Solana DEX volume crossed $143 billion in a single month in late 2025. Over 70% of that volume flows through Jupiter. And increasingly, when Jupiter routes your swap, it lands on a venue you’ve never heard of — a proprietary AMM with no UI, no LPs, and no token.

This is the most important structural shift in Solana DeFi right now, and most people are still thinking in terms of Raydium vs. Orca.

Three models, one routing layer

Every spot DEX on Solana falls into one of three categories:

Model	Pricing	Liquidity	Examples
CLOB	On-chain order book	Makers	Phoenix, Manifest, OpenBook v2
AMM / CLMM	Passive curve (xy=k, CL ranges)	LP deposits	Orca, Raydium, Meteora
Prop AMM	Keeper-updated curve	Protocol-owned vault	HumidiFi, Obric, ZeroFi

Jupiter queries all three. It doesn’t care about your ideology — it cares about which venue returns the best quote for 4.2 SOL → USDC right now.

CLOBs: the obvious model that never scaled

A central limit order book is how every serious exchange works — Binance, NYSE, CME. Makers post limit orders, takers lift them. Price discovery happens through the book.

On Solana, Phoenix is the cleanest implementation. Fully on-chain order matching, cranked settlement, ~100K–300K compute units per order operation. It works. Manifest — the spiritual successor to the Serum → OpenBook lineage — is the first formally verified DEX on Solana. Zero trading fees, and a global order system where the same tokens can back orders across multiple markets without locking capital. It already captures ~7% of Solana’s total DEX volume and 24% of stablecoin swaps. OpenBook v2 remains the permissionless default.

The problem is that on-chain CLOBs are expensive per operation and require active market makers posting and canceling orders constantly. Each order placement, amendment, and cancellation is a transaction. On a 400ms block time, you’re competing for blockspace just to maintain quotes. Phoenix did around $8.6B in Q1 2025 — real volume, but a fraction of what AMMs handle on spot pairs.

CLOBs also have limited composability with aggregators. A Jupiter route that touches a CLOB order book has to consume whatever resting liquidity exists at that moment. There’s no guarantee of fill, and partial fills create routing complexity.

AMMs and CLMMs: the incumbents being bled dry

Passive liquidity pools dominate Solana by raw volume. Raydium crossed $1 trillion cumulative in July 2025, running both standard constant-product pools (CPMM) and concentrated liquidity (CLMM). Orca’s Whirlpools pioneered tick-based concentrated liquidity on Solana. Meteora’s DLMM uses discrete bins instead of ticks for the same purpose — letting LPs concentrate capital in specific price ranges.

The numbers look healthy:

Protocol	2025 Volume (est.)	Model
Raydium	~$353B	CPMM + CLMM
Meteora	~$114B	DLMM (bin-based CL)
Orca	~$104B	Whirlpools (tick-based CL)

But there’s a structural problem eating these protocols from the inside: LVR.

The LVR problem

Loss-Versus-Rebalancing, formalized by Milionis et al. in 2022, quantifies what passive LPs actually lose. The intuition: a passive AMM always trades at stale prices. When ETH moves from $2,000 to $2,010 on Binance, the AMM still offers $2,000 until an arbitrageur corrects it — pocketing the $10 difference.

For a constant-function AMM (xy=k), the instantaneous LVR rate is:

LVR rate = σ² / 8  (per unit time, as fraction of pool value)

Where σ is the annualized volatility of the risky asset. For SOL at ~60% annualized vol, that’s roughly:

0.60² / 8 = 0.045 = 4.5% per year

This means passive SOL/USDC LPs lose ~4.5% of their position annually to arbitrageurs, before fees. On a $100M pool, that’s $4.5M/year extracted by bots — value that flows to arbitrageurs, not to LPs.

Concentrated liquidity (Orca Whirlpools, Raydium CLMM, Meteora DLMM) makes this worse, not better. Concentrating liquidity increases capital efficiency and fee capture, but it also increases LVR exposure proportionally. You’re earning more fees on a narrower range, but you’re getting arbed harder on that same range.

The result: on major pairs like SOL/USDC, passive LPs frequently lose money net of fees. The fee APY looks attractive until you subtract the LVR.

Prop AMMs: how professional market makers took over

Proprietary AMMs are the response to LVR. Instead of a passive curve that waits to be arbitraged, a prop AMM has a keeper — an off-chain engine run by a professional trading firm — that continuously updates the on-chain pricing curve to match external markets.

The architecture is simple:

1. On-chain program holds a vault with inventory (e.g., SOL + USDC). No external LPs — the trading firm owns the capital.
2. Off-chain keeper monitors Binance, Coinbase, other DEXes, and pushes oracle/price updates to the on-chain program.
3. Jupiter integration — Jupiter queries the prop AMM during routing. If the prop AMM’s quote is best, the swap executes against its vault.

The critical technical detail is compute efficiency. HumidiFi’s oracle update instruction costs ~143 CU. A typical Jupiter swap costs ~150,000 CU. This 1,000x efficiency gap means prop AMMs can update quotes multiple times per second at negligible cost, while keeping their Jito tips low enough to maintain priority without burning capital.

Because the keeper knows the real-time price on Binance, it can quote a spread of 1-3 bps on SOL/USDC — far tighter than a passive AMM’s effective spread after LVR. The keeper doesn’t get arbitraged because it is the arbitrageur: it’s already pricing at the global best price.

The numbers don’t lie

By late 2025, prop AMMs captured over 50% of Solana’s spot DEX volume on major pairs. HumidiFi alone hit 62.4% of Jupiter-routed volume at its peak. These are venues with no public UI, no token, no community — just professional trading firms providing liquidity through Jupiter’s routing.

For context on how dependent prop AMMs are on Jupiter: 99.2% of GoonFi’s volume, 97.3% of ZeroFi’s, and 92.5% of Obric’s came through aggregator routing.

Lifinity — the original oracle-based AMM on Solana, launched in January 2022 — shut down in November 2025. It proved the concept but couldn’t compete with the next generation of prop AMMs backed by professional trading firms with superior execution infrastructure.

Jupiter is the real exchange

This is the part most people miss. Jupiter isn’t just an aggregator — it’s the de facto exchange for Solana. With 93.6% of aggregator market share and over 70% of all Solana DEX volume flowing through its routing, Jupiter is the venue. Everything downstream — Raydium pools, Orca Whirlpools, HumidiFi vaults — is a liquidity source.

Jupiter Ultra V3 makes this explicit. It combines traditional AMM routing with an RFQ (request-for-quote) system where market makers can bid directly on user order flow. The line between “aggregator” and “exchange” is gone.

For a prop AMM, Jupiter integration is the only distribution channel that matters. You don’t need a website. You don’t need a token. You need a single integration with Jupiter’s routing engine, and you instantly access the majority of Solana’s swap volume.

DFlow, a competing aggregator that routes 98% of SOL-stablecoin volumes to prop AMMs (vs. ~80% for Jupiter), has demonstrated that aggregators optimized for prop AMM routing can achieve better execution quality on major pairs. The aggregator layer is where the real competition is happening.

What this means

For LPs: Passive LP-ing on major pairs is a losing trade unless you’re actively managing positions with off-chain tooling. The days of depositing SOL/USDC into a pool and collecting yield are over for liquid pairs. Passive AMMs still work well for long-tail pairs (memecoins, new tokens) where there are no prop AMM vaults — but on anything with a Binance listing, you’re competing against firms that run the same delta-hedging strategies as traditional market makers.

For protocols: If you’re building a DEX, you’re building a liquidity source for Jupiter. The user-facing product is Jupiter’s swap interface. Your protocol’s job is to return competitive quotes. Passive AMMs compete on breadth (thousands of pairs). Prop AMMs compete on execution quality (tight spreads on major pairs). CLOBs compete on price discovery and transparency for pairs where traders want visible depth.

For Solana: The prop AMM model is uniquely enabled by Solana’s architecture. Sub-second block times, cheap transactions, Jito’s block-builder auction for priority ordering, and low CU costs for oracle updates create an environment where professional market makers can operate on-chain with near-CEX performance. This doesn’t work on Ethereum L1 (12-second blocks, expensive gas) and works poorly on most L2s (centralized sequencers introduce MEV risk).

The irony is that DeFi’s original promise — permissionless, decentralized market making by anyone — is being replaced by the same firms that make markets on Binance, just using Solana as the settlement layer. Whether that’s a pragmatic evolution or a betrayal of the vision depends on what you think DeFi is for.