Backrunning on Solana: After the Wave Passes

I'm Done Trying to Race Ahead

For weeks I think about MEV as a kind of high-stakes race — get there first, push to the front, edge out the other by a hair. But the more I look at the actual data on Solana, the more I realize the strategy I keep gravitating toward isn't a race to the front at all. It's something more like surfing. Wait for someone bigger than you to push the water around, then ride the wake they leave behind.

That's backrunning. And I want to spend this entry actually pinning down what it is, why people argue it's the only acceptable face of MEV, and what makes it so hard in practice that the people doing it well are paying record-breaking tips for a single bundle.

The Definition I Keep Coming Back To

The textbook answer is short. Backrunning is "the execution of a transaction immediately after another, high-value transaction" — typically a large swap — in order to capture the price gap that swap leaves behind, according to CoW Protocol's explainer on the topic. It does not change the original trader's price. It does not push them into a worse fill. It happens after the fact.

That last detail is what separates backrunning from its more notorious cousins. a16z Crypto's research team, in their MEV explainer, defines the three classic MEV strategies cleanly: frontrunning happens before a target transaction, sandwiching wraps the target in two trades, and backrunning is "a third party who spots this... can insert a sell transaction right after Alice's to profit from this price differential." The wording matters. The backrunner isn't a participant in the original trade. They're a janitor showing up after the dance is over and sweeping up what's left on the floor.

CoW Protocol goes further and calls backrunning "the least harmful" of MEV strategies, noting that it "doesn't worsen prices for the original trader but instead captures leftover arbitrage — essentially 'money left on the table' for sophisticated users." That phrase — money left on the table — is the one I keep writing in my notebook. The whole strategy is a bet that big traders aren't going to bother picking up their own change.

Why an AMM Even Leaves Change Behind

To understand why this works, I have to stop thinking of decentralized exchanges like Walmart and start thinking of them more like an old-school self-service kiosk where the price changes only when someone uses the machine. Paradigm's primer on AMM price impact puts it about as bluntly as anyone:

The AMM does not update its price as other markets move around it. The market price only moves as the reserve ratio of the tokens in the pool changes, which happens when someone trades against it.

That single sentence is the engine that makes backrunning possible. Every other venue in the world can have its price drift — and the AMM just sits there, frozen at whatever ratio its last trade left behind. A large buy on the AMM pushes its quoted price up; the rest of the market hasn't moved, so suddenly the AMM is overpriced relative to every other venue holding the same asset. A large sell does the inverse — the AMM is now cheap, and the rest of the world hasn't caught up.

That dislocation is the wake. And the wake doesn't last long. Paradigm's same essay notes that "traders can still expect the price quoted by an AMM to closely track the global market price because of continuous arbitrage." In other words, the very fact that prices stay coherent across DeFi at all is because there is always somebody — usually a bot — willing to ride that wake.

This is the part that reframes the whole strategy for me. Backrunning isn't a parasite on the system. It is the system. Without it, AMM prices would drift away from reality the moment any sizable trade happened, and every casual swapper after that would be transacting at a stale price.

The Wave That Made Everyone Notice

The single event that crystallizes backrunning for me is something a researcher documents in their 2025 Solana MEV report. On January 10, 2024, someone bought roughly $8.9 million worth of DogWifHat (WIF) in a single transaction. That kind of size on a memecoin pool is a tsunami. The AMM's reserves got hammered, the price snapped upward, and the rest of the WIF market hadn't yet caught up.

A backrunner was watching. They wedged a trade in immediately after the buy, swept the now-mispriced WIF on the imbalanced pool, and unwound it elsewhere. According to the report, the searcher walked away with a net profit of 17,442.24 SOL — about $1,794,746 at the time — from one transaction. To win the slot priority, they paid a tip of 890.42 SOL (around $91,621), which the report describes as "one of the largest tips in Jito's history."

Let me sit with the numbers for a second. The original buyer paid roughly nine million dollars for their WIF. The backrunner paid roughly ninety thousand dollars to a validator to be the next person in line. And the backrunner cleared nearly $1.8 million in profit in the time it takes to blink.

The original trader was not harmed. Their swap had already executed at the price the AMM quoted them. What the backrunner captured was the gap between that quote and the rest of the world's price for WIF — a gap that existed for, at most, a few hundred milliseconds. The whole strategy is a bet that you can be the fastest person to notice that gap and the most aggressive person willing to pay for the privilege of closing it.

The Scale of It Is Almost Boring

The WIF event is the headline, but the more I dig into the report, the more the day-to-day picture stops looking like home-run trading and starts looking more like the loose change in the couch cushions of the entire ecosystem.

Over the year covered by the report, the analysts identify more than 90 million successful arbitrage transactions on Solana — they specifically count 90,445,905 — generating a total of $142.8 million in profit, at an average of $1.58 per arbitrage. That last figure is the one that keeps me grounded. A buck fifty-eight per trade. The size of a vending-machine soda. The whole industry, on Solana alone, is built on doing that trade often enough that the buck-fifty-eights stack up to nine figures.

And this is just one chain. An analysis overview of Solana MEV economics estimates the broader MEV economy across all chains at around $7.5 billion in total value. The 2025 report attributes 88.7% of Solana arbitrage profit — $126.7 million of the $142.8 million — to SOL-denominated routes, which tells you something important: most of the work is happening on the most liquid pairs, not on exotic pairs you've never heard of.

The most lucrative single arbitrage identified in the data window cleared $3.7 million. The least lucrative — well, plenty cleared cents. The report cites one example of an Orca/Phoenix route that netted about $0.026 on a 45-USDC swap. The arbitrage world is barbelled: tiny trades all day plus the occasional jackpot when a tsunami like the WIF event hits.

What I Mean When I Say "Backrunning Is Cleaner"

I keep being careful with the word "cleaner." CoW Protocol calls backrunning "the least harmful" of the three classic MEV attacks, and I've been going back and forth on whether that's marketing language or actually accurate. The more I dig, the more I think the answer depends on which counterfactual you compare to.

If you compare backrunning to sandwich attacks, it isn't even close. A sandwich attack works by detecting a pending trade, frontrunning it to inflate the price, letting the victim fill at the worse price, and then backrunning to dump the inventory at a profit. The victim eats the entire spread the attacker created. The 2025 report on Solana sandwich activity is brutal reading — a single sandwich bot operator they identify generated $13.43 million in profit over about a month, conducting an average of 51,600 sandwiches per day with an 88.9% success rate. Memecoin traders, who tolerate high slippage to chase listings, get hit the hardest. Sixteen of the top twenty most-sandwiched tokens in their data started on Pump.fun.

That is harm. Each sandwich is a direct extraction from the victim's wallet. The backrun part of a sandwich is only profitable because the frontrun part artificially pumped the price first.

A pure backrun without a frontrun doesn't do any of that. The original trader chose their size, chose their slippage tolerance, signed their transaction, and got filled at the AMM's quote. Whether some bot trades right after them or the AMM sits dormant for ten seconds until the next organic trade — from the original trader's perspective, those two worlds look identical.

But. There's a real cost backrunning imposes on the network, and it isn't on the victim — it's on everyone else trying to use the chain. CoW Protocol notes that "competition among backrunning bots can lead to negative externalities, including network congestion and increased gas fees as bots spam transactions." The report confirms this is not theoretical: on Solana in April 2024, 75.7% of all non-vote transactions were reverted, the vast majority being failed arbitrage attempts from bots racing each other. Three out of four transactions in the most active month of the year did nothing except waste blockspace.

So when I say backrunning is "clean," I mean it doesn't have a victim in the same way a sandwich does. I don't mean it's free.

The Race Is Not Where I Expected

The part that's been reorganizing my mental model the most this week is what the race actually looks like. I assume for a long time that backrunning is about who can submit a transaction first — get to the gate before the other guy. That model is mostly wrong for Solana.

On Ethereum, the public mempool is the hunting ground. Bots watch pending transactions sitting in the queue and construct backrun trades around the ones that look juicy. The MEV Wiki back-running entry describes the classic version: bots monitor for a new Uniswap pair listing event, submit a buy transaction sized to land immediately after the liquidity-initialization transaction, and unload at market price later. The economic shape of the trade is the same as Solana, but the plumbing is different.

On Solana, the public mempool effectively doesn't exist in the same form. The way most serious backrunning happens is through a bundle auction. A searcher constructs a small group of transactions — sometimes just one, sometimes a few — that must execute atomically, all of them or none. They attach a tip and submit it to a block engine that runs an off-chain auction. The bundle with the most efficient tip wins the right to be sequenced into a leader's slot. The same analysis characterizes it directly: "Blockspace is no longer filled purely by native transaction queues, it's actively auctioned off to the highest-value bundles submitted by searchers."

This is closer to a tip-based bundle submission than it is to a footrace. You don't beat the other searcher by being faster — you beat them by figuring out, in the moment, what your edge is worth and being willing to give the right percentage of it to the validator. According to the report, top arbitrage searchers tend to surrender 50 to 60% of their expected profit in tips. The remaining margin is what the bot keeps. That ratio is not generous. It's competitive. If you tip too little, somebody else wins the slot; if you tip too much, the trade isn't worth doing.

Which means the central skill in backrunning is not coding speed. It's pricing your own edge in real time, on a per-opportunity basis, against opponents whose strategies you cannot directly observe.

What This Asks of a Builder

Reading all of this, I keep trying to translate it into what I'd actually have to be good at to run this strategy. A few things stand out.

First, the latency floor is brutal. The price dislocation a backrunner is chasing might last a few hundred milliseconds before the next bot or the next organic trade closes it. The infrastructure that wins backrunning is co-located close to validators and processes pool state changes the instant they happen — not a polling loop, but a push stream. I keep finding references to top searchers running on bare-metal servers in the same data centers as high-stake validators. This is not a strategy you run from a laptop with a free public RPC.

Second, the math has to be right and it has to be cheap to compute. For every potentially interesting swap that crosses your view, you have to simulate the resulting price impact across the relevant pools, identify whether any arbitrage route closes that gap profitably after fees and tip, decide what to tip, and ship the bundle — all in the time window before someone else does the same. The actual on-chain economics are simple xy=k arithmetic in the simplest case. But across hundreds of pools, with concentrated liquidity ticks and bin-based curves and dynamic fees layered in, the simulation gets ugly fast. Doing it correctly and quickly is most of the engineering challenge.

Third, the strategy is unforgiving of false positives. The data on failed arbitrage spam — 75.7% of non-vote transactions reverted at peak — is what happens when bots fire on weak signals. Every failed attempt costs base fees and validator tips even if no value is captured. The good searchers aren't necessarily the ones firing the most shots. They're the ones firing the shots that hit.

Fourth, even after all that, you still lose most of the profit to the validator. The 50–60% tip ratio described in the report is not a tax on a few transactions. It's the operating equation. A bot that captures $1.58 of theoretical profit per trade keeps maybe sixty cents of it after the tip. The arbitrage economy works at all because there are tens of millions of these trades per year — the 90.4 million figure averages out to roughly a quarter million arbitrages per day across Solana. Scale matters more than per-trade margin.

What I'm Sitting With Right Now

The thing I find most interesting about backrunning, after reading through all of this, is that it sits at this strange spot between extraction and infrastructure. Done badly, it's a spam factory that clogs blockspace and burns fees that everyone else has to pay. Done well, it's the mechanism that keeps DEX prices honest, the reason an AMM quote isn't twenty seconds stale every time you swap. The same activity, with the same code, can look like a parasite or a market-maker depending on whether the bot is winning or losing.

What that means for me — sitting in front of a project that's still trying to make its first profitable trade — is that I have to be honest about which version I'm chasing. The lottery shot, the WIF-style $1.8 million event, is real but it's exactly one observation. The everyday economy is the buck-fifty-eight per trade, repeated until the numbers add up. The infrastructure cost to compete at the second version is high. The probability of catching the first version, from where I'm sitting, is whatever you'd call slightly above zero.

I don't know yet which side of that line I'll land on. Probably neither, in any pure form. But it helps to know that the wave I keep watching for isn't the trade itself — it's the wake it leaves behind.

Key Takeaways

• Backrunning trades after a target transaction, not before or around it. The original trader is not affected; the bot is capturing a price gap left behind on the AMM versus the rest of the market.
• AMM mechanics make backrunning possible. Constant-product pools don't update their prices unless someone trades against them, so any sizable swap creates a dislocation that exists until an arbitrageur closes it — a process Paradigm describes as continuous arbitrage.
• The scale on Solana is enormous but the per-trade profit is tiny. The 2025 report identifies more than 90 million successful arbitrages in one year totaling $142.8 million in profit, at an average of $1.58 each.
• Backrunning is widely considered the least harmful MEV strategy, per CoW Protocol's framing, because no victim eats the spread — but bot competition still produces real externalities, including a peak of 75.7% reverted Solana transactions in April 2024.
• The competitive edge is pricing, not raw speed. Top searchers reportedly pay 50–60% of expected profit in tips to win bundle slots, which means the skill is valuing your own opportunity correctly against opponents you can't see.