The ad says passive income. The data says you’re the exit liquidity for three firms in Amsterdam.
There is a version of this pitch you have seen a hundred times. Spin up a trading bot, plug in an API key, let it run while you sleep, wake up richer. The strategy is always some flavor of the same idea: markets are inefficient, prices drift apart across venues, and a fast enough program can harvest the difference forever. Arbitrage. Free money. Passive.
It is a beautiful story, and it is largely dead on arrival for anyone reading this. Not because the inefficiencies don’t exist — they do, and they’re enormous in volume — but because the money that falls out of them has already been captured, industrialized, and centralized by a tiny number of firms that won the race before you’d heard there was one. Someone deploying a bot in 2026 is not entering an open field. They’re entering a race that roughly three firms already won, and buying a ticket to be their counterparty.
Let me show you the numbers, because the numbers are the whole argument.
Huge volume is not the same as available profit
The headline strategy in crypto is cross-venue arbitrage: a price is cheaper on a centralized exchange than on a decentralized one (or vice versa), so a bot buys low on one and sells high on the other. On Ethereum this shows up as “non-atomic” arbitrage — the two legs don’t settle in the same transaction, so someone has to carry inventory and eat the risk in between.
The volume is staggering. From September 2022 through October 2023, researchers identified roughly $132 billion in non-atomic arbitrage trades across the five largest decentralized exchanges — nearly 30% of all volume on those venues over that stretch. [Heimbach et al., IEEE S&P 2024] If you stop reading there, you conclude there is a river of money flowing past and all you need is a bucket.
But volume is the trap. A separate, longer study tracked the realized extracted value — the money actually kept after costs — from these CEX-DEX arbitrages. Over 19 months (August 2023 to March 2025), across 7.2 million individual arbitrages moving $241.7 billion in volume, the total value extracted was $233.8 million. [Wu et al., AFT 2025] That’s a margin of roughly one tenth of one percent. The river is real. The bucket comes back nearly empty, because ten thousand other buckets already went in ahead of yours and the water is priced to the last drop.
This is the first thing the ads never tell you: in a mature, competitive arbitrage market, volume dominance vastly overstates the money available. Thirty percent of DEX flow sounds like an ocean. The profit inside it, spread across millions of trades and years, is a puddle — and, as we’ll see, it isn’t even a puddle you’re allowed to drink from.
The edge is captured, and the gate is closing
Here is where the passive-income story doesn’t just weaken — it inverts. That $233.8 million in extracted value was not spread across a healthy crowd of independent bot operators. Three firms — Wintermute, SCP, and Kayle — captured $170.8 million of it, about 73%. By the first quarter of 2025 those same three were taking roughly 90% of all extracted value. [Wu et al., AFT 2025] Wintermute alone pulled $71.4 million; SCP another $71.1 million.
And the crowd is thinning, not growing. The number of active, labeled searchers in this game fell from 23 to 14 to 11 over the study window. The volume side tells the same story: in the earlier Ethereum data, just 11 searchers accounted for over 80% of all non-atomic arbitrage volume, and the top two alone — running bots with names like beaversearcher1 and jumpsearcher — controlled about 49%. [Heimbach et al., IEEE S&P 2024]
The researchers are not subtle about what this means. Quoting Wu et al. directly: “this further centralization significantly elevates barriers to entry, discouraging market participation by smaller entities and fostering economies of scale.” [Wu et al., AFT 2025]
Read that again. This is not a blogger being cynical. This is a peer-reviewed paper stating, in the flat language of academic finance, that the door is closing on you specifically. The barriers they name are exactly the ones a retail bot cannot buy at the App Store: deep inventory capital, low-latency infrastructure sitting next to the venues, the ability to eat inventory risk across venues, uncertainty over whether your transaction even makes it into the block, and a direct pipeline between the searcher and the party that builds the block. That last one — searcher-builder integration — is a structural relationship, not a setting you toggle in a config file.
So when the marketing copy says passive income, the accurate translation is: you are volunteering to be the marginal, slower, undercapitalized counterparty to Wintermute. You are the exit liquidity.
”But flash loans remove the capital barrier!”
The sharper pitch has an answer for the capital problem. In some strategies — liquidation MEV, where a bot repays an undercollateralized loan and seizes the collateral at a discount — you can borrow the entire principal inside a single transaction with a flash loan and pay it back in the same breath. No capital required. The playing field is level.
Except it isn’t, and the reason is instructive. When flash loans neutralize capital as a barrier, the contest doesn’t disappear — it just moves to the one dimension you still can’t win: getting your transaction included and ordered first. That comes down to who bids the highest priority fee and who is fastest to the block, a dynamic researchers documented as priority gas auctions. [Daian et al., PGA] The economically important sentence: the contract logic is not your competitive advantage. Everyone can see the profitable liquidation. Everyone can write the same fifteen lines of Solidity. The trade goes to whoever pays the most gas fastest — which loops right back to latency and infrastructure, the things the incumbents already own. Remove capital as the moat and you simply expose the deeper moat underneath it.
The one bot trade that provably paid — and its trap
To be fair, and to keep this honest: there is a category of trade that unambiguously worked, and it’s worth understanding precisely because it proves the rule rather than breaking it.
In March 2023, Silicon Valley Bank failed. Circle held a chunk of USDC’s reserves there, and the market panicked that the stablecoin was no longer fully backed. USDC broke its dollar peg and fell to about $0.87 — roughly 13% below where it’s supposed to sit. [Chainalysis, March 2023] Traders who believed the peg would be restored bought discounted USDC on Curve and waited. It re-pegged within days. One firm, Folkvang, and an associated wallet booked around $16.5 million in profit across 59 swaps. [Chainalysis, March 2023] A real bot-executed arbitrage, real money, no ambiguity.
Now look at what that trade actually required. You had to be fast (the discount closed in days), you had to be already capitalized (you can’t flash-loan a multi-day directional bet), and you had to deploy both during a rare, genuinely terrifying tail event where the correct read was that a systemic bank failure would not take the stablecoin down with it. That is not a strategy. That is a coin-flip on the end of the world, made by people who were already positioned to act on it in minutes.
And the symmetric bet — the same shape, run into a different outcome — was ruin. Ten months earlier, TerraUSD lost its peg and the “buy the discount, bet on the re-peg” trade led straight off a cliff: UST and its sister token LUNA went to zero, wiping out roughly $60 billion in days. The traders who “bought the dip” on that depeg didn’t collect an arbitrage; they were annihilated. Same setup. Opposite result. The depeg trade pays like a lottery and loses like a margin call, and knowing which it will be in the moment is precisely the edge you don’t have.
Where the retail bot platforms actually sit
None of this is a knock on the tools. Hummingbot, Freqtrade, 3Commas and the rest are genuinely capable software, and they do exactly what they say. But notice what they are: they are the build-on layer. They hand you a strategy framework, connectors, a backtester. What they cannot hand you is co-located low-latency infrastructure, a nine-figure inventory book to absorb risk across venues, or a private relationship with the entity assembling the block. They put you at the starting line of a race — and the race, per the data above, is being won at the top of the block by firms operating at a scale and speed that a retail deployment cannot approach.
Giving retail the tools was never the same as giving retail the edge. The edge lives in latency, capital, and integration, and those three have consolidated into fewer hands every quarter the researchers measured.
The honest bottom line
Strip the story down and here is what the data supports, and not one inch more:
- The inefficiencies are real and the volume is gigantic — but the capturable profit is razor-thin (~0.1% of volume) and shrinking as competition prices it out.
- That thin profit is overwhelmingly captured by a handful of firms — three took ~73%, rising toward 90%, while the field of participants contracts.
- The barriers are structural, not skill-based: capital, latency, infrastructure, block-inclusion certainty, and builder integration. Peer-reviewed work says outright that these barriers are rising and are meant to discourage exactly the small entrant the ads are recruiting.
- The strategies that provably paid required being fast and already-capitalized and right during rare tail events — and their mirror images were financial death.
Could you run a bot in 2026 and make money? On the margins, occasionally, in niches the whales haven’t bothered to fully drain — sure, the way you can make money counting cards until the casino notices. But the “spin up a bot and collect passive income” promise is a story told to people who will supply the volume, not capture the value. The edge has a name, an address, and a co-located server. It is not you.
The market isn’t inefficient because nobody’s looking. It’s efficient because three firms in Amsterdam and their machines never stop.
Sources & Methodology
Every quantitative claim above traces to a primary source. Figures are quoted as stated in the original research; no numbers were estimated or interpolated.
- Non-atomic arbitrage volume & concentration — Heimbach, Pahari, Schertenleib, “Non-Atomic Arbitrage in Decentralized Finance,” IEEE Symposium on Security and Privacy 2024. ($132B volume / ~30% of top-5 DEX flow, Sept 2022–Oct 2023; 11 searchers >80% of volume; top-2 ~49%.) arxiv.org/html/2401.01622v2
- CEX-DEX realized extracted value & firm concentration — Wu et al., “CEX-DEX Arbitrage,” Advances in Financial Technologies (AFT) 2025. ($233.8M extracted / 7.2M arbitrages / $241.7B volume over Aug 2023–Mar 2025; Wintermute + SCP + Kayle = $170.8M ≈ 73%, rising to ~90% by Q1 2025; active searchers 23→14→11; the barriers-to-entry quotation.) arxiv.org/html/2507.13023v1
- Priority gas auctions / latency as the real moat — Daian et al., “Flash Boys 2.0,” on priority gas auctions and transaction-ordering competition. arxiv.org/abs/2305.16468
- March 2023 USDC depeg & the Folkvang re-peg trade — Chainalysis, analysis of the March 2023 USDC / Silicon Valley Bank depeg (USDC to ~$0.87; Folkvang and associated wallet ~$16.5M across 59 swaps). chainalysis.com
- TerraUSD / LUNA collapse — the May 2022 UST depeg destroyed roughly $60B; contemporaneous reporting via CoinDesk and others.
Nothing here is financial advice — it’s the opposite of a pitch. If you want to see markets priced honestly, with the data out in the open and nothing sold to slant it, that’s why MarketCrystal is free.