, down 44.2% from the prior 30-day period. Its 7-day average fell from an early-August peak of $5.36 billion per day to $2.01 billion per day.
• Entropy (io) briefly led on Nebius, a market where it competed head-on with the category leader, then gave that lead back. Its share on Nebius moved through a three-week path of 8.7%, 53.1% and 20.4%, while its own volume has now declined for three straight weeks. At the same time, open interest rose 37% to $51.4 million.
• Settlement asset choice still looks decisive. Every venue that used a non-USDC stablecoin has already stopped trading. Every venue still operating uses USDC. IOSG describes the running score as six to six.
• Listing slots are not hard to buy. At the current auction floor, one slot costs about $39,900. Buying every active Paragon market would cost roughly $1.04 million, about two weeks of fee income for Trade[XYZ].
• No one is really competing on price. Stock-focused venues all converged on deployerFeeScale = 1.0 with Growth Mode enabled. Measured fee rates came in at 0.427 bp for Trade[XYZ] and 0.400 bp for Entropy. Outside the leader, all venues combined, dead or alive, have earned only $747,000 in deployer revenue over their entire lives.

Ten deployers in total, and no eleventh entrant

IOSG said only 10 teams have ever registered a perpetual DEX under HIP-3, and no eleventh entrant has appeared. Four are still trading, five have stopped and one never launched.

Trade[XYZ] represented 97.8% of HIP-3 volume over the last 30 days and 97.6% over the last 7 days. The report notes that public API data cannot derive a count of distinct traders, so “trade count” in the study is the sum of the n field in daily candles over the selected window.

Among the challengers, Entropy generated $1.03 billion across 26 days and six live markets. IOSG says the venue’s edge came from its in-house oracle rather than its asset list. Paragon was described as the only challenger whose order books actually looked like order books, with 26 live markets and a broader long tail. Even after Trade[XYZ] switched on five of Paragon’s core names in one go, Paragon’s monthly volume still rose 49.9%. Markets by Kinetiq bought 23 names, but 95% of its volume sat in two index perpetuals. HyENA has already ended operations: every market was delisted, open interest fell to zero and lifetime revenue was just $33,414.

Why HIP-3 share is easy to misread

Measured daily, HIP-3’s share of total Hyperliquid perpetual volume crossed 50% in mid-July, touched nearly 57% in early August, then fell below 30% and has not recovered since Aug. 20. On Aug. 18, one builder venue briefly traded more than Hyperliquid’s entire validator-operated set. It has not happened again.

IOSG argues that the ratio is often misread because it is mainly telling a denominator story. The numerator is a stock-linked book, while the denominator is a crypto book, and the more volatile leg is the crypto side. The 57% print came during quiet crypto conditions. The later 26% reading came when the same stock-linked book ran into an active crypto market. During the same period, core perpetual volume rose 117%. Over the latest 7-day window, HIP-3’s share rebounded to 28.6%, even as Trade[XYZ] kept shrinking.

The report’s point is straightforward: before quoting any HIP-3 share figure, it matters to say what crypto was doing in that same period.

IOSG says the more important line is absolute volume, and that trend has deteriorated. Trade[XYZ] traded $64.60 billion over the last 30 days, down 44.2% from the previous period. Its 7-day average dropped from $5.36 billion a day in early August to $2.01 billion a day. The venue’s own book retraced 62%, and its biggest market, SK Hynix, fell to $8.50 billion. In IOSG’s reading, both legs of the share decline are real, but most of the weakness on the HIP-3 side is not a competition story.

Volume fell mostly because the theme cooled, not because share was lost one-for-one

IOSG says the easiest interpretation of a 44% decline is that the leader is losing. The data do not support that cleanly. If competition were the main cause, the underlying names would still be trading normally and Trade[XYZ] would simply be capturing a smaller slice. What actually happened, the report says, is that the names themselves became quieter.

There was no broad selloff in price terms. Using the early-August volume peak as the reference point, every major market in the book was trading at a higher price by the time of measurement. What collapsed was the distance those names traveled in a day, and venue volume fell almost in step with that move.

Because equity markets shut on weekends while Trade[XYZ] keeps trading, IOSG used a weekday-only lens for this part. On that basis, daily average volatility in the storage basket had a +0.47 correlation with daily venue volume across a 45-trading-day sample. Gold served as the natural control group: it was the only large market in the set with rising intraday volatility during the month, and its volume rose as well. Silver did not fit that pattern.

Volatility was only a proxy, though. A more direct test compared the same nine names with their real-world stock-market turnover. That exercise explained only about half the drop. Actual trading in the storage and AI capex basket fell 25.7%, and IOSG accepts that half as real sector weakness. But XYZ fell 49.7%, nearly twice as much. The extra 24 percentage points did not come from the sector itself.

The largest gaps appeared in the venue’s core names: SanDisk at -26.0 percentage points, Micron at -25.4 points, Intel at -21.5 points and SK Hynix at -17.2 points. Trade[XYZ] actually outperformed the real market in Nvidia by +36.5 points and Nebius by +18.6 points, but both of those books were small.

Competition also failed to explain most of the gap. Entropy traded $523 million of SanDisk over the 30-day window, while XYZ’s own SanDisk book was down $6.14 billion. In that market, challengers could account for only about 8% of the decline at most.

What remains looks more like a rotation. Hyperliquid’s core perpetual book grew 117% during the same period while HIP-3 fell, yet the two together still expanded 26%. IOSG’s conclusion is that money did not leave Hyperliquid. It rotated out of stock-linked books and back into crypto books.

Scale matters here too. Across the same 30-day window, the nine underlying names traded $2,004.7 billion on their home exchanges. XYZ did $23.5 billion in those same names, or 1.2% of that total. Its full 104-market book, at $64.60 billion, was only 3.2% of the real-world turnover in those nine names. Penetration was highest in SK Hynix at 9.1% and lowest in Broadcom at 0.1%. IOSG says that curve says a lot about what this business actually is: assets that crypto-native traders cannot easily access elsewhere penetrate better, while liquid U.S. mega-caps that anyone can already buy penetrate less.

So IOSG splits the recent decline in two: about half was sector beta, and about half belonged to the venue itself.

Settlement asset choice was the cleanest predictor of survival

Six venues have already stopped trading. IOSG says the variable that separated the survivors from the shut-down names was not asset selection, team quality or historical turnover. It was the settlement stablecoin.

The mechanism is simple. Traders have to convert into a specific stablecoin before placing the first trade, and many will not bother. Felix is the clearest example. The fee edge that once supported USDH was effectively erased once Growth Mode went live, leaving only friction behind.

Kinetiq provided the control case. It was the only operator that survived after shutting down once. Its fix was to kill the USDH venue and relaunch the same index products on USDC. Historical volume had no predictive power here. dreamcash once traded $19.51 billion, more than the whole June cohort put together, and still shut down. Entropy entered in August with more capital than any previous entrant and showed no hesitation in choosing USDC.

AQAv2 did not tilt fees in favor of HIP-3

IOSG says it is natural to read USDC’s sweep as a protocol design choice, but Hyperliquid’s own documentation points the other way.

Under Aligned Quote Assets v2, USDC was enabled on Hyperliquid in late August, with Coinbase as treasury deployer and Circle as technical deployer. The documentation says roughly 90% of USDC’s cost-adjusted reserve income on Hyperliquid goes to the protocol and into the Assistance Fund. Interest accrues in 30-day periods and is paid on day eight after each period closes, meaning the first payment is due in early October. Nothing has been paid yet.

AQAv2 explicitly does not favor HIP-3. The documentation says there is no preferential treatment in trading fees or volume credit, and that other quote assets remain supported on HIP-3 perpetuals. Fee benefits belong to AQAv1, which gives venue collateral assets lower taker fees, higher maker rebates and higher volume credit. USDC is not in AQAv1 and cannot fit structurally because that version requires the stablecoin to be exclusive to Hyperliquid. The actual privilege AQAv2 points toward is for event contracts and validator-operated perpetuals, and even that awaits future upgrades. It is not the market IOSG measured here.

That is why the settlement outcome was shaped by liquidity and by a company-level event, not by fee engineering, according to the report. USDH was discontinued on July 17, 2026, with holders able to redeem 1:1 into USDC. Coinbase took over its brand assets and became treasury deployer for USDC. Today, USDC accounts for 98.3% of stablecoin supply on Hyperliquid, USDT for 1.2%, and residual feUSD, USDe and USDH each sit around one-thousandth. A venue using some other settlement asset was not losing on fees. It was asking its traders to leave the only pool with depth.

IOSG also attempted to size AQAv2 without official numbers. Using $6.77 billion of USDC on Hyperliquid, SOFR at about 3.6%, and a 90% protocol share, the line points to something around $200 million a year. A third-party estimate using a $5 billion base lands in a $135 million to $160 million range. IOSG stresses that the cost-adjustment element inside the AQA rate is not public, since it comes from a validator-reported oracle, so every figure here is an estimate rather than a directly observed number.

HyENA supplied a second mechanism check. Because it listed crypto assets, it was excluded from Growth Mode. On the same names it was quoting around 5 bp while the core book beneath it was around 3 bp, resulting in worse execution quality. It spent about $0.88 million on listing slots and earned just $33,414 over its lifetime.

IOSG adds that a trading halt does not necessarily mean a complete exit. HyENA has already delisted all 25 markets and its open interest is zero, but its stake remains 508,915 HYPE, worth about $40.6 million at $79.73, and no withdrawal has been initiated over the last 12 days. Felix and dreamcash have both fully withdrawn their stakes and now read zero, while Ventuals is down to 7,967. A venue that has removed every market but still leaves roughly $40 million staked onchain is either unwinding slowly or using the deployer seat for something else, IOSG says.

Ventuals gets its own post-mortem. IOSG says the real failure was not just weak liquidity but the funding-rate mechanism itself. Pre-IPO perpetuals had no convergence anchor, and the funding rate once ran toward roughly 8,700% annualized. In that setup, longs could be liquidated regardless of whether the mark price was sensible. Entropy capped annualized funding around 10% and settled to the TWAP of its own mark rather than chasing an outside reference. IOSG reads that contract design almost as a direct checklist of fixes for how Ventuals died.

What Trade[XYZ] actually sells, and why OpenAI is not on the board

Trade[XYZ]’s top 10 markets made up 66.7% of its 30-day volume, but the long tail outside the top six still contributed $32.4 billion on its own. Nvidia was just 3.5%. Apple, Tesla, Alphabet and Microsoft together came to 3.3%, only a quarter of SK Hynix by itself.

IOSG says the standard tokenized-U.S.-stocks narrative does not describe this business very well. Trade[XYZ] is really running a 24/7 venue for trading storage and AI capex exposure, plus crude, metals and index products: Korean and Japanese semiconductors, a synthetic DRAM index, SpaceX, its in-house XYZ100 basket and a licensed S&P 500 product. Its territory is the set of assets crypto-native traders cannot touch elsewhere at 3 a.m.

That is also where Entropy chose to attack, by going after SanDisk and Nebius rather than Apple.

The report spends time on why Trade[XYZ] does not list OpenAI. The obvious answer is that it avoids private companies, but IOSG says that is wrong. Pre-IPO is actually one of the venue’s stronger businesses. SpaceX alone traded $2.80 billion over 30 days, or 4.3% of the book, ranking ninth. Below that were Unitree at $511 million, CXMT at $317 million, Zhipu at $156 million, MiniMax at $92 million, SHEIN at $27 million, with YMTC already registered and waiting.

Those names share two traits: they have observable secondary prices and known share counts. SpaceX regularly runs tender offers with a visible per-share price. The Chinese names trade in active domestic pre-IPO gray markets, and their share counts can be assembled from corporate filings and financing rounds. That lets the venue quote them on a per-share basis like other assets.

OpenAI and Anthropic do not fit that mold. Their secondary trades are often wrapped in SPVs, where what changes hands is a claim on a fund interest and the number being negotiated is an enterprise valuation, not a per-share price. Quoting them per share would mean inventing a denominator. IOSG says the bottleneck is a pricing-convention problem, not a willingness problem.

Entropy’s solution was to avoid per-share pricing entirely and quote the company itself, with one contract equal to $1 billion of market value. At current midpoint prices, Anthropic sits near $2.17 trillion and OpenAI near $1.53 trillion.

That still is not a moat, in IOSG’s view. If the leader wanted to add a market-value contract, it could do so for roughly $39,900. Just as important, its own roadmap points elsewhere. Trade[XYZ] already has 16 registered but inactive names in queue, including uranium, aluminum, the U.S. dollar index, VIX, corn, wheat, TTF, KRW, India’s Nifty, Brazil’s Ibovespa, Ibiden, KSTR, plus YMTC and H100. That is a macro and commodities roadmap, not a frontier-AI-lab roadmap.

IOSG highlights one structural detail. Trade[XYZ] does not set an oracleUpdater and instead pushes mark prices with its own deployer key. Entropy and Felix both point to the same third-party updater address, 0x94757f8d…. Entropy has publicly said RedStone is the pricing source for its Anthropic market, which could explain why two otherwise unrelated venues share an updater address, though the address is not labeled onchain. An in-house oracle is one thing when there is a clear reference price. It is something else when the mark itself has to be constructed, and constructed mark prices are exactly the business Entropy chose.

Head-to-head overlap, and the one week Entropy was ahead

There are now nine names that are live on two HIP-3 venues at the same time. Books are no longer neatly separate. If a name is worth listing twice, overlap is becoming normal.

IOSG notes that Trade[XYZ] had registered five of Paragon’s core names without activating them, then switched all five on in a single day on Aug. 18. It now leads in all five. Filling that gap cost roughly three days of fee income and took one afternoon. Whether that was deterrence or simply an onboarding queue reaching completion cannot be proven from chain data alone, the report says.

The result is easier to read. Four weeks later, Paragon still held roughly 20% to 25% share in four of those five names, and its overall monthly volume was still up 49.9%. Unitree showed the mechanism most clearly. Trade[XYZ]’s Unitree book was about 11 times the size of Paragon’s, and Paragon held only 8.2% of the pair, yet Paragon’s own Unitree volume almost doubled over the same period. IOSG’s interpretation is that the leader did not simply steal volume from the challenger. It expanded the market around the challenger. Entering a market and owning a market are not the same thing.

Entropy’s Nebius battle was more direct. For one week, Entropy’s Nebius volume actually exceeded Trade[XYZ], something no HIP-3 challenger had managed before in a market where the leader was actively quoting. The following week, Trade[XYZ]’s Nebius book rose 61% while Entropy dropped 63%, pushing the challenger back to about one-fifth of the pair. SanDisk told the same story more quietly. Entropy’s share there has hovered around 13% in recent weeks and was 8.5% over the full 30-day window.

IOSG’s reading is careful: Entropy proved it can break into a market where the leader is actively making prices, but it has not yet proved it can hold that position. Its total volume has declined for three straight weeks, from $417 million to $254 million.

The counter-signal is inventory. While weekly turnover dropped 39% from its high, Entropy’s open interest rose 37% to $51.4 million, with Anthropic alone accounting for $29.9 million. Wash-like volume tends to cancel itself out and does not leave inventory behind. So when share is slipping but open interest is still accumulating, IOSG says the signal points more toward real positioning than pure churn. The more useful object to watch is the tension between those two facts.

That volume still needs a discount. Entropy has no token and has not confirmed an airdrop, but a pointsMultiplier parameter is already visible in the backend and its own leaderboard page says “coming soon.” IOSG says that means part of the traffic may be farming expectations rather than using the product, and outside observers cannot separate the two.

Scale remains decisive. Entropy did about $1.0 billion in a month. Trade[XYZ] did $64.6 billion. The challenger is 1.6% of the leader. It won specific battles, not the category.

Its registered-but-inactive names hint at the next battle. Entropy holds EWY, SBE, TCNT and a DRAM index. The DRAM index is already Trade[XYZ]’s fourth-largest product, and EWY points straight at Korea. IOSG says the next test appears set to run directly into the leader’s core territory rather than into another uncontested pre-IPO name.

The report also lays out two counting rules. First, only live books are counted. HIP-3 deployers often register assets long before turning them on, and those markets can return an oracle markPx while midPx is null, isDelisted is true, open interest is zero and there is no candle history. Trade[XYZ] has 16 such names, mkts has 19, Paragon has 9 and Entropy has 4. HyENA’s 25 are different because they were once live and later shut down. Second, ticker strings are not the same thing as assets. para:STX is Seagate with a midpoint around 799, while core-book STX is Stacks at around 0.27. Matching by code alone would fabricate a tenth overlapping market that does not really exist. Mid-prices have to be checked first.

The economics: large costs, thin revenue and almost no room to undercut on fees

IOSG says the fee structure leaves almost no price competition. Two parameters matter at the asset level, and both are public in metaAndAssetCtxs: growthMode and deployerFeeScale. The all-in fee is base × (1 + s), where base is the standard perpetual fee schedule and s is the deployer coefficient. It can be set from 0 to 3.00 and is capped at 1.00 under Growth Mode. The deployer receives s / (1 + s), so at s = 1.00 the deployer gets half. Growth Mode then cuts the all-in number by at least 90%, provided the market does not overlap with validator-operated perpetuals, which excludes crypto assets and crypto indices.

Every stock-focused venue independently converged on the same setup: max out the deployer share and turn on Growth Mode. Even Entropy, which entered with differentiated products and venture backing, did not choose to compete by lowering price. The only venue outside that template was the one that had just stopped trading.

IOSG says deployer revenue can be measured without relying on an aggregator. Fee income accumulates in sub-accounts tied to fee-receiving addresses. clearinghouseState with the dex field shows undistributed balances, while actual payouts appear in userNonFundingLedgerUpdates as sends whose sourceDex equals the venue name. Because transfers are irregular, the correct way is to measure between transfers.

On that basis, Trade[XYZ] has accumulated $1,380,592 since its Aug. 27 payout, against $32.30 billion in volume. That implies a measured rate of 0.427 bp, or about $79,000 a day. Entropy has never made a transfer, so its cumulative fee revenue can be read directly: $1.036 billion in volume against $41,468 in fees, equal to 0.400 bp.

Two cost buckets, with very different economics

Staking is the most intimidating number, but it comes back. No one takes that principal away in the normal case. It is delegated to validators, continues earning staking rewards, and is returned intact on exit. It must remain locked for at least 183 days from deployment. Malicious market operation, such as pushing bad oracle prices, can still trigger slashing by weighted validator vote, and the position remains slashable during the 7-day unstaking queue. That makes a clean exit roughly a 190-day process. Felix and dreamcash have both already withdrawn in full and now read zero.

Current stakes are 500,973 HYPE for Entropy, 500,712 for Paragon, 508,915 for HyENA, 588,489 for Kinetiq, which covers both km and mkts, and 500,488 for Trade[XYZ] plus another 500,269 at a second address. ABCDEx has only 1,004 HYPE and never staked.

The cost that really does not come back is the listing slot. Every perpetual DEX gets the first three assets free. After that, each additional market has to be bought with HYPE in a 31-hour Dutch auction. The auction opens at twice the last clearing price and decays linearly to a floor of 500 HYPE. On Sept. 14, the auction opened at the 500-HYPE floor and also closed there, implying a slot cost of about $39,900. IOSG says demand for slots has already cooled from the 582-HYPE clearing price seen a week earlier.

The report says the key column is payback time. Trade[XYZ] needs about eight weeks of fees to cover the full bill for all its markets. Every challenger except Entropy would need longer than HIP-3 itself has existed. Entropy clears the arithmetic only because it bought seven slots rather than thirty.

Time should widen this asymmetry. Pausing a market is free and reversible, and a paid-for slot can be mothballed and reopened later. That is what many “registered but not live” names are. Reserve slots also accumulate with deployment history, using the formula 7 + 0.2 × historical auction deployments. That leaves Trade[XYZ] with about 30 immediately usable slots, while a new entrant starts with only 7. A venue launching today with plans for 20 markets could list 10 at once and then wait through auction cycles for the rest, at best one every 31 hours. Entropy’s answer was not to play that game at all. It launched only five markets and tried to make each one count.

The staking-yield trap

Apart from Entropy, IOSG says every challenger has earned more from the passive yield on its entry ticket than from running the exchange itself. Paragon is the clean example. Its lifetime deployer revenue is $64,281. But a $39.87 million stake at roughly 2.2% annualized yields about $877,000 a year, roughly 14 times as much.

That is not comfort, the report argues. Those rewards come from newly emitted HYPE in future protocol reserve emissions. In economic terms, they look more like inflation compensation than business income, and they are paid in the same asset the operator is forced to remain long. If HYPE falls 30%, the stake loses about $12.0 million, more than a decade of yield. Over the last eight days alone, HYPE fell from $88.37 to $79.73, cutting about $4.3 million from each staked position.

How big the fee pool really is

Growth Mode pins measured rates near 0.4 bp, with deployers taking half. With a 97.8% share and annualized volume of about $786 billion, Trade[XYZ] is on pace for about $29 million a year in deployer revenue. IOSG says this is less a leader’s ceiling than an estimate of the total venue-level prize available under today’s volume and today’s fee floor.

Outside the leader, venue operators collectively have about $167 million in HYPE tied up today. Against that, all of them, alive or dead, have earned only $747,000 in deployer revenue over their whole histories. Set beside a $14 million seed round, IOSG says the arithmetic is clear: under current volume and current fee floors, HIP-3 operators cannot justify a business valuation through trading fees alone. Whatever value challengers have must come from somewhere else, whether that is a token, a frontend, customer relationships or some product layer the protocol has not priced yet.

What can actually be defended

IOSG argues that HIP-3 has intentionally commoditized most of what a venue might otherwise defend. Staking can be bought. Listings can be bought. The fee floor is shared. Distribution is shared too, because every HIP-3 book can be reached from the same frontend.

That leaves a different filter question. Not “which assets will you list,” because assets can be bought. The better question is: what do you have that the leader cannot buy for the price of one listing slot? Across every team that has operated a HIP-3 venue, IOSG says only one has had a clear answer, and that answer was an oracle stack and a settlement design, not a watchlist of assets.

Entropy is the exception for reasons that go beyond its market names. The company was founded by researchers and traders from Citadel Securities, Optiver, Millennium and Polymarket. IOSG says that bench showed up in two places: the depth available from day one, which was what mattered in Nebius, and the funding-rate and settlement design, which reads like a direct response to the way Ventuals failed. Its $14 million seed round was led by Ribbit Capital, whose home field is retail brokerage and fintech distribution rather than DeFi. In IOSG’s framing, that points toward an ambition to own customers rather than merely extract protocol incentives.

The report also warns against name confusion. Entropy Advisors, which is closely tied to Arbitrum DAO, and a custody startup named Entropy backed by a16z are completely separate companies from the Hyperliquid-linked venue discussed here. IOSG says no relationship with Hyper Foundation should be inferred from the shared name.

Put together, the lane still looks thin. A shrinking operator controls 97.8% of volume. A ring of challengers has staked about $167 million in HYPE and received only $747,000 in historical fees combined. The one team that appears genuinely able to price markets proved it could win a market for a week, then failed to keep it.

IOSG adds that a simpler alternative may be to take the protocol’s half of fees rather than the operator’s half. That route comes without lockups, slashing risk or operating burden. Even then, it is not the main body of Hyperliquid’s fee base, and the data already show that HIP-3 clearing 50% share was more a quiet-crypto illusion than a durable trend.

The routing layer: the closest thing to a distribution business

Builder codes are, in IOSG’s view, the layer that comes closest to a distribution business on HIP-3. Frontends tag their own orders and take a builder fee, all without posting the large capital required to operate a venue. Flowscan counts 819 such codes.

The volume they touch is small. Routed volume totals about $52.6 billion, roughly 9% of HIP-3’s historical $587 billion. Over the latest 30 days, routed volume was about $5.3 billion against total HIP-3 volume of $66.09 billion, or around 8%. More than 90% of all flow carries no frontend tag at all, which IOSG says is exactly what one would expect in a market dominated by market makers and API traders.

The routing leaderboard also needs careful denominators. Its relevant base is the $5.30 billion routed by builder codes over the last 30 days, not the full $66.09 billion of HIP-3 volume. Against the larger number, even the biggest entry, CoinDCX, would be only 0.7%. The top 10 codes in the table account for 59.7% of routed flow, while Flowscan counts 819 builder codes in total, leaving the other 809 to split the remaining 40%.

Two entries stand out. Entropy’s $423 million of routed flow all occurred in the last 30 days, while its own venue traded $1.03 billion during the same period. That implies roughly 40% of its book came through a frontend it controls. IOSG says the retail-brokerage-style distribution ambition associated with the Ribbit round shows up in the data, not just in narrative, and that this is a different business from being a deployer.

The second example is dreamcash, which supplies a cleaner lesson. Its venue has been dead since July 2 and its own book now reads zero, yet its builder code still routed $17.3 million over the last 30 days and $3.54 billion in total. Venue operation and frontend distribution can be separated cleanly, and only one of those businesses requires about $40 million to enter.

IOSG also flags a common dashboard trap. Venue rankings are often sorted by lifetime volume by default, which means a venue that shut down months ago can still appear to have meaningful share. dreamcash shows 3.3% of HIP-3 volume in an all-time view and $0 in any recent window. Before citing any share number, the report says, it is necessary to verify the window behind it.

That is why the frontends worth watching may not be deployers at all. Coinbase announced a simplified perpetuals interface inside its wallet on Sept. 12, powered by Hyperliquid and covering crypto, tokenized stocks and prediction markets for users outside the United States. Kraken’s parent company is separately in talks to bring Hyperliquid-related perpetuals into a regulated U.S. venue. Neither company is likely to stake 500,000 HYPE.

Bottom line: IOSG is not optimistic about adding another HIP-3 deployer

Put together, the numbers are stark. Excluding the leader, all HIP-3 volume over the last 30 days totaled just $1.49 billion, or about $18.1 billion annualized. At the observed 0.400 bp rate, that is about $725,000 in annual deployer revenue to be split across four venues. Those four have about $167 million of HYPE staked today. The same capital, passively staked at 2.2%, would earn roughly $3.67 million a year.

In other words, running these exchanges earns about one-fifth of what passive staking of the same capital would generate.

IOSG says this is not an undeveloped market. It is a market already priced close to zero, for reasons that can all be measured. Listing slots cost about $39,900 each, so assets are not defensible. The fee floor is shared, so there is no room to cut price. Distribution is shared, and builder codes only touch 9% of flow. Settlement assets have already converged six to six in favor of USDC, and AQAv2 explicitly gives HIP-3 no fee tilt, so the protocol is not trying to subsidize this layer.

The harder part is the ceiling. Even with 97.8% share and every structural advantage, XYZ reached only 1.2% of real-world trading in the nine names where it actively quotes, and that relative share is still slipping in its core storage names. A new entrant is not facing “a giant incumbent.” It is facing a leader that is already small against the underlying market and still shrinking.

IOSG says there is only one real exception: a team that controls a right others cannot buy, an oracle others cannot build, or a funding and settlement design that can survive in thin books. Entropy is the only example on the board, and even it won a market for one week before getting pushed back. An asset list is not the answer. Two separate sections of the report point to the same conclusion.

What could change the view

• Entropy holds a contested market for a month rather than a week, and stabilizes its share in SanDisk rather than drifting around 13%. IOSG says the most important signal is still open interest rising while volume share falls.
• Entropy launches its DRAM index and Korea-linked names. Those strike directly at the leader’s core territory and would provide a cleaner test than another uncontested pre-IPO name.
• Token terms eventually make the equity story work. The fee story does not work today, and no token exists yet.
• The pre-IPO category proves durable. Anthropic faded after its first week, and OpenAI has fallen from $5.3 million on opening day to around $4 million. If even one of those books can stabilize, the category may be real rather than a launch-week trade.
• The first AQAv2 payment arrives in early October. That will offer the first observable read on the actual size of the protocol’s USDC income line, against today’s third-party estimate range of $135 million to $200 million.
• The fee floor loosens. Hyperliquid signaled in early August that a future upgrade may allow HIP-3 deployers to raise fees by as much as 3x on individual assets, effectively clawing back some Growth Mode discount. There is no timeline yet, and IOSG’s fee analysis is built on the current floor.
• Permissioned markets become real. HIP-3*, announced on Sept. 3, is an optional onchain whitelist that lets deployers limit which wallets can trade a market. It is framed around compliance and institutional access and is still on testnet. IOSG says this is the first mechanism that could make access rights rather than assets the scarce good.
• A team appears with exclusive data or index rights tied to assets that really do have 24/7 crypto-native demand. That is the one configuration in which slot-grabbing logic could truly fail.
• Evidence emerges that the leader’s volume cannot survive the end of Growth Mode. Its measured 0.427 bp is only about one-tenth of what the same book would collect under the standard fee schedule. If that exemption is load-bearing, the venue is less secure than market share alone suggests.
• HyENA unstakes its 508,915 HYPE. That would confirm the venue is finished rather than dormant.

Limits noted in the report

• The “slot-grabbing” interpretation is grounded in measured market-share changes, but motive remains inference. Five names launching on the same day could also mean a deployment pipeline happened to complete that day.
• The Entropy analysis is based on 26 days of data. Nebius leadership and the subsequent reversal are both single-week observations in a single mid-sized market.
• Team backgrounds and fundraising history come from company statements and media coverage, not from onchain verification.
• The volatility result in section two is correlation over a 45-working-day sample, not causal decomposition. Volume and realized volatility may both be reacting to the same factor, with cooling AI capex trading the most obvious candidate.
• Trade count is not trader count. Public APIs do not reveal unique users.
• Capital-return figures are based on HYPE at $79.73 and a 2.2% annualized staking rate. IOSG notes that the latter is a protocol parameter, not a contractual promise, and would decline if network-wide staking rises.

The report was published under IOSG Ventures and written by Mario Chow.