How Perp DEXs Work: On-Chain Perpetual Futures in 2026

Perpetual futures are the most heavily traded product in all of crypto. They let a trader bet on the price of Bitcoin, Ethereum or almost any token with leverage, hold the position for as long as the margin lasts, and never worry about an expiry date. For years that activity lived almost entirely on centralized exchanges. That is changing. On-chain venues now settle a meaningful slice of the global perpetuals market, and in 2026 that slice crossed into double digits for the first time, reaching about 12 percent of all perpetual trading volume according to data compiled by Castle Labs.

Perpetual decentralized exchanges, or perp DEXs, are the venues doing that work. They run the matching, pricing, funding and liquidation logic that a centralized derivatives desk would normally keep behind closed doors, except here most of it sits on a public blockchain and your collateral stays in a wallet you control. This guide explains what a perp DEX actually is, the three very different ways engineers build one, how funding and liquidations keep the whole machine honest, who the biggest venues are in 2026, where the money and the risk sit, and how regulators in the United States and Europe now treat the product.

What a perp DEX actually is

A perpetual future, or perp, is a derivative contract that tracks the price of an underlying asset without ever settling or expiring. A trader posts collateral (usually a stablecoin such as USDC), chooses a direction and a leverage multiple, and the contract gains or loses value as the reference price moves. A perp DEX offers those contracts through smart contracts and public infrastructure rather than through a company that holds your account balance. Two differences from a centralized venue matter most.

The first is custody. On a centralized exchange you deposit funds and the exchange records your balance in its own ledger; you are trusting it to stay solvent and to let you withdraw. On most perp DEXs your assets stay in your own wallet, and the protocol only has permission to touch the margin you have posted to a position. That design is a direct reaction to the collapse of FTX in late 2022, which wiped out customers who had treated an exchange balance as if it were their own money. The second difference is transparency. Open positions, funding payments and liquidations are written to a public ledger, so anyone can audit the exchange’s exposure in real time rather than taking a monthly attestation on faith.

The trade-off is legal rather than technical. Because perps are leveraged derivatives, venues that offer them without a license tend to block users in regulated markets. That is why United States residents usually hit a geofence when they try to connect a wallet to the largest offshore venues. The table below sums up the practical contrasts.

A five-minute primer on perpetual futures

Before comparing venues it helps to be clear on the instrument itself, because every design decision downstream exists to serve it. A traditional future has a settlement date: you agree today to buy or sell something at a set price on, say, the last Friday of the quarter. A perpetual has no such date. You can hold it indefinitely, which makes it feel like trading spot with borrowed money, but it creates a problem. With no expiry to force the contract price back in line with the real market, what stops a perp from drifting away from the asset it is supposed to track?

The answer is the funding rate, the single most important mechanic in the whole product. At regular intervals, commonly every hour or every eight hours, longs and shorts exchange a small payment with each other. When the perpetual trades above the spot price, funding turns positive and longs pay shorts, which discourages new longs and nudges the price back down. When the perp trades below spot, shorts pay longs. Nobody pays the exchange; the money flows between the two sides of the market, and arbitrage traders step in to capture the gap. That gentle, constant pressure is what tethers a contract with no expiry to the asset it mirrors.

Two prices run side by side on any perp venue. The index price is a reference spot price, usually an average drawn from several large markets through a price oracle. The mark price is the value used to calculate unrealized profit, loss and liquidation; it is anchored to the index so that a brief, thin-order-book spike cannot wrongly liquidate healthy positions. Leverage sits on top: posting 100 dollars of margin at 10x controls 1,000 dollars of notional exposure, and offshore venues often allow 50x or more. Finally, pay attention to open interest, the total value of all positions currently open. Because volume can be inflated by wash trading and incentive programs, open interest is usually the cleaner gauge of how much real risk a venue is carrying.

Why put perps on-chain at all?

If centralized exchanges already match orders faster and deeper, why rebuild the machine on a blockchain? The honest answer is that early on-chain versions were slow, clunky and thinly traded, and most volume stayed where the user experience was good. The pitch for perp DEXs is that they can keep self-custody and transparency while closing the performance gap. Hyperliquid co-founder Jeff Yan has described the project in exactly those terms, telling Fortune that “centralized exchanges had a really great UX, and almost all the volume was happening on centralized exchanges, but no one in DeFi was, I think, really trying to match that.”

Beyond user experience, four arguments recur. Self-custody means a venue cannot quietly lend out or lose your collateral. Transparency means solvency is observable rather than asserted. Censorship resistance means an account cannot be frozen on a whim, though as we will see that cuts both ways. And composability means a perp position can plug into the rest of on-chain finance, serving as collateral elsewhere or being wrapped into automated strategies. Yan, whose team famously launched without venture funding or an insider token allocation, has argued that credible neutrality is the point, saying the goal is to build “a credibly neutral platform on which everyone else can build” where a “really important principle is to sort of not have insiders.”

None of this is free. Putting an order book or a pricing engine on a public ledger means contending with block times, validator incentives and the ever-present risk that a clever attacker reads the same state you do. The history of the sector is largely a story of engineers trading those constraints against one another, which is why there is no single architecture but three.

Three ways to build a perp DEX

Almost every perp DEX in use today is a variation on one of three engines. The choice shapes everything else: how prices are discovered, who stands on the other side of your trade, how deep the liquidity feels, and what breaks under stress.

The first is the central limit order book, or CLOB, the same model a stock exchange or a centralized crypto venue uses. Buyers and sellers post bids and offers, and a matching engine pairs them. Price discovery is endogenous, meaning it emerges from the order flow itself rather than being imported from outside. The second is the oracle-plus-pool model, sometimes called peer-to-pool. There is no order book; traders transact against a shared pot of capital supplied by liquidity providers, and every fill happens at a price delivered by an external oracle. The third is the virtual automated market maker, or vAMM, which prices trades along a mathematical bonding curve using virtual reserves rather than real ones. The table lays out how they compare.

The vAMM came first and mattered historically; Perpetual Protocol pioneered it and Drift used a version of it on Solana. But virtual reserves are prone to slippage on size and can drift from the index without strong arbitrage, and by 2026 the design has largely been absorbed into hybrids or retired. The real contest today is between order books and pools.

The order-book model up close

The order-book venues are trying to feel exactly like a centralized exchange while keeping the ledger public. Hyperliquid is the standard-bearer. It does not run on Ethereum or an existing layer 2; it runs its own purpose-built blockchain. Its core component, HyperCore, holds, in the project’s own words, “fully onchain perpetual futures and spot order books,” matched by a custom consensus algorithm called HyperBFT that is tuned for the latency a trading venue needs. In February 2025 the team added HyperEVM, a general-purpose smart-contract layer built into the same chain, so developers can write applications that tap directly into the exchange’s native liquidity rather than bolting on a separate network.

dYdX took a different route to the same goal. Its earlier version ran on StarkEx, a zero-knowledge rollup on Ethereum, with a matching engine the company operated centrally. In October 2023 it migrated to dYdX Chain, a standalone proof-of-stake blockchain built with the Cosmos SDK. On that chain each validator runs an in-memory order book off-chain, orders and cancellations propagate across the network, and only matched trades are committed on-chain. That hybrid of off-chain matching and on-chain settlement is a common compromise, because a fully on-chain book is hard to make fast. Other order-book venues such as Aevo, Paradex on Starknet and the zero-knowledge newcomer Lighter chase the same balance of speed, decentralization and a familiar limit-order experience.

The strengths of the order-book model are genuine price discovery, tight spreads when market makers show up, and native support for limit orders. The weakness is that it needs those professional market makers to provide depth; without them a book is thin and a large order moves the price sharply. It is also the hardest design to decentralize fully, because low latency and high throughput pull against the overhead of a public blockchain.

The pool model up close

The pool venues solve the liquidity problem by removing the order book entirely. Instead of matching you with another trader, they match you with a pot of money. GMX, which runs on Arbitrum and Avalanche, is the archetype. In its first version, liquidity providers deposited a basket of assets into a single shared pool called GLP, and that pool was the direct counterparty to every trade; when traders won, the pool paid, and when they lost, the pool collected. GMX version 2, live since August 2023, split that one pool into isolated per-market pools (called GM) so that a blow-up in one market cannot drain the liquidity backing another, and it prices trades using Chainlink Data Streams, a low-latency oracle verified on-chain.

On Solana, Jupiter runs the same idea through its JLP pool, a basket of SOL, ETH, wBTC and USDC that traders borrow against to open leverage. There is no order book and no slippage on a standard fill; a price-impact fee protects the pool, prices come from an oracle, and liquidity providers earn roughly 75 percent of all fees as what the protocol calls real yield. Gains Network takes the model one step further into synthetics. Its gTrade platform settles every position in DAI against a single vault rather than requiring real liquidity for each market, which lets it list not just crypto but forex, stock indices and commodities, all backed by the same pot of stablecoins.

The appeal of the pool model is obvious: instant, zero-slippage fills, deep virtual liquidity from day one, and no need to court market makers. The catch is equally stark. The venue is a price-taker, not a price-discoverer, so it is only ever as accurate as its oracle, and liquidity providers are, in aggregate, short every trader’s winning bet. Those two facts explain most of the sector’s worst accidents, which we come to shortly.

Funding rates, mark price, and how the peg holds

Funding deserves a closer look because it is where traders quietly bleed or earn money regardless of whether their price call is right. On an order-book venue, funding is typically calculated from the gap between the perp’s own trading price and the index price, then charged across all open positions at each interval. If funding sits at a positive 0.01 percent every eight hours and you are long, you pay that fraction of your notional to the shorts three times a day. Over a week of crowded bullish positioning, that drip can quietly eat a real chunk of a leveraged position even if the price never moves.

Pool-based venues handle the same job a little differently. Because liquidity providers are the counterparty, the venue charges borrowing fees and funding that compensate the pool for the directional risk it is carrying, and skew between longs and shorts pushes those rates. The economic purpose is identical: make the expensive side of the market pay the cheap side, so that arbitrageurs are rewarded for closing any gap between the contract and spot. When funding runs extreme, it is itself a signal that positioning has become lopsided and a sharp reversal is more likely.

The mark price is the quiet guardian of the whole system. If a venue calculated your liquidation off the last trade in its own book, an attacker could jam a single print through a thin market and trigger a cascade of forced closures. By anchoring the mark to an oracle-derived index that aggregates several large markets, a venue makes that manipulation far more expensive. The flip side is that the oracle becomes the thing worth attacking, which is a recurring theme across every design that leans on one, and a subject our companion piece on who front-runs your trade examines in detail.

Liquidations, margin, and the insurance backstop

Leverage works only if the venue can close losing positions before they go underwater, and that is what the liquidation engine does. When you open a trade you post initial margin, the minimum collateral needed to open it. As the position moves against you, your equity falls toward the maintenance margin, the lower threshold below which the position is no longer adequately backed. Touch that line and a liquidation is triggered: the engine closes your position, often charging a penalty, to protect the rest of the market from your shortfall.

Traders choose how that risk is scoped. Isolated margin walls off a single position so that only the collateral assigned to it is at risk; cross margin pools all your collateral behind all your positions, which is more capital-efficient but means one bad trade can drag down the rest. The sizing of that buffer is why high leverage is dangerous: at 50x, a 2 percent move against you can be enough to wipe the position.

Sometimes a market gaps so fast that a position is closed only after it is already underwater, leaving a deficit someone must cover. Venues handle this with an insurance fund that absorbs the bad debt, and, as a last resort, with auto-deleveraging, where the system trims the most profitable opposing positions to balance the books. On Hyperliquid, a community vault plays the role of liquidity provider and liquidation backstop at once, which makes that vault central to both the venue’s returns and its worst day, as the next two sections show.

Who takes the other side: the LP vaults

On a pool-based venue, and increasingly on order-book venues too, ordinary users can become the house by depositing into a liquidity vault. Hyperliquid’s version is HLP, described in its documentation as a vault that “provides liquidity to Hyperliquid through multiple market making strategies, performs liquidations,” and “accrues a portion of trading fees.” It is community-owned: anyone can deposit, share in the profit and loss, and withdraw after a short lock-up. GMX’s GLP and GM pools and Jupiter’s JLP work on the same principle, paying their depositors the lion’s share of fees in exchange for standing behind traders.

This is one of the more honest forms of yield in on-chain finance, because it is paid out of real trading fees rather than freshly minted tokens. But the risk is specific and often misunderstood. A liquidity vault is, in net terms, short the aggregate winning trades of everyone using the venue. In calm markets, where most leveraged traders lose over time and fees pile up, the vault earns steadily. In a violent, one-directional move where a cohort of traders happens to be right, the vault can take a sharp drawdown. Depositors are underwriting an insurance book, not parking cash in a savings account, and the headline yield is the premium for that.

The design also concentrates systemic risk. If a single vault backs most of a venue’s liquidity and liquidations, then anything that forces a huge loss onto that vault threatens the whole exchange. That is not a hypothetical, as the JELLY episode would soon demonstrate.

The 2026 landscape: who leads and by how much

The sector grew up fast. Perp DEXs recorded their first trillion-dollar month in September 2025, with combined volume around 1.05 trillion dollars. At that peak the newcomer Aster briefly led with roughly 420 billion dollars, ahead of Hyperliquid at about 282 billion and the private-beta venue Lighter at around 164 billion. For a single day that October, the top two DEXs combined (about 78 billion dollars) nearly rivaled Binance’s futures desk at roughly 83 billion. Then the frenzy cooled; by early 2026 monthly volumes had fallen more than half from the October high.

Reading market share is where people get fooled, because volume can be juiced by zero-fee and incentive programs. Open interest, which is far harder to fake, is the better lens, and by that measure Hyperliquid dominates. Castle Labs data put its share of perp DEX open interest near 57 percent in late September 2026, with Aster and a surprise climber called Variational trailing well behind. Against the entire global perpetuals market, centralized venues included, Hyperliquid’s open interest reached records above eight billion dollars and roughly a tenth of all perp open interest, a share that CoinDesk noted was climbing through the year. Readers who want a live snapshot can watch the leaderboard on DefiLlama.

Aster is the cautionary tale of the cohort. Its ASTER token launched in September 2025 with backing from YZi Labs, the investment outfit spun out of Binance Labs, and the venue briefly overtook Hyperliquid on fees and volume. Within months the token had given back most of its gains, a reminder that incentive-fueled volume is not the same as durable usage.

Tokens and fee capture

Most major perp DEXs have a token, and the interesting question is what, if anything, it does with the fees the venue earns. Hyperliquid’s HYPE is the headline case. The project ran one of the largest airdrops in crypto history in November 2024, distributing 310 million tokens, about 31 percent of the genesis supply, to roughly 94,000 wallets with no private-investor allocation at all. Revenue then feeds an Assistance Fund that uses the large majority of protocol fees to buy HYPE on the open market, with the accumulated tokens treated as effectively removed from circulation. In effect the exchange converts trading activity into a steady, fee-funded bid for its own token.

Other venues wire the loop differently. dYdX directs trading revenue to the stakers who secure its chain. GMX shares fees with both its token stakers and its liquidity providers. Jupiter sends most fees to JLP depositors and buys back its JUP token. The common thread, and the reason this generation of venues is taken more seriously than the yield farms of 2021, is that the rewards are paid out of genuine revenue rather than token emissions. The caveat is that revenue is cyclical: the same CoinDesk analysis that charted Hyperliquid’s growth also flagged that a shift in its trading mix was eating into the fee revenue that backs those HYPE buybacks, a reminder that a token’s support is only as strong as the volume underneath it.

When it breaks: JELLY, oracle exploits, and MEV

The clearest way to understand a design is to watch it fail. On 26 March 2025, a trader opened a large short on JELLYJELLY, a thinly traded memecoin perp on Hyperliquid, while a whale dumped the token on spot markets. The engineered move forced Hyperliquid’s HLP vault to inherit the toxic short, and the vault’s unrealized loss swelled toward 12 million dollars while HYPE dropped as much as 22 percent. Hyperliquid’s validators then convened and voted to delist the market and force-settle it at a price that spared the vault, promising that ordinary users would be made whole by its foundation.

The rescue worked, but it detonated a debate about what decentralization means. If a small set of validators can step in, freeze a market and set a settlement price when the house is about to lose, how different is that from a centralized exchange pulling the plug? Bitget chief executive Gracy Chen was blunt, warning that “Hyperliquid may be on track to become FTX 2.0” and arguing that forcing settlement at a favorable price “sets a dangerous precedent.” The episode is a live example of the governance questions explored in our look at DAO security councils, where emergency powers that protect users one day can override them the next.

The pool model has its own signature failure: the oracle. In July 2025 an attacker drained roughly 40 million dollars from GMX’s version 1 pool on Arbitrum, exploiting a reentrancy flaw to manipulate the average price the contract used to value short positions, which in turn inflated the pool’s calculated value and let the attacker mint and redeem liquidity tokens at a rigged rate. GMX offered a bounty, the attacker returned most of the funds, and the protocol later finalized a payout of about 44 million dollars to make affected liquidity providers whole, as crypto.news reported. A technical post-mortem by the security firm Rekt traced the root cause to a years-old patch. The lesson repeats across the sector: when your price comes from outside the market, the thing that feeds that price is the thing worth attacking.

Who regulates this: the CFTC, the SEC line, and Europe

Here is the piece most explainers get wrong. In the United States, perpetual futures are derivatives, so they fall under the Commodity Futures Trading Commission, not the Securities and Exchange Commission. The line between the two agencies tracks the underlying asset: the SEC oversees securities and spot securities markets, while the CFTC oversees commodity derivatives. A March 2026 joint interpretation from the two agencies classified Bitcoin, Ethereum and more than a dozen other major tokens as digital commodities, and it is precisely that classification that cleared the way for commodity-style perps without the SEC in the room.

Through 2025 and 2026 the CFTC pushed hard to bring the product onshore. Bitnomial self-certified the first United States perpetual futures contract in April 2025; Coinbase launched perpetual-style futures for US customers that July; and in May 2026 the agency issued its first affirmative approval of a true crypto perpetual, a Bitcoin contract listed by the exchange operator KalshiEX. Kraken then rolled out CFTC-regulated perpetual futures to US traders in June 2026. Then-acting CFTC chair Caroline Pham framed the campaign in political terms, saying the agency was “wasting no time in fulfilling President Trump’s vision to make America the crypto capital of the world.”

That same statutory logic explains the geofences. Offering leveraged retail derivatives to Americans requires a registered futures exchange; the CFTC stated plainly that “leveraged retail commodity trading can only occur on futures exchanges.” Offshore perp DEXs hold no such registration, so they block US users by IP rather than risk enforcement. Hyperliquid is trying to change that from the other direction, petitioning the CFTC in 2026 for rules that would let licensed firms offer perps built on its chain, on the argument that the current rulebook was written for traditional intermediaries and does not fit self-custodial markets.

Europe draws the line somewhere else entirely. In February 2026 the European Securities and Markets Authority reminded firms that products marketed as perpetual futures are, in substance, contracts for difference that fall under MiFID II rather than the bloc’s dedicated crypto rulebook. That matters because crypto CFDs sold to retail investors are capped at 2:1 leverage under measures in force since 2018, a world away from the 50x on offer offshore. For how the broader European framework fits together, see our guide to what the MiCA rulebook does for you.

How to actually use a perp DEX (and the risks)

If you are in a jurisdiction where these venues are accessible, the mechanics are straightforward. You need a self-custodial wallet, and the smoother experience increasingly comes from the smart-account wallets described in our primer on what makes a wallet smart. From there you fund the wallet with a stablecoin on the relevant chain, connect, and the venue’s interface looks much like any exchange: pick a market, choose long or short, set your leverage and your margin mode, and place the order. New markets are added constantly, and the way a venue lists them, sometimes permissionlessly, is a topic in its own right, covered in our piece on how perp markets go live.

The discipline is in the risk management, and the risks here are not the same as on a centralized exchange. Before you size a position, walk through the ones specific to this product.

• Liquidation risk: high leverage means a small adverse move can wipe you out. Prefer isolated margin and modest multiples until you know the venue’s quirks.
• Funding drag: holding a crowded position through many funding intervals can cost real money even if the price goes nowhere.
• Oracle and smart-contract risk: a bug or a manipulated price feed can produce losses no trader could have hedged, as the GMX exploit showed.
• Governance override risk: validators or token holders may intervene in a market, as Hyperliquid did with JELLY, in ways that help or hurt your position.
• Regulatory and access risk: geofences can change, and a venue you rely on may cut off your region.
• Key security: self-custody puts the burden on you; a lost seed phrase or a compromised signing device means lost funds, with no support desk to call.

None of that is a reason to avoid the product, only to treat it as the sharp instrument it is. The appeal of a perp DEX is that it hands you both the upside and the responsibility that a centralized venue would otherwise manage on your behalf.

Frequently Asked Questions

What is a perp DEX?

A perp DEX is a decentralized exchange for perpetual futures, leveraged contracts that track an asset’s price and never expire. Instead of depositing funds with a company, you trade from your own wallet while smart contracts handle pricing, funding and liquidations. Hyperliquid, GMX, dYdX and Jupiter are leading examples.

Is Hyperliquid the biggest perpetual DEX?

Yes. By open interest, the harder-to-fake metric, Hyperliquid held well over half of the perp DEX market through 2026 according to data cited by Castle Labs, and it also tops most volume rankings. Aster briefly rivaled it in late 2025 before cooling off.

Can people in the United States trade on perp DEXs?

Most offshore perp DEXs geoblock US residents, because US law requires leveraged retail derivatives to trade on a CFTC-regulated futures exchange. In 2025 and 2026, licensed venues such as Bitnomial, Coinbase and Kraken began offering CFTC-regulated perpetual futures to US customers as a compliant alternative.

What is the funding rate on a perpetual future?

The funding rate is a recurring payment exchanged between long and short traders, often every one or eight hours, that keeps the perpetual price close to the spot price. When the perp trades above spot, longs pay shorts; when it trades below, shorts pay longs. The exchange does not keep the payment.

Are perp DEXs safe?

They remove the custody risk of handing funds to an exchange, but they carry their own: smart-contract bugs, oracle manipulation, liquidation cascades, drawdowns for liquidity-pool backers, and the chance that validators or governance intervene in a market, as Hyperliquid did during the 2025 JELLY incident. Use modest leverage and isolated margin, and only risk what you can afford to lose.

By the HOGE Wire markets desk. This explainer is for information only and is not investment advice.