Perp DEXs in 2026: Who Takes the Other Side of Your Trade

Bitcoin tore through the first week of September 2026 and pulled the entire leverage complex up with it. Few tokens moved harder than HYPE, the asset behind the largest on-chain perpetuals venue, which changed hands near $85.89 on 6 September according to CoinGecko. Decentralized perpetual exchanges, or perp DEXs, have been setting volume records for the better part of a year, and the September move, the same jobs-driven surge we covered in our September countdown, sent traders piling back into leverage.

The boom conceals a question almost nobody stops to ask when they click “long” with 20x size: who is actually on the other side of that bet?

On a centralized exchange the answer is dull and comforting. A company runs the order book, matches you against another customer, keeps an insurance fund on its own balance sheet, and exists as a legal entity you could, in principle, sue. On a perp DEX none of that is true. There is a smart contract, a set of validators or a sequencer, and a pool of other people’s capital. The house is still there. It just has no front desk, no phone number, and in some cases no discretion left to exercise once the code is deployed.

This article is about that house: who takes the other side of an on-chain perp, how it earns, how it loses, and what happens on the days it nearly breaks. We mapped the plumbing (order books, oracle pools, hybrids) in an earlier explainer on the on-chain futures boom. Here we follow the money and the risk instead of the architecture.

On a CEX, the house has a name

Start with what a perp DEX is replacing. When you trade a perpetual future on a big centralized exchange, three things sit behind the screen. First, a matching engine pairs your order with another trader’s. Second, when a position goes so far underwater that the trader’s margin cannot cover it, an insurance fund absorbs the shortfall so the winning side still gets paid. Third, when even the insurance fund is not enough, the exchange reaches for auto-deleveraging, forcibly closing profitable traders to keep the book solvent.

All three are backstopped by a company with a balance sheet, shareholders, and a compliance department. If the insurance fund is drained, the firm can top it up. If something goes wrong, there is an entity to complain to, a jurisdiction, and sometimes a court. You are trusting a counterparty you can name.

A perp DEX keeps the same three jobs (matching, backstopping, deleveraging) but hands them to code and to strangers. Nobody can top up the fund from a corporate treasury. There is no compliance desk to freeze a suspicious account, at least not in the way a bank would. And because these venues are non-custodial, your collateral sits in a contract you approved rather than in a company’s omnibus wallet. The trade-off that makes them censorship-resistant is the same one that leaves you without anyone to sue. That shift, from custodial accounts to keys and contracts, is the same one reshaping wallets, as we explored in our piece on the exchange becoming your wallet.

Three answers to who is my counterparty

Perp DEXs solve the counterparty problem in three broad ways, and each puts a different entity in the dealer’s chair.

The first is the on-chain order book. Venues like Hyperliquid and dYdX match traders against each other, exactly like a centralized exchange, so your everyday counterparty is another human (or bot) taking the opposite view. dYdX runs its order book off-chain, in memory across its validators, committing only matched trades to the chain, a design its own team has framed as a deliberate trade-off between speed and decentralization. What sits behind the book is a backstop vault or insurance fund that becomes the counterparty of last resort when a liquidation cannot be filled.

The second is the peer-to-pool model. There is no order book at all; traders open positions directly against a shared liquidity pool at a price fed by an oracle. GMX (with its GLP and newer GM pools) and Jupiter (with JLP) are the archetypes. Here the pool is unambiguously the house: every long and short is a bet against the depositors’ capital.

The third is the hybrid: matching happens off-chain for speed, settlement happens on-chain for custody. Aster, Lighter, Vertex and others fit here. Your counterparty is still another trader, but the sequencer or matching layer is a trust point that the other two designs try to remove.

HLP, the house you can buy a seat in

Hyperliquid is the clearest place to watch the house work, because it publishes it. The Hyperliquidity Provider vault, universally shortened to HLP, is a protocol-owned pool that anyone can deposit USDC into. HLP does three jobs, per CoinGecko’s breakdown: it market-makes across the venue’s hundred-plus perpetual listings, it captures funding, and it acts as the backstop that inherits positions when liquidations cannot be cleared in the open market. Crucially, HLP is not the counterparty to your average trade (another trader usually is); it is the market maker and the liquidator of last resort.

The economics are unusually transparent. HLP charges no performance fee; profits flow to depositors in proportion to their share. And those profits are lumpy rather than smooth. HLP’s assets peaked near $603.9 million in September 2025 and had fallen roughly 55% to about $268.6 million by June 2026, yet its cumulative profit since launching in May 2023 reached about $136.9 million. Two days did much of the work: the flash crash of 10 October 2025 handed HLP an estimated $40 million to $41.5 million (close to a 10% gain in under 48 hours), and a liquidation cascade on 31 January 2026 added around $15 million. Those two events alone account for roughly 41% of the vault’s all-time profit.

Read that again, because it is the whole thesis of the vault business: the house’s biggest paydays come from chaos. When leverage gets flushed out of the market, the depositors on the other side collect. The rest of the time they are quietly warehousing risk. And they cannot always leave in a hurry: vault capital is subject to lockups, so an LP’s money tends to be least accessible in exactly the turmoil that might make them want it back.

When the pool is the counterparty: JLP and GM

On Solana, Jupiter makes the house even more explicit. Its perpetuals run entirely peer-to-pool: traders open positions against the JLP pool, and JLP holders take the other side of every single trade. When traders lose, JLP holders profit; when traders win, JLP holders eat the loss. The pool holds a basket of USDC, SOL, ETH, BTC and USDT (stablecoins are roughly 35% of it), has carried a total value well above $1 billion, and pays holders 75% of all platform fees, with the remaining 25% routed to the protocol, according to a breakdown of Jupiter Perps.

GMX pioneered this structure on Arbitrum and Avalanche. Its first-generation GLP pool was a single multi-asset basket that served as counterparty to every trade; its v2 redesign split liquidity into isolated GM pools so that a blow-up in one market cannot drain the others. For depositors that changed the risk calculus: rather than backing the entire basket, an LP can choose to stand behind only the markets they understand and sit out the long tail. Chainlink oracles feed the prices. The pitch to depositors is the same everywhere: park assets, collect fees, earn a yield that looks like a bond coupon.

It is not a bond coupon. A perp-pool deposit is a leveraged short position on the aggregate profit and loss of everyone trading against you, dressed up as yield. Which brings us to the uncomfortable accounting question.

Is it yield, or the other side of a bet?

The honest way to describe a perp-vault return is: fees, plus funding captured, plus (or minus) the net profit and loss of the traders on the other side. In a market where retail is structurally long and structurally loses, that nets out positive often enough to advertise double-digit annual returns; JLP has historically shown figures ranging from around 30% to 80% a year, denominated in the underlying basket rather than a farm token.

But calling it “yield” flatters it. Real yield, as we have argued in a framework that treats every yield as a spread over T-bills, is a claim on genuine cash flow: fees a protocol actually earns from users. A perp vault earns real fees, yes, but a large slice of the headline number is directional; it is the traders’ losses, which vanish (and reverse) the moment the crowd is right. That is closer to selling volatility than to clipping a coupon. It behaves nothing like the protocol issuance plus fees you get from staking a proof-of-stake asset, the kind of spread we compared across Lido, Rocket Pool and Frax. Staking pays you for securing a network; a perp vault pays you for being willing to lose money on the days the traders win.

How funding keeps both sides paying

None of this works without funding, the small periodic payment that tethers a perpetual to the spot price it is supposed to track. When more traders are long than short, longs pay shorts; when the crowd flips, the shorts pay. That is what stops a contract with no expiry from drifting away from the asset it references, and it is a second, quieter channel through which money moves between you and the house.

In the order-book model, funding flows between traders, and the backstop vault collects it only on the positions it happens to be holding. In the peer-to-pool model the pool is always on one side, so funding (often charged as an hourly borrow fee on the trader’s position) becomes a steady toll paid to depositors for renting out the pool’s liquidity. On-chain venues have tended to run funding hotter than centralized ones, because their traders skew retail and long-biased and will pay up for leverage. For the house, that persistent long bias is the edge: most of the time, the crowd is paying to hold the very position the vault is short.

This is why a perp-vault return can look attractive even in a flat market. Depositors are not only waiting for traders to be wrong about direction; they are collecting rent the whole time the book is lopsided. It is when price finally moves hard against that lopsided crowd, the same days the vault books its largest mark-to-market gains, that funding briefly flips and the toll runs the other way.

The JELLY squeeze: the day the house rewrote the rules

The single best illustration of what “no front desk” really means came on 26 March 2025, and it involved a memecoin called JELLYJELLY.

A trader opened a large leveraged short on JELLY on Hyperliquid, then went into the illiquid spot market and pumped the token, reportedly by more than 400%. The short was liquidated, and under the venue’s rules the position was handed to the HLP vault, which suddenly found itself involuntarily long a rocketing memecoin it never chose to hold. Unrealized losses on the vault briefly reached $12 million to $13.5 million, per CoinDesk.

What happened next is the part worth remembering. Hyperliquid’s validators convened and voted, within minutes, to delist the JELLY contract and force-settle every position at $0.0095, a price far below where JELLY was trading. That nullified the attacker’s paper profit and left HLP with a small gain of roughly $703,000 instead of a large loss. The attacker had already pulled about $6.26 million off the exchange before withdrawals were frozen.

The vault was rescued. But the rescue is exactly what unsettled people: a small set of validators changed the settlement price of a live market by fiat to protect the protocol’s own pool. It was, depending on your point of view, either responsible risk management or proof that the house can rewrite the rules mid-hand when it is losing. Either way, it answered the counterparty question with brutal clarity. On this venue, when the backstop is in danger, the backstop can vote.

The episode reshaped how Hyperliquid manages the vault. It tightened position limits, revisited how thinly traded assets can be listed, and leaned harder on the principle that HLP should never be forced to warehouse a market it cannot hedge. Rival venues took notes too, because the JELLY playbook (open a position, then shove the illiquid spot market that prices it) is the most reliable way to attack any perp DEX where a vault is the backstop of last resort.

When the oracle is the house’s blind spot

The pool model has a structural weakness the order-book model does not: its price is imported. Because trades execute at an oracle price with no slippage, anyone who can move the oracle can extract value from the pool.

GMX learned this early. On 18 September 2022 a trader used GMX’s zero-slippage execution to manipulate the price of AVAX, which fed GMX’s oracle, extracting roughly $565,000 from the GLP pool and prompting GMX to impose open-interest caps, as CoinDesk reported. The episode was less a hack than a demonstration. Joshua Lim, then head of derivatives at Genesis Trading, described it as GMX simply “working as designed” (Cointelegraph): the pool promised to trade at the oracle price, so it did, even when the oracle was being gamed. The lesson stuck; every serious peer-to-pool venue now caps open interest, throttles low-liquidity assets, and leans on sturdier oracle designs. But the dependency never goes away. If you are the pool, the oracle is your eyes, and a blind spot in the oracle is a blind spot in your balance sheet.

The modern fix is defense in depth: more than one independent price feed, time-weighted averages that are expensive to move, per-block sanity checks, and hard caps on how much size any one thin market can carry. Jupiter, for one, prices its pool from more than a single oracle source. None of this removes the dependency; it just raises the cost of moving the price faster than the pool can react. For a depositor, the quality of a venue’s oracle stack is not a technical footnote; it is the difference between a vault that survives a manipulation attempt and one that funds it.

When the code is the house’s blind spot

The other way the house bleeds is through its own software. On 9 July 2025 GMX suffered a far larger loss: about $42 million drained from a v1 market through a reentrancy bug in the executeDecreaseOrder function, which let an attacker desynchronize GLP’s average-price accounting and conjure value out of thin air, according to a post-mortem by security firm Halborn. Most of the funds were ultimately returned in exchange for a white-hat bounty, and GMX arranged compensation for affected GLP holders, but the incident was a reminder that a liquidity pool is only as safe as the least-audited line of code with access to it.

This is why offensive security has become a core cost center for these protocols rather than an afterthought. The best teams now pay firms to attack them before criminals do, the discipline we profiled in our look at why Halborn breaks in first. For a depositor, the takeaway is blunt: when you buy into a perp vault, you are underwriting two very different risks at once, the market risk of the traders being right, and the technical risk of the contract being wrong. The advertised return compensates you for both, whether you priced them or not.

Auto-deleveraging: when the house makes you the counterparty

There is a third way the house protects itself, and it can reach into your account even if you did nothing wrong. Auto-deleveraging is the mechanism of last resort: when a liquidated position cannot be closed and the backstop cannot (or should not) absorb it, the protocol forcibly closes the most profitable, highest-leverage traders on the opposite side to keep the system solvent.

On 9 April 2026 a trader tried to manipulate the memecoin FARTCOIN on Hyperliquid with a long position worth around $145 million; the price crashed roughly 50% in a single hourly candle, and the venue’s deleveraging engine forcibly closed short positions on the other side, handing two wallets a combined profit near $849,000 at zero fee, as CoinDesk documented. Those traders were winning; the system closed them anyway, because socializing the shortfall was the only way to keep the book whole. Auto-deleveraging is the on-chain answer to the question a CEX solves with a corporate balance sheet: if the loser cannot pay, someone has to, and on a perp DEX that someone is whichever winner the algorithm can reach.

Centralized exchanges run the same last-resort tool, but they reach for a corporate insurance fund first and deleverage winners only when that fund is exhausted. A perp DEX with a thin backstop has less room before it must socialize a shortfall onto the traders who read the market correctly. That is the quiet cost of a smaller house: the less capital standing behind the book, the more often the winners themselves end up covering the losers.

The house is enormous, and it is tiny

The strange thing about the biggest house in on-chain derivatives is how few people run it. Hyperliquid now carries eleven-figure open interest, and its markets in real-world assets (equities, commodities and indices listed through its permissionless HIP-3 framework) have grown so fast that real-world-asset open interest set records near $3.6 billion and overtook Bitcoin as the venue’s single largest market, per CoinDesk. Builder-deployed HIP-3 markets now drive roughly half the venue’s volume, though a single deployer, Trade.xyz, accounts for more than 90% of HIP-3 open interest.

The scale has caught the attention of the people who run the incumbent exchanges. Jeffrey Sprecher, founder and chief executive of Intercontinental Exchange (the owner of the New York Stock Exchange), told a Bernstein conference in May 2026 that Hyperliquid is “bigger than NASDAQ” and marveled that “it’s 11 people,” adding that he had met its founders (CoinDesk). A house that clears eleven-figure open interest with a team you could fit around one table is a new kind of financial institution, and a new kind of concentration risk.

That concentration cuts into the token story too. Because HIP-3 deployers keep half of their markets’ fees, Hyperliquid’s own gross revenue has fallen every quarter from a peak near $357 million in the third quarter of 2025 to roughly $202 million by the second quarter of 2026, thinning the cash flow that funds HYPE buybacks even as volumes hit records. The house is busier than ever; it just keeps less of the take.

That tiny-team, community-vault structure is the flip side of the JELLY story. The same small set of validators that can rescue the house in a five-minute emergency vote is the set that writes the rules the rest of the time, which is why critics keep asking how many independent parties it would really take to move a market. Decentralization on a diagram and decentralization under fire are not always the same thing.

The volume mirage, and whose numbers to trust

Knowing who the house is also means knowing whose reported numbers to trust. Perp DEX monthly volume crossed $1 trillion for the first time in 2025, and much of that surge came from Aster, a BNB Chain venue endorsed by Binance founder Changpeng Zhao and backed by YZi Labs, which briefly out-printed Hyperliquid on headline volume, according to reporting on DefiLlama data. But a chunk of Aster’s volume runs on aggressive trading incentives and is partly self-reported, and on stickier measures such as open interest (capital actually committed rather than churned), Hyperliquid has stayed well ahead. Aster’s token, ASTER, traded near $0.78 in early September per CoinGecko, riding the same rally as the rest of the sector.

The distinction matters for the counterparty question. Incentive-driven volume tells you nothing about how much real capital is sitting in the venue’s vault to backstop your trade. When you size up a perp DEX, open interest and vault balances describe the size of the house; raw volume describes how loudly it is advertising.

This is where public dashboards earn their keep. Trackers like DefiLlama publish open interest, fees and pool balances next to volume, and the gap between a venue’s volume rank and its open-interest rank is often the tell. A protocol paying users to trade can top the volume table for a month; it cannot fake the capital sitting in its vault, because that capital has to actually be there to pay out a winning trade. Follow the balances, not the headlines.

The regulated house arrives, and the CME is furious

For US traders, the most important development of 2026 is that a named, regulated counterparty is coming back to perpetuals, and it lives at the Commodity Futures Trading Commission, not the Securities and Exchange Commission. Perpetual futures are derivatives, which puts them under the CFTC; the SEC’s interest is largely confined to whether a venue’s token is a security.

On 28 May 2026 the CFTC approved KalshiEX’s BTCPERP, a cash-settled bitcoin perpetual that trades around the clock with no expiry, the first true perpetual listed on a registered US contract market. Chairman Michael S. Selig called it a “watershed moment,” noting the agency had “not approved a new type of derivative in over a decade” (CoinDesk). Kalshi has since asked to list perpetuals on a 500-stock US index and on copper, and Coinbase has moved to offer domestic stock perpetuals under a security-futures framework.

The incumbents are not amused. CME Group sued the CFTC in June 2026, arguing that a contract settled through recurring funding payments is a “swap” under Dodd-Frank rather than a “future,” and that the agency changed course without proper process. On 3 September 2026 the CFTC asked the court to throw the case out, dismissing it as “much ado about nothing” and noting that nothing stops CME from listing the same contracts itself (CoinDesk). The fight is really about who gets to be the house for regulated perpetuals, and the funding mechanism, the very thing that makes a perp a perp, is now the crux of a federal lawsuit.

The contrast with the DEX world is stark. On a regulated venue, a clearing member and an exchange stand behind your position, and the CFTC stands behind them. On a permissionless, no-KYC perp DEX, the house is a smart contract and a vault of strangers, and if it fails you, there is no regulator to call. Same product, entirely different answer to who takes the other side.

For an American trader the practical map is now three-tiered: fully regulated venues such as the Kalshi contract, where a US entity is accountable and a dispute has somewhere to go; offshore centralized exchanges that ring-fence US users behind compliance walls; and permissionless perp DEXs that ask no questions and offer no protections. Each tier trades a measure of recourse for a measure of access, and the vault-backed DEX sits at the far access-over-recourse end of that spectrum.

What to check before you deposit or click long

Whichever side of the table you sit on, the counterparty question has practical consequences. If you are the trader, read the settlement and liquidation rules before you size up:

• Who can force-settle or delist a market, and at what price, as the validators did with JELLY?
• Which oracle prices your position, and what are the open-interest caps?
• Does the venue use auto-deleveraging, and where would you sit in the queue if it fired?
• How large is the backstop vault relative to open interest?

On a perp DEX the fine print is not a customer agreement; it is code and governance, and it can be exercised against you in minutes.

If you are the depositor, the liquidity provider, be honest with yourself about what you are buying. A perp-vault deposit is not a savings account; it is a leveraged, undiversified bet that the traders across from you will keep losing, wrapped around live smart-contract and oracle risk. There is no deposit insurance, no company to reimburse you, and no court that will care if the code does exactly what it was written to do. The yield is real, and so is the reason it is being paid.

The perp DEX did something genuinely new: it turned the house from a private company into an open, on-chain vault that anyone can join or trade against. That is a real democratization of a business that used to be a licence to print money for a select few. It is also a reminder that in a market with no front desk, the house is still the house, and now it might be you.

Frequently Asked Questions

Who is the counterparty when I trade on a perp DEX?

It depends on the design. On order-book venues like Hyperliquid and dYdX your counterparty is usually another trader, backstopped by a protocol vault or insurance fund. On peer-to-pool venues like GMX and Jupiter, a shared liquidity pool is the direct counterparty to every trade, so its depositors profit when traders lose and lose when traders win.

What is Hyperliquid’s HLP vault?

HLP is a protocol-owned vault that anyone can deposit USDC into. It market-makes, captures funding, and backstops liquidations that cannot be filled in the open market. It charges no performance fee, so profits and losses pass straight to depositors, and they cluster around volatile events like the October 2025 flash crash.

Is providing liquidity to a perp DEX pool really yield?

Only partly. The return is fees plus funding plus the net losses of traders on the other side. The fee portion is genuine cash flow, but the large directional portion is a short-volatility bet that reverses when traders win, so it behaves more like selling insurance than clipping a bond coupon.

Are perp DEXs regulated in the United States?

Perpetual futures fall under the CFTC, not the SEC. Regulated venues such as Kalshi and Coinbase are bringing perpetuals onshore, with Kalshi’s BTCPERP approved in May 2026. Permissionless, non-custodial perp DEXs generally sit outside that perimeter, so US retail using them has little recourse if something goes wrong.

What was the Hyperliquid JELLY incident?

In March 2025 a trader manipulated the memecoin JELLYJELLY to squeeze a short that had been passed to the HLP vault, pushing it to double-digit-million unrealized losses. Hyperliquid’s validators voted to delist the market and force-settle it at a favorable price, rescuing the vault but raising questions about how decentralized the venue really is.

Liam Brennan is a markets correspondent at HOGE Wire, covering crypto derivatives and on-chain trading.