JIT Liquidity: When Capital Shows Up for One Block

A pit crew, not a parking lot

I keep getting stuck on the same mental image. A Formula 1 pit crew doesn't live on the track. They wait in the box, the car comes screaming in, four tires go on in seconds, and they're gone before the next lap. They never owned the asphalt. They were there for one moment, did one job, and walked away with the paycheck.

That's the cleanest analogy I've found for JIT liquidity. Traditional AMM liquidity providers behave like long-term lot tenants — they park their tokens in a pool, accept the price exposure, and hope fees make up for it. JIT liquidity providers behave like the pit crew. They mint liquidity right before a big swap, let the swap rip through it, and burn the position immediately after, all inside a single block. They never own the price exposure. They never wear the impermanent loss. They just collect the fee from that one swap and go.

I keep coming back to this because the more I read into it, the more I think it explains something fundamental about where MEV is heading on Solana — and why the chain looks the way it does in 2026.

The original idea, born on Ethereum

The canonical version of JIT liquidity grew out of Uniswap V3's concentrated liquidity model. Concentrated liquidity let LPs aim their capital at a narrow price range instead of smearing it across the whole curve. A JIT provider takes that idea to the extreme: aim all your capital at a single price band, for a single block, around a single trade.

One DEX research glossary I keep coming back to spells out the loop in four steps: observe a large pending swap, mint concentrated liquidity right before it executes, capture the trading fee as the swap routes through your position, and burn the position as soon as the swap is done — "all within the same block," per that glossary. Because the LP is in and out within milliseconds, price barely moves, so the impermanent loss you'd normally eat over days or weeks just doesn't exist. The fee is effectively risk-free — minus the cost of the infrastructure and the gas to do it.

Uniswap's own blog put real numbers on how niche this was on Ethereum. Between May 2021 and July 2022, the protocol counted 8,287 JIT transactions attempted, supplying around $2 billion in total liquidity, per the Uniswap post. Sounds enormous until you read the next line: that activity filled "~0.3% of all liquidity demand" against about $600 billion in trading volume, also from the Uniswap post. And it gets stranger — the same post reports that "over 95% of JIT liquidity was supplied by one single account" and that fewer than 20 addresses had ever attempted the strategy.

When I first read that, I had to stop and reread it. The entire ecosystem of Uniswap V3 JIT — the strategy everyone talks about, the strategy academic papers write about, the strategy that scares passive LPs — almost all of it was one wallet, on one chain, for one stretch of time. The original JIT wasn't a movement. It was basically one professional shop running a very specific play in a very specific pool. The same blog post notes that a single USDC-WETH 5bps pool accounted for over half of all JIT liquidity supplied. Pit crew at one corner of one track.

Why JIT stayed niche on Ethereum

The academics put a number on the friction that kept everyone else out. A peer-reviewed paper from a London-based university research group, "Demystifying Just-in-Time (JIT) Liquidity Attacks on Uniswap V3," found that a JIT provider needs capital equal to roughly 269 times the swap volume on average, versus only about 6 times the swap volume for a sandwich attack. The same paper reports average JIT ROI of about 0.007%, while sandwich attackers earn meaningfully more.

So on Ethereum, the trade was simple: enormous balance sheet required, microscopic return per swap, and a single dominant actor already collecting most of what's there. For anyone starting from zero — which is exactly where I am — this strategy on this chain is basically a closed shop. You can't out-capitalize a market maker with a $1 million balance. You can't out-capitalize them with $10 million. The economics demand size you don't have.

But the academic paper also flagged something quietly important: the same JIT activity that's brutal for passive LPs is actually good for traders. The paper reports that takers got execution prices 0.139% better on average when JIT was involved. So the trade is structurally weird: it hurts the lazy capital sitting in the pool, while helping the active capital moving through it. That tension is the whole story of why JIT keeps showing up no matter how much people complain about it.

Solana lacks a public mempool — so how does JIT even work here?

When I started studying this for Solana, the first wall I hit was the mempool problem. The classic Ethereum-style JIT relies on watching a pending transaction queue, noticing a big swap coming, and inserting yourself in front of it. Solana doesn't work that way. There's no in-protocol mempool. Transactions go straight to validators, and there's no public queue to snipe from. One infrastructure provider's writeup describes this as part of why "Solana's continuous block production and lack of an in-protocol mempool changes the default behavior and social dynamics of the chain," per a provider's writeup.

Jito Labs ran a private mempool for a while, which did enable some sandwich-style behavior. Then in March 2024, Jito Labs shut it down, citing growing concerns about MEV-driven harm to the ecosystem. CoinDesk reported at the time that the team viewed sandwich attacks and similar predation as a risk to ecosystem health. That single move closed the easiest door for mempool-style JIT on Solana.

So for a while, the obvious answer was: classic JIT just doesn't run on this chain. But what actually happened is much more interesting. Solana didn't get JIT, exactly — Solana got something that walks like JIT, breathes like JIT, and is arguably more capital-efficient than JIT ever was on Ethereum. The protocols doing this go by the name proprietary AMMs, or prop AMMs, sometimes "dark AMMs."

How prop AMMs do JIT without a mempool

The trick is to stop thinking about JIT as "watch the queue, snipe the trade," and start thinking about it as "have your pricing ready exactly when the trade lands."

A prop AMM runs an off-chain quoting engine — basically a market maker's brain — that knows the fair price of an asset at any given millisecond. When a major aggregator shops a user's order around, the prop AMM submits a transaction that updates a tiny on-chain price value, and then the user's swap executes against that freshly updated value in the same atomic operation. The Solana Foundation's "Understanding Proprietary AMMs" describes how PropAMMs "use predictive price feeds to update minimal data onchain." The data being updated is small — a single 8-byte u64 value, per the Solana Foundation post, versus 80 to 320 bytes for traditional orderbook updates.

The cost difference is even more striking. A provider's writeup reports that one of the more active prop AMMs runs an oracle update of 143 compute units, costing about 9,998 lamports — roughly $0.001784. A typical aggregator-routed swap, by contrast, uses around 150,000 CUs, per the same provider. The pricing refresh is something like a thousand times cheaper than the swap that uses it. That's the magic. Refreshing the quote is so cheap that you can do it before every single swap that comes your way, for every single asset pair, as a matter of routine.

Functionally this is still JIT. The liquidity isn't a permanent pool sitting at a wrong price waiting to be picked off. It's a fresh quote that arrives just in time, the swap fills against it, and the market maker's inventory rebalances. Same pit crew, same one-block window, different mechanism — no mempool sniping required, just very fast, very cheap on-chain re-quoting.

Reading the prop AMM scoreboard

The data on how dominant this has become is, frankly, hard to believe if you're still picturing DEX as "Uniswap-style passive pools." One research firm's report finds that prop AMMs now account for roughly 65% of on-chain trading volume on Solana, surpassing traditional CLAMMs. Another research firm's analysis puts it at over 80% market share in core SOL-stablecoin pairs, with roughly 40% of Solana DEX swap volume routing through aggregators.

The daily volume range is striking on its own. According to a provider's writeup, daily volumes consistently exceeded $1 billion over a 60-day window, with peaks near $2 billion and slow days down to several hundred million, and SOL/USDC averaging around $1.5 billion per day over a three-month period. One publication pegged the weekly total at several billion in trading volume for these so-called dark AMMs, representing roughly 30% of all blockchain trading. One protocol, HumidiFi, was processing volume on the order of $3 billion in a single week per the same publication — somewhere around 15% of Solana's entire trading volume on its own.

What happened to the traditional AMMs as this took over? Per a provider's before/after comparison, the major classic AMMs lost meaningful share in aggregator routing — venues that previously commanded 30–60%+ of aggregator volume compressed down into the low 20% range. That's not a small reshuffle — that's an entire layer of the DEX stack getting eaten from underneath. Within a couple of years, the chain went from "Uniswap-style AMMs are the market" to "professional market makers running per-block JIT are the market."

And reading those numbers, I keep coming back to the original Uniswap data point — the Ethereum version of JIT was one wallet, less than a percent of volume, fewer than 20 addresses ever trying it. The Solana version of the same idea, executed differently, is now most of the volume on the chain. Same underlying insight — capital that shows up for one block beats capital that sleeps for a year — completely different infrastructure outcome.

What the spreads actually say

The market share story is impressive on its own, but it would be hollow if execution quality was bad. It isn't.

One research firm's report includes a tight breakdown of effective spreads in basis points across protocols and trade sizes. For SOL-USDC, the report shows the dominant prop AMM quoting 0.4 to 1.6 basis points across most trade sizes, widening only to about 5 bps at the $1 million size. Two other prop AMMs come in at roughly 1.3 to 3 bps. Traditional AMMs run 5 to 9 bps for trades between $1,000 and $50,000, and they widen further above $100,000.

The BTC-USDC and TRUMP-USDC comparisons in the same report tell a similar story. The dominant prop AMM quotes 2 to 4 bps on BTC-USDC with minimal size sensitivity. A traditional concentrated-liquidity AMM in the report runs about 8 bps on small trades, climbing to 9 to 10 bps above $50,000.

Another research firm's analysis puts a TradFi comparison on it that I've been thinking about a lot. Per that firm, sub-$100K trades on Solana's prop AMMs "match TradFi execution quality," noting that S&P 500 names typically range from 0.5 to 8 bps. In other words, retail-sized trades on a chain that didn't exist a decade ago are now executing at spreads competitive with the most liquid equities market in the world. That's not a thing you can hand-wave away.

What made it click for me is the size invariance. On a classic AMM, the spread grows with trade size because the curve simply gives a worse price as you push along it. On a prop AMM, the quote is set by an off-chain model that already knows where the market is, so it can quote pretty much the same spread on a $1,000 trade as on a $100,000 trade. That's the Costco effect, basically — once you've already built the warehouse and the inventory system, the marginal cost of moving one more pallet is essentially flat. The model is the warehouse. The on-chain quote is the pallet.

The hidden cost

None of this is free, and the criticism is real. Every current Solana prop AMM is closed-source. Per the Solana Foundation's writeup, none of them open up the quoting logic, and most don't accept community liquidity provision. The capital comes entirely from the operator. One publication quotes a market-making executive with the line that's been stuck in my head: "Better execution is winning out over transparency."

That's the tradeoff in one sentence. Users get pricing that beats the old open AMMs by a wide margin, but they're trading against a system whose internals are opaque. There's no way to LP into one of these venues the way you can LP into a classic AMM. There's no community pool. The fees go to the operator, period. The role of "liquidity provider" — the role that made DeFi a thing — is functionally being deprecated for these flows.

The same publication also surfaces a structural concern: per its reporting, dark AMMs "lack public websites, don't allow community liquidity provision, and rely solely on creator-provided liquidity." That's not a security claim, it's a participation claim. Open AMMs were supposed to let any random person rent out their capital. Prop AMMs replace that with professional shops running closed software. It's the difference between a community garden and a private greenhouse — the produce coming out of the greenhouse might be better, but you're not allowed in.

There's also a worry about leader-dependent inclusion. Because prop AMMs need their oracle update to land in the same block as the swap, they depend on block leaders cooperating. The Solana Foundation post acknowledges this, noting that "some validators prioritize fair ordering while others may censor competitor price updates or delay updates to extract MEV," per the Foundation. That's an open question for the chain's long-term neutrality, and I haven't seen it cleanly resolved anywhere.

A founder quoted in the same publication frames where this is heading with a line I find almost too direct: "Probably only a matter of time before active liquidity entirely outcompetes passive liquidity." The Uniswap-style model where you deposit and forget was a beautiful idea, but if the data above is right, it's losing decisively to the model where professionals quote in real time and clear their books every block.

What I'm taking away as a solo developer

I have to be honest with myself about what these numbers mean for somebody starting from zero. The Ethereum-style JIT play is closed to me. The capital requirement that the academic paper estimated — about 269 times the swap size — assumes I can deploy seven or eight figures into a pool for the duration of one swap. That's not happening on a personal budget. The 0.007% ROI per swap means even if I could, I'd need to be running at industrial scale for it to clear infrastructure costs.

The Solana prop AMM play is, if anything, even more out of reach as an operator. Building the off-chain quoting model is a real piece of trading infrastructure — one of the prop AMMs in the comparison was built by a well-known professional market-making firm, per one publication. Another was built by the same team behind a leading on-chain orderbook venue, per one provider's reporting. The barrier to entry isn't really code anymore — it's a market maker's pricing model, the inventory management around it, and the relationships with aggregators to get flow. None of those scale down to a kitchen-table operation.

But understanding this changes how I think about my own search space. When the dominant venues on Solana are running per-block JIT against professionally-priced inventory, the inefficiencies an amateur bot can hunt aren't going to live there. They're going to live around the edges — in the tail, in the long-tail tokens these prop AMMs don't quote, in the moments when aggregator routing is suboptimal, in the small pools where professional MMs haven't bothered to deploy. The food trucks, basically, parked outside the territory the chain restaurants have already locked down.

I also keep coming back to a quieter observation. There's a parallel design lineage where JIT is treated as a feature rather than an exploit. One protocol's documentation describes a Dutch auction JIT mechanism where market orders route through a short auction and competing makers fill them, with the AMM acting as a backstop. Another protocol's documentation describes a similar inversion, where the design "naturally incentivises liquidity providers to front-run each other to the benefit of the user." Both are designed JIT — JIT baked into the protocol with open participation. Whether that model wins out over closed prop AMMs, or whether the two coexist, feels like one of the genuinely open questions in this space right now.

For today, all I can do is sit with how strange the shape of this market is. Most of the trading volume on one of the largest chains in crypto is now executed against liquidity that exists for one block at a time, priced by off-chain models I cannot see, run by operators I do not know, with closed code I cannot audit. And the user experience is better than it has ever been. The pit crew has eaten the parking lot. I'm still figuring out where, exactly, somebody starting from zero fits into that picture.

Key Takeaways

• JIT liquidity is a one-block trade, not a long-term LP position. The provider mints concentrated liquidity right before a swap, captures the fee, and burns the position before any meaningful price movement can hit them.
• The Ethereum version stayed niche. Per Uniswap, over 95% of JIT volume between May 2021 and July 2022 came from a single account, and fewer than 20 addresses ever tried the strategy.
• Solana lacks a public mempool, so classic queue-sniping JIT doesn't work here. Prop AMMs instead refresh on-chain quotes immediately before each swap, achieving the same "just in time" effect through cheap oracle updates rather than mempool front-running.
• Prop AMMs now dominate Solana DEX volume. One research firm puts them at roughly 65% of on-chain trading volume, with the leading venue processing several billion in volume per week per one publication.
• Execution quality on prop AMMs rivals TradFi. Per multiple research firms, effective spreads on SOL-USDC at the dominant prop AMM sit in the 0.4 to 1.6 bps range — competitive with S&P 500 names, which run 0.5 to 8 bps.
• The cost is transparency and participation. All current Solana prop AMMs are closed-source and accept no community LP. Better execution has, in one executive's words to a publication, won out over openness.