On-chain Perpetuals: app-chain vs Solana vs Ethereum L2s

A single app-chain (proprietary application chain) moves more perpetuals than all Ethereum chains combined: Hyperliquid processes approximately $205.5 billion per month, compared to ~$74 billion for all of Solana and ~$32 billion spread across Ethereum's archipelago of Layer 2s (July 10, 2026, DeFiLlama). Most on-chain perpetuals coverage compares brand against brand —Hyperliquid vs. GMX, Jupiter vs. Pacifica— and ranks DEXs (decentralized exchanges) by product. This piece does something different: it aggregates volume by infrastructure model —the dedicated single-product app-chain, the generalist Layer 1 where multiple DEXs compete, and the Ethereum Layer 2 archipelago fragmented across half a dozen networks—, explains exactly which protocols each bucket includes and its source, and answers the infrastructure question that the product-view leaves open: where should a new perpetuals DEX be launched in 2026?

Why compare by infrastructure layer instead of protocol?

A team preparing to launch a perpetuals DEX first chooses its substrate: the chain it runs on. With Hyperliquid moving about $205.5 billion a month from its own chain, that decision — whether to use an app-chain, a shared Layer 1, or one of the many Ethereum Layer 2s — dictates the liquidity that can be captured, the infrastructure cost, and even the token economics before a single line of the matching engine is written. The instrument itself is not the novelty — the perpetual swap was invented by BitMEX in 2016, and we covered its shutdown and legacy here—what is being decided in 2026 is the substrate.

DeFiLlama ranks perpetuals by protocol and by chain, but not by architecture. This architectural summation is our own work: we gathered individual protocols into three buckets —the Hyperliquid app-chain, all Solana perpetual DEXs as a block, and everything running on Ethereum Layer 2s— from disaggregated data. The protocol-level comparison —which token to buy, who has better execution— was resolved in Hyperliquid vs. GMX vs. dYdX; this piece is the layer beneath, the substrate layer.

The thesis, in advance: in 2026, concentration beats fragmentation. A chain dedicated to a single order book accumulates more real liquidity than three DEXs fighting over the same Layer 1, and significantly more than a dozen perp DEXs scattered across incompatible Layer 2s. The retained open interest of each layer confirms this cell by cell, as seen in the table.

Which protocols does each model include and by what method?

To ensure the aggregation is reproducible —so anyone can redo the math— we set three rules: a single snapshot date for all three categories (July 10, 2026), a single volume source per cell (DeFiLlama, /perps dashboard and its chain views), and an explicit assignment of each protocol to a bucket. These are the three models and what is included in each:

• Dedicated App-chain — Hyperliquid. A Layer 1 blockchain built exclusively for its order book (HyperCore). A single protocol dominates the category; dYdX, an app-chain on Cosmos with ~$2.66 billion over 30 days —residual compared to Hyperliquid—, is excluded from the bucket to avoid mixing ecosystems. Source: DeFiLlama, hyperliquid-l1 chain. The technical details of why a proprietary chain changes the rules can be found in Hyperliquid architecture.
• General-purpose Layer 1 — Solana (Jupiter + Drift + Pacifica). A general-purpose blockchain where several perpetual DEXs compete for the same liquidity. We aggregate the three largest: Jupiter Perps, Drift, and Pacifica. Source: DeFiLlama, solana chain (which already aggregates the network total).
• Ethereum Layer 2 Archipelago — GMX + Vertex + StarkWare cluster. The perp DEXs that settle, directly or indirectly, against Ethereum through its Layer 2s: GMX on Arbitrum and Vertex —migrated from Arbitrum to Ink (Kraken's L2) after withdrawing its VRTX token in 2025, now rebranded as Nado—, plus the StarkWare cluster (Paradex on Starknet, Extended and edgeX on StarkEx validiums —a Layer 2 that stores data off-chain—). Ink is also an Ethereum Layer 2, so the bucket holds up. Source: DeFiLlama by protocol, manually aggregated. This is the most difficult bucket to close because activity is spread across networks that do not share liquidity.

Three methodological honesty warnings. GMX also has a deployment on Avalanche, an independent Layer 1 and not an Ethereum Layer 2: we count its Arbitrum branch as representative and leave Avalanche out of the model. The StarkWare cluster is "Ethereum" in a loose sense —it uses its proof technology and anchors against Ethereum— not because they all settle on Layer 1 with the same mechanism. And a freshness warning: at the end of July, the Drift adapter on DeFiLlama reports zero volume, so the live Solana total in /perps/chains (~$34.7 billion on Jul-31) underestimates the bucket; the Jul-10 cutoff is the last snapshot with Drift included.

The second, regarding what fits into no specific bucket: Aster runs on BNB Chain, which is neither a single-product app-chain, nor Solana, nor an Ethereum Layer 2. It represents a fourth category — a perp DEX on a centralized issuer's alternative Layer 1 — outside the scope of this comparison by design. We declare it rather than hide it: Aster reached nearly 70% of the tracked perp DEX market share at its peak in September 2025, before being delisted, and stands at around 8% at the close of July 2026 (~$41.6 billion in 30 days). This figure should be taken with a grain of salt: DeFiLlama delisted it in October 2025 due to volume suspiciously correlated with Binance, and later relisted it while warning of "verification gaps" in its own data. Including it in any bucket would muddy the total; leaving it out, with this note, keeps it clean.

How much does each architecture move in July 2026?

The table summarizes the snapshot from July 10, 2026. The 30-day volume is the most robust data (direct from DeFiLlama chain views); the aggregated open interest is composed by protocol and should be read as an approximation, with cells marked for refreshing on the day of publication.

In the CleanSky aggregation (cutoff 10-jul-2026, methodology above), Ethereum layer 2s total ~$32 billion over 30 days, compared to ~$205.5 billion for Hyperliquid. The cutoff holds at month-end: on 31-jul DeFiLlama reported ~$198.4 billion for Hyperliquid (−3.5% in three weeks) and ~$32.3 billion for the same layer 2 bucket, with no change in ranking.

Ethereum, which in 2021 was the default home for almost everything on-chain, is today competing for second place in perpetuals volume with zkLighter and Solana —as of the end of July, less than $2 billion separate the three and the ranking rotates daily—, at more than five times the distance from the app-chain that did not even exist back then.

The fifth column is the quality detector. The ratio between open interest (OI: the value of live positions at a given time, actual capital locked) and daily volume measures whether the volume represents positioning or rotation. Hyperliquid —at ~1.5— retains ample capital relative to its trading activity; Solana —around 0.3— moves much more than it retains; and the archipelago sits at a midpoint (~0.7), with edgeX's open interest almost single-handedly supporting the bucket. That gap is the difference between liquidity that stays and volume that passes through.

Why Pacifica's 54% on Solana is misleading?

The Solana bucket hides the clearest example of why raw volume lies. In October 2025, the entire network's perpetual volume hit a record $65 billion for the month, and a single platform, Pacifica, captured 54% of that share after overtaking Jupiter in daily volume in September —the jump was from less than 200 million on September 9 to over 1 billion on the 30th (SolanaFloor). Within six months of its launch, Pacifica had surpassed Jupiter Perps ($264 billion in all of 2025) and Drift ($92 billion).

And yet Pacifica retained, with that 54% of the volume, less than 5% of the network's open interest: about $68.8 million in OI while crossing more than $1 billion a day at the end of September 2025. It is an OI/daily volume ratio in the range of 0.07 —more than four times worse than the average for Solana itself, around 0.3 at the July 2026 cutoff—, the unmistakable signature of incentivized volume: short-term capital rotating to farm points before an airdrop, not positions that anyone is holding. It is the same pattern we documented by protocol in the Hyperliquid-GMX-dYdX comparison, where newcomers with ratios of 0.12-0.18 inflated their nominal share.

The counterpoint within the same bucket is provided by Drift: it regained Solana's OI leadership after the April 2026 exploit —over $700 million by mid-year— despite falling behind Pacifica in volume. It is the exact reverse, and explains why summing Solana as a block makes sense: the generalist Layer 1 captures real activity (Drift) and farming activity (Pacifica) simultaneously, and only the aggregate with its OI alongside tells how much of that share is solid. The story of how Drift weathered that attack and regained OI leadership is in the Drift hack of April 2026.

Why is perpetual volume falling on Ethereum L2s?

The third bucket is the one that has changed the most, and to the downside. The StarkWare cluster —Paradex, Extended, and edgeX, the perp DEXs built on StarkWare's Layer 2 technology— dropped from ~16% of the total tracked perp DEX volume in January 2026 (BlockEden) to ~5% in July (DeFiLlama: edgeX 3.75%, Extended 1.26%, Paradex 0.07%). There was a peak along the way: edgeX managed to surpass Aster in April 2026 with about $73 billion monthly, before deflating. GMX, the veteran of the bucket, long ago stopped fighting for active trading and repositioned itself as a yield protocol on Arbitrum and Avalanche; its monthly volume hovers around $2.8 billion. The context of that retreat, involving the GMX rewards pause, was covered in the 2026 GMX rewards pause. The newcomer to the bucket is RISEx, the perp DEX of the RISE L2 with a fully on-chain order book: it closed July with about $110 million daily and ~$2.4 billion over 30 days —outside the cutoff sum, which defined the bucket before its takeoff— after launching its Ignite points program on July 24 with more than $3 billion in cumulative beta volume. It is the newest example of the archipelago thesis: real growth, in yet another compartment that does not share liquidity with its neighbors.

The lesson of the archipelago is structural. Spreading a perpetuals market across Arbitrum, Starknet, and several StarkEx validiums fragments liquidity into non-communicating compartments: a large trader on edgeX cannot find the counterparty that would exist in a single book. For perpetuals —a product where book depth is everything— this slicing is precisely the problem, and it is reflected in the bucket's drop from 16% to 5% in six months.

Why does the dedicated app-chain win?

Hyperliquid's ~$10.5 billion in OI alone exceeds, by a wide margin, the combined open interest of all of Solana and the entire Layer 2 archipelago. With the layer-view in front of us, its advantage ceases to be brand and becomes architecture: a chain built for a single order book concentrates all activity into a single matching engine, while Solana divides it between Jupiter, Drift, and Pacifica, and the archipelago between half a dozen networks. Liquidity attracts liquidity, and that is the only one of the three layers capable of closing that loop.

That concentration translates into the only metric that cannot be farmed with incentives: retained open interest. When measuring committed capital instead of crossed dollars, the app-chain doesn't just win by a small margin: it wins by an order of magnitude. It is worth, however, specifying the share: Hyperliquid hovers around 35-40% of the perp DEX tracked by DeFiLlama, but captures around 70% of decentralized perpetuals across all chains, and between 6% (The Block, April 2026) and ~10% (BlockEden, MEXC Research) of global derivatives volume including centralized exchanges, depending on which tracker measures the denominator. Three denominators, three figures: the common trap is to cite 70% without stating the basis.

Where to launch a new perpetual DEX in 2026?

The layer-view turns the abstract infrastructure question into a decision with three measured options. There is no universal answer, as each layer optimizes for something different, but the snapshot data narrows down the trade-off for each:

• Proprietary App-chain (Hyperliquid model). Maximum liquidity concentration and total engine control, in exchange for the massive cost of starting a Layer 1 with validators, security, and bootstrapping liquidity from scratch. It is the winning bet in volume and OI, but only if the liquidity loop can be sustained before initial capital runs out.
• Generalist Layer 1 (Solana model). Inherit an existing user base, wallets, and liquidity, and launch quickly; in exchange, compete within the same chain with other DEXs for that liquidity, and market share is volatile —Pacifica went from nothing to leader in six months, and its OI warns of how fragile that leadership can be.
• Ethereum Layer 2 (Archipelago model). Access Ethereum's security and brand, but fall into a fragmented ecosystem where liquidity is not shared between networks; the January to July 2026 series shows this bucket shrinking from 16% to 5%. It is the option with the strongest structural headwind for a perpetuals product.

To decide without being swayed by volume rankings, there is only one rule to order the three layers: divide the aggregate open interest by the model's daily volume. A high ratio (Hyperliquid, ~1.5) indicates that the layer retains capital; a low ratio (Solana, ~0.3) suggests that a large part of that volume is turnover that will evaporate once incentives cease; the middle ground of the archipelago (~0.7) indicates that what little remains there is, at least, positioned. The same mechanics of funding rates that make or break this ratio are explained in how funding rates work on a perpetual DEX.

Aggregation by model leaves three conclusions that brand-vs-brand comparison cannot provide: concentration beats fragmentation; volume without open interest alongside it is a hollow metric —Pacifica's 54% with less than 5% of OI is the proof—; and substrate matters before product. This piece looks at the market through its infrastructure layer; the protocol view —which token to buy, who executes a large order best— lives in Hyperliquid vs. GMX vs. dYdX, and the full landscape of platforms, airdrops, and risks in DeFi perpetuals in 2026.

Sources and links: DeFiLlama — Perp DEX volume by chain · DeFiLlama — Perp DEX rankings · SolanaFloor — Pacifica and OI · SolanaFloor — Pacifica reaches $1B in daily perp volume · Osprey Funds — record $65 billion on Solana · The Block — perp DEX share of global market · BlockEden — StarkWare cluster share (Jan 2026) · Yellow Research — Hyperliquid's share