Author: kokii, Investor at Summer Capital

TL; DR

Solana has dominated the on-chain spot market (with ~30% of total on-chain DEX share and spreads approaching CEX levels), primarily driven by the PropAMM architecture. However, in the Perp market, HyperLiquid holds an overwhelming dominance with over 70% of OI and volume share.
The root cause of this gap is not speed, but the unpredictability of sequencing: Solana's Leader scheduling mechanism cannot provide market makers with a deterministic guarantee that "cancellations will be prioritized over fills," forcing MMs to proactively widen spreads and reduce depth, creating a vicious cycle of deteriorating liquidity.
Solana is counterattacking on two fronts: the protocol layer (Alpenglow compressing finality to 150ms, MCP breaking the single Leader monopoly, BAM/ACE enabling programmable sequencing at the application layer), and the application layer (Phoenix's asymmetric CU scheduling, Bulk's sidechain risk engine, JTX's retail-professional channels).
However, Solana's long-term winning strategy lies not in replicating HyperLiquid, but in what HyperLiquid cannot do: DeFi composability, the natural fit of the oracle pool model for long-tail/RWA Perps, and the ecosystem flywheel that converts massive spot retail flow into Perp participants. This is a structural ceiling of the dedicated chain architecture, and also Solana's long-term moat.

1. The Contrast: King of Spot, Laggard in Derivatives

Solana DEX trading volume has long accounted for 30% of total on-chain DEX volume, exceeding hundreds of billions of dollars monthly and repeatedly surpassing the combined total of Ethereum and L2s.

In terms of price discovery efficiency, the spread for mainstream assets (like SOL-USDC, BTC-USDC) compared to CEXs like Binance and OKX is already minimal. The effective spread for large trades can be as low as 0.4-1.6 bps, significantly better than the 5-9 bps level of traditional AMMs. The core mechanism driving this result is the PropAMM (Proprietary Automated Market Maker). Currently, over 50% of spot trading on Solana occurs on PropAMMs: professional market makers like Wintermute/Jump encode their proprietary pricing strategies directly into on-chain contracts, pushing quotes on-chain at high frequency via oracles. The market-making logic shifts from "passive liquidity provision" to "actively pushing the latest market price."

Feature	Prop AMM	Traditional Passive AMM
Capital Source	100% market maker proprietary capital, no public deposits accepted	Public LP deposits
Pricing Mechanism	Off-chain prediction model → On-chain parameter updates (Active Pricing)	Fixed mathematical formula xy=k / Concentrated Liquidity
Quote Updates	Multiple updates per block, even every 50ms	Moves passively only when trades occur
Liquidity Type	Active Liquidity	Passive Liquidity
Frontend Interface	Usually no frontend, routed through aggregators	Has an independent frontend
Code Transparency	Closed-source or partially obfuscated	Open-source
Applicable Assets	Mainstream short-tail assets (SOL, BTC, ETH, etc.)	Full spectrum of assets including long-tail
Capital Efficiency	Extremely high: $10M TVL can handle $1B+ daily volume	Lower: large amounts of idle capital
Spread	SOL-USDC 0.4-1.6 bps (~5 bps for large trades)	SOL-USDC 5-9 bps (double digits for large trades)

In contrast, Solana Perp DEX's trading volume and Open Interest (OI) market share pale in comparison. Hyperliquid dominates the on-chain Perp market with about 70% share, while Aster (BNB ecosystem), Variational, Lighter, and edgeX (ETH ecosystem) are also striving to catch up. Looking back at Solana, which was considered "centralized and high-performance" and suitable for Perp trading, its Perp sector has been lagging behind this cycle. Drift was recently hacked for over $285M+, significantly weakening its momentum.

2. The Evolution Logic of Perp DEX: Why Dedicated Chains Won

The fundamental reason is that spot is about asset issuance, while derivatives are about price discovery. Solana's past success came from repeatedly catching the rhythm of new asset issuance. However, Perp pricing has long occurred on CEXs; what Perp DEXs do is bring prices on-chain through various means.

Opening the spot and perpetual order books on Binance, they look almost identical, but underneath they are two completely different engineering entities:

Spot Matching: Matching = Settlement. Buyers and sellers perform a one-time asset swap, with no "position" continuing after the transaction is complete. The engine requires no external dependencies, and risk is zeroed out at the moment of execution.
Perp Matching: Matching is not the end of settlement, but an Event Source——a domino. Every match triggers a complex downstream chain: mark price update → position update → margin recalculation → unrealized PnL refresh → potential liquidation trigger. The matching engine and the risk engine are deeply coupled.

A spot DEX can run with a simple xy=k formula and a few hundred lines of code, but a Perp DEX needs to maintain much more complex and high-frequency operational logic:

Matching: Maintains the order book, executes price-time priority matching
Oracle/Index Price: Maintains the mark price system
Price Anchoring: Calculates and executes funding rates
Margin Management: Enforces initial and maintenance margin rules
Liquidation: Force-closes positions when account equity falls below maintenance margin
Solvency Guarantee: Prevents systemic defaults through insurance funds and ADL mechanisms

The early practice of Perp DEXs using the Oracle-LP Pool model, where the oracle pulls prices from quote sources and the LP pool acts as the counterparty for trades, successfully created significant trading volume on-chain with simple contracts (GMX's peak daily volume was ~$1B). This was the golden age of DeFi Summer, where the industry believed that elegant mechanism design could simply replicate all complex off-chain financial infrastructure on-chain, just as Uniswap and Aave did. However, practice proved this does not hold true for Perps:

For Traders: Order types, capital efficiency, matching latency, and OI caps (limited by TVL) all lagged far behind CEXs
For Market Makers: For large-scale professional market-making capital seeking precise risk control, LPs could not actively manage quotes and risk like order book market makers

Professional market makers and traders require low latency, deterministic execution, high capital efficiency, and predictable risk engines. To replicate the CEX experience on a DEX, one must rely on highly optimized, near-vertically integrated infrastructure. General-purpose public chains have inherent architectural disadvantages. The evolution of Perp DEXs has gone through three generations of paradigms, constantly balancing performance, liquidity, and decentralization (permissionless access, transparent matching, withdrawal security), ultimately settling on architectures like Hyperliquid / Lighter / Aster, which are highly centralized yet retain decentralized characteristics:

Self-developed Full-Stack Infrastructure: Custom consensus mechanisms and execution engines provide deterministic transaction sequencing, extremely low and predictable latency, and a gas-free experience for cancellations and order placements.
Self-Reinforcing Liquidity: Market makers provide extremely narrow bid-ask spreads and deep order book depth -> reducing friction costs for large and high-frequency traders, attracting large order flow -> rapid increase in trading volume attracts more professional market makers to connect and provide better quotes

Dimension	Centralized Exchange (CEX)	Phase 1 (2021-2023)	Phase 2 (2023-2024)	Phase 3 (2025+)
Architecture	Centralized Limit Order Book (CLOB) matching engine deployed on high-performance cloud	Liquidity Pool Passive LP acts as counterparty, oracle pricing	Hybrid Order Book Off-chain matching (in validator RAM)	Dedicated L1 CLOB Matching engine embedded in consensus mechanism
Representative Projects	Binance, Bybit	GMX, Jupiter	dYdX V4	Hyperliquid, Lighter
Matching Latency	<50µs	400ms-2s (limited by oracle/block)	~1s (off-chain matching)	5-15ms (on-chain)
Order Throughput	1,000,000+ TPS	1K-3K TPS (limited by public chain)	1.5K-10K TPS (hybrid architecture)	200K TPS (dedicated optimization)
Order Type Support	Market, Limit, OCO, Trailing Stop, TWAP, etc.	Basic (Market/Limit)	Advanced (OCO/TWAP)	Full near-CEX experience (Conditional Orders)
Capital Efficiency	Cross-asset / Portfolio Margin	Low (locked liquidity pool)	High (unified margin)	Very High (cross-margin)
Liquidity Provision	Market Making via API/FIX	Passive (AMM/vAMM)	Active (Vault/HFT)	Native (HLP+CLOB)
Settlement Finality	Instant (internal database)	1-15s (depends on public chain)	s~1 second (consensus confirmation)	~0.2s (HyperBFT)
Withdrawal Delay	2-5 minutes (standard)	5-30 minutes (network confirmation)	L2 instant / L1 requires 7 days	Instant (ZK-proofs)
API Round-Trip Time	30ms-80ms (AWS)	300-500ms (Public RPC)	150-400ms (Gateway)	150-300ms (Dedicated Channel)

3. Solana's Structural Constraints

Solana is a general-purpose L1, and compared to a dedicated Perp L1, there is inevitably a performance gap. However, Solana's biggest constraint on the Perp market is not speed and throughput, but ordering unpredictability.

Structural limitations of general-purpose block space. The matching and liquidation logic for perpetual contracts must share block space with Memecoins, NFTs, and DeFi, lacking dedicated block lanes. During periods of high market volatility, market makers cannot guarantee timely quote updates.
Conflict between on-chain risk engines and compute units. Running a comprehensive liquidation and margin control model on-chain (such as multi-position cross-margin, asset correlation discounts, etc.) requires extremely complex mathematical computations for every transaction, consuming a large number of Compute Units (CU). Solana has strict limits on the maximum CU available for a single account within each Slot and the total block CU. This restricts the possibility of designing complex, advanced risk engines for on-chain Perp DEXs.
Non-deterministic block production and ordering. Solana uses a Leader-based block production model. The Leader node rotates among network validators every 4 slots (approximately 1.6 seconds), and different validators run clients (like Firedancer, Agave, etc.) with differences in transaction scheduling and packaging logic. In a CEX or dedicated L1 (like Hyperliquid), the order book follows strict FIFO or deterministic matching rules. On Solana, whether a market maker's quote update or a taker transaction gets packaged first depends entirely on the current Leader's scheduler. When external market prices move sharply, if a market maker initiates a cancellation order but fails to get it packaged on-chain before an arbitrageur's taker order, the trade is executed, resulting in a direct loss.
Multi-layered fee stack increases uncertainty. Transaction costs include L1 base fees, priority fees, Jito Tips, and RPC landing service fees. The complex and dynamically changing fee stack makes it difficult for MMs to accurately predict the comprehensive cost of a single quote update, weakening the profitability stability of high-frequency strategies.

Suppose a market maker places a BTC sell order at $64,000. The market suddenly drops, so they need to cancel the order and place a new one at $63,000, but

On Solana CLOBs like Phoenix, updating an order consumes 100,000–300,000 CU/order, and cancellation has a cost.
If this coincides with a Meme launch or NFT mint, the new order might not be immediately included on-chain. The market maker can only increase the probability of the cancellation being included on-chain by paying higher Priority Fees and Jito Tips.

Ordering unpredictabilitydirectly erodes market maker profits. Market makers are forced to price risk into the spread, compelling MMs to actively widen spreads and reduce depth to buffer the uncertainty window, creating a vicious cycle: market makers exit → liquidity thins → slippage increases → user attrition.

In terms of trading experience, Blockworks calculated the round-trip friction cost (Base Taker Fee + Market Impact Cost Slippage/Impact) for BTC under different Perp DEXs. Solana Perps showed significantly higher slippage than Hyperliquid and Binance in large-size scenarios:

Trade Value	Hyperliquid	Pacifica (Solana offchain-CLOB)	Drift (Solana hybrid CLOB)	Binance (CEX)
$10K	~5 bps	~3.5 bps	~15 bps	~6 bps
$100K	~5.5 bps	~4.5 bps	~18 bps	~6.2 bps
$500K	~7 bps	~12 bps (impact cost expands rapidly)	~26 bps	~7.5 bps
$1M	~9 bps	~18 bps	~42 bps	~9 bps

4. Solana's Counterattack: Protocol Layer + Application Layer

The Solana ecosystem is advancing reforms on two levels simultaneously, precisely targeting the core pain points of market makers.

Protocol Layer

Alpenglow Upgrade

Alpenglow (led by Anza) is Solana's largest consensus layer overhaul to date. It will replace the existing PoH and Tower BFT, introducing Votor voting and Rotor propagation mechanisms, compressing finality from the current ~12.8s to 100-150ms. This significantly shortens the blank period market makers wait for trade feedback, reducing passive one-sided position risk. Expected to land by 2026.

MCP (Multiple Concurrent Proposers)

Currently, Solana operates under a single Leader model, where a single node monopolizes block production and packaging rights for each block, introducing significant censorship risk and MEV extraction risk. The Constellation architecture proposed by Anza introduces MCP technology:

Breaking the single Leader monopoly: Allows multiple concurrent validators to propose transaction batches and submit them to the consensus network simultaneously. Because multiple proposers can package transactions concurrently, the pace of "state transitions" and logic execution within a block is greatly accelerated, enabling the network to process state updates continuously.
Neutrality guarantee: Since arbitrageurs cannot achieve front-running through a single channel (e.g., spamming a specific Leader with junk packets), the probability of market maker quotes being "sniped" decreases exponentially, making Solana's L1 physical environment more neutral and fair.

CU Doubled to 100M

To address the computational preemption problem between financial contracts and liquidation engines without sacrificing global composability, the Solana community is advancing a plan to double the per-block compute unit (CU) limit from its previous cap to 100M CUs (expected to land by mid-2026):

Increased Capacity: With the CU limit doubled, liquidation modules and complex cross-margin accounting logic will no longer be constrained by strict local CU resource ceilings during execution, effectively reducing transaction failure rates under high concurrency.
Further Reduction in Transaction Fee Rates: The substantial expansion of block physical capacity further lowers the actual base cost allocated per transaction, providing ample and low-cost resource space for on-chain high-frequency market making and massive order volumes.

Jito BAM and ACE Framework

Before a complete L1 consensus overhaul, Jito took the lead by providing the market with a locally available "high-speed execution channel" through its pluggable client patch — Jito BAM (Block Assembly Marketplace). It went live in September 2025. Currently covering approximately 30% of network stake, combined with the original Jito client's 30%, it forms an extremely strong network coverage.

Jito BAM is a new type of block assembly algorithm and micro-auction marketplace. It changes the previous logic of randomly packing transactions based solely on priority fees, replacing it with a strict intra-block deterministic auction based on Fee/CU (tip efficiency per compute unit). Its core is a TEE (Trusted Execution Environment, AMD SEV-SNP) encrypted mempool — transactions are invisible to anyone (including node operators) before execution, structurally eliminating the information asymmetry required for sandwich attacks. Simultaneously, it generates an on-chain cryptographic audit trail (Attestation), allowing any observer to verify ordering fairness.

The core killer feature of the BAM framework is the introduction of ACE (Application-Controlled Execution) logic, and the first plugin to be widely piloted and deployed is the Maker Priority Plugin:

Sub-slot Deterministic Cancels: The plugin allows market makers to submit special "quote updates" and "cancel resting orders" commands to BAM through specific cryptographic proofs or dedicated channels.
Deterministic Packing Guarantee: At the sub-slot level, market maker cancellations and price updates are forcibly prioritized at the front of the block before taker execution transactions.
Value Proposition: Even during network congestion, as long as a market maker's cancellation transaction is sent, it can clear expired resting orders with a near-100% success rate before arbitrage takers take effect. This completely cuts off the sniping behavior of arbitrageurs targeting market makers' "stale quotes," providing the physical basis for on-chain order book spreads to narrow to levels approaching CEXs.

If these upgrades are implemented and achieve target adoption rates, they will systematically improve the execution determinism and capital efficiency of Solana Perps.

Application Layer

To resolve the conflict between L1 performance ceilings and trading demand, a variety of differentiated Perp architecture systems have emerged in the Solana ecosystem. These solutions make different trade-offs and engineering choices between "fully decentralized composability" and "ultimate trade execution performance":

Phoenix

Phoenix, as the first high-profile fully on-chain CLOB on Solana, was built by the Ellipsis Labs team behind Prop AMM SolFi. Due to their SolFi experience, the team has deep insights into Solana computational optimization.

Technical Architecture: A fully on-chain non-custodial CLOB. All buy and sell order matching logic, resting order queue states, and position liquidations are entirely driven by smart contracts on Solana L1.

Computational Layer Asymmetry Countermeasure: A single quote update transaction consumes only 500 CU, while taker execution logic consumes 150,000 CU. This cleverly achieves a probabilistic near-300x priority scheduling rate for market makers under Solana's existing CU scheduling mechanism.
Advantages: Extreme composability. All trading states are on L1, meaning held open positions can easily serve as collateral for other lending protocols and support atomic basis trades with other DeFi legos.
Disadvantages: Trading latency bottleneck is limited by the Solana L1 Slot performance ceiling, with potential for congestion lag during extreme volatility.

JTX

JTX represents Jito's major expansion into upstream trading terminals/consumer-grade applications. Strictly speaking, it is not a Perp DEX but a front-end trading terminal.

Technical Architecture: Deeply integrates with Phoenix Perps for Perp liquidity, using it as the preferred decentralized contract matching layer. Implements a genuine on-chain order mechanism where limit orders queue directly on Solana and are executed by smart contracts within the same Slot when the price is met. Supports CEX-grade TWAP, stop-loss/take-profit, and various automated order types.
Advantages: Leveraging Jito's optimization experience and dominant position in Solana transaction flow, combined with Jito BAM's Maker Priority Plugin, market makers and users can achieve lower trading latency.

Bulk

Bulk proposes a highly innovative heterogeneous scaling architecture. Its core logic is: moving the matching and risk control engines to a dedicated sidechain for execution, while keeping fund settlement and custody on Solana L1.

Technical Architecture: Users deposit and lock margin in Solana L1 smart contracts (Kamino, Loopscale, etc.), but subsequent matching, trading, and liquidation engines are processed on a sidechain jointly run by Solana validators (running the Bulk-Agave client). Testing already has support from over a dozen large validator nodes (representing over 5% of total network Stake).
Advantages: Market-maker-oriented mechanisms, gas-free high-frequency order placement, 20ms quote update frequency, FIFO (First-In-First-Out) matching with built-in "cancel priority." Since sidechain validators are also Solana validators, users can theoretically use collateral like that on Kamino to provide margin and trade on the sidechain via a delegation mechanism, ensuring a degree of composability.

Pacifica

Technical Architecture: Fund deposits, locking, and settlement withdrawals remain in Solana L1 smart contracts. All order book resting orders, matching, and position accounting are processed asynchronously on centralized off-chain servers.
Advantages: Extremely strong performance, almost identical to the CEX experience. Once captured 62% of Solana Perp market share in January 2026 through point reward incentives.

Disadvantages: Completely sacrifices on-chain composability. Various trade-related state information, such as positions, orders, and liquidation status, cannot be verified on-chain, preventing other Solana applications from integrating with it. The project's roadmap plans to migrate it to a self-operated Substrate application chain in the future, which will also pit it directly against Hyperliquid in head-to-head competition.

Jupiter Perps

Technical Architecture: Adopts the JLP (Jupiter Liquidity Pool) oracle pricing pool mechanism. There is no matching engine or order book; users directly engage in leveraged betting against the pool's mixed assets (SOL, BTC, ETH, USDC, etc.) using Pyth's latest oracle execution price.
Advantages: Extremely simple experience, highly friendly to retail users. JLP tokens offer high annualized yields and are widely accepted as high-quality collateral on platforms like Kamino.
Disadvantages: Fixed trading fee rate, rigidly determined by oracle quotes (simulating slippage through additional Price Impact), unable to support high-frequency and massive professional trading. Current monthly trading volume oscillates around 10B, serving as the retail user base.

Protocol Name	Matching Location	Asset Custody & Settlement	Risk Engine Characteristics	Latency & Throughput	Core Innovation & Representative Mechanism
Phoenix Perps	Fully On-Chain L1	Solana L1 Smart Contracts	On-chain liquidation model, anti-manipulation mark price hybrid algorithm	Limited by L1 Slot (~400ms latency)	Asymmetric scheduling: 500 CU quotes vs 150k CU takers
Bulk	Independent Validator Sidechain	Solana L1 (delegatable linkage with Kamino, etc.)	CME SPAN Portfolio Risk Engine (70%+ margin reduction for hedged positions)	20ms sidechain matching/tick	12.5% USDC fee rebate to validators, BulkSOL ecosystem binding
Drift Protocol	Fully on-chain L1	Solana L1 smart contracts	On-chain cross-margin, supports 40+ collaterals	V3 slot fill acceleration and Swift protocol (within 400ms)	Swift off-chain signature broadcast pre-notifies market makers, Jito BAM partnership
Jupiter Perps	None (oracle pricing)	Solana L1 smart contracts	JLP pool acts as counterparty; liquidation depends on Pyth oracle price	Extremely low (executes directly against oracle)	JLP liquidity pool model, protects LPs via Price Impact
Pacifica	Centralized off-chain	Solana L1 smart contracts	Off-chain computation for liquidation and margin risk	Microsecond-level matching latency (comparable to CEX)	Efficient off-chain matching, points-based volume farming mechanism, future app-chain
Bullet	Dedicated off-chain sequencer	Solana L1 deposit/withdrawal custody	Off-chain computation, asynchronous settlement and liquidation	Microsecond-level matching latency, block time positioning ~400ms	Reborn from the former Zeta, network expansion (ZK Rollup style)
JTX	L1 native (via Phoenix)	Solana L1 smart contracts	Inherits Phoenix on-chain risk controls, supports CEX-grade TWAP/SL/TP	Extremely low (CEX-grade millisecond notification and execution)	80/20 fee distribution, genuine on-chain order book, CEX price comparison mechanism

5. Solana's Long-Term Advantage: What HyperLiquid Cannot Do

There is only one eternal secret to business competition: attack the weakness that arises from your competitor's strength.

In pursuit of extreme performance, Hyperliquid abandoned a general-purpose smart contract execution environment — Hypercore cannot natively run arbitrary DeFi logic. This means Perps on HyperLiquid are isolated: they cannot be composed with lending protocols (e.g., using a collateral position from Aave/Kamino to directly open a Perp position), cannot be referenced by on-chain options protocols as a price source, and cannot serve as a DeFi Lego block. Although HyperEVM attempts to bridge this gap, there remains a fundamental disparity in the depth of native composability compared to Solana.

Long-tail/RWA Perp Pool Model Advantage: An order book always faces the "chicken or egg" dilemma; market makers are only profitable when there is sufficient taker flow, while traders are only willing to enter when there is sufficient liquidity. Breaking this deadlock requires massive market maker incentives (token rewards, fee rebates, etc.). However, RWA assets are characterized by fragmented liquidity and limited market maker coverage, making the oracle-LP pool model more suitable than CLOB for such scenarios.

DeFi Composability Flywheel: When Solana's Perp protocols can read the on-chain state of Kamino, Jupiter, and Orca, and be invoked by on-chain options, structured products, and automated strategies, the depth of its ecosystem will surpass any single application chain. Jupiter's JLP was once the largest source of carry trade and yield on Solana, a capability that a single-purpose Perp chain does not possess.

Spot-Perp Ecosystem Synergy creates a unique network effect. Solana already dominates on-chain stock trading volume. Combined with cross-chain asset gateways like Sunrise, it can rapidly provide Solana-side liquidity after the TGE of new assets on Monad, Coinbase, and others. Terminals like JTX achieve a unified spot and Perp user experience, making it easier to convert massive retail Memecoin traffic into derivatives participation through a simplified UI, sharing infrastructure to reduce cold-start friction.

These advantages point to Solana's long-term positioning in the dimensions of "composability + long-tail/RWA + retail access," rather than simply replicating Hyperliquid's specialized CLOB model.

In the past 24 hours, the total network contract liquidation was US$588 million, mainly due to the short position

Popular Articles