What It’s Like Running A Billion-Dollar Market-Neutral Fund In Crypto

Summary

• Evgeny Gokhberg’s core thesis is that DeFi differs from TradFi in exactly one way: the platform is separated from the liquidity provider, and his fund fills the bank’s seat. Uniswap has an interface and matches buyers and sellers “but it has no capital,” so capital provision is outsourced to the market — “we’re not getting paid because DeFi is risky. We’re getting paid because there is economic activity on chain and then we’re the ones making it happen.” He says they are getting 11%–12% consistently in DeFi yield.
• His firm’s research into whether AI makes DeFi uninvestable found that all—or 99% of hacks that actually happened—involved AI exploiting legacy OPSEC vulnerabilities, not AI breaking smart contracts. His metaphor: “AI doesn’t invent an amazing drill that breaks through a metal vault. What AI does, it finds 100 old banks with wooden doors that you can knock out with a kick.” March–April saw a record number of hacks, but dollar losses were normalized at below 2% annualized. Of roughly 60 hacks this year, his team didn’t even know 50 of the platforms; he says this indicates those platforms were probably not large or serious and that the security trend is moving in the right direction.
• The most exciting yield source right now is tokenized stocks, because his yield is “a derivative of the market environment” and there is a bull market in AI stocks. “The same things we used to be doing with memecoins… we’re doing with tokenized NVIDIA” — which puts him in explicit competition with TradFi, even as banks look to lend stablecoins into his vaults and tokenize equity and credit into them.
• Evgeny says agents can replace a junior analyst doing repeatable tasks, but not a senior analyst or expert. Avi agrees that AI makes research easier but creates false positives: CCOI “had a massive ripper” after appearing early on Claude as a data-center downstream play, then “sold off 70% because nobody had actually done the real research.” Avi says a specialist can process what Claude provides and find alpha in the spread. The warning, from Avi, is that AI “doesn’t replace the need for you to think.”
• The edge, he admits, is mostly survival plus a risk framework: he red-flagged Anchor and did not chase its 20% yields while “lots of people were looking at me like I was an idiot” — “and some of those people are no longer around.” He treats DeFi like a high-yield credit book (yield must exceed cost of default) with a probabilistic AAA-to-CCC rating system, and argues retail can’t replicate it: “when people go into DeFi, realistically they go blindly… on the back of a brand that they trust and a CEO who is active on Twitter.”
• On Saylor comparisons to Luna: “Not really. Not at all actually.” Luna was “debt backed by equity and that equity was worth nothing,” whereas Strategy is “basically like a Lombard loan.” Using his illustration, it is an asset worth $100 against say, $30 borrowed; he does not recall the latest figures. A 90% Bitcoin drawdown forcing sales is possible, “but at the current levels of leverage, it doesn’t feel like it’s a risky thing.”
• He’s declining the temptation to expand into TradFi despite crypto’s slow patch, because his edge is on-chain and he’s “more bullish on the market than bearish.” “When I think of… telling them that I can trade Apple better than Citadel — I don’t think I can do that.” The convergence trades — spreads, arbitrage, and DeFi lenders and market makers meeting tokenized real-world assets — are where he wants to sit; the coolest recent DeFi innovation, in his view, is tranching finally being executed in DeFi.

Deep dive

1. EM long/short taught him to work backwards down the supply chain

• Evgeny’s decade in TradFi (a long/short hedge fund, then UBS and Deutsche) meant buying listed Egyptian real-estate companies at “30 cents on the dollar” after the Arab Spring closed the exchange — access came through sell-side brokers because “you had capital.” His contrast with crypto: now “it’s an anon person on Twitter that you have to find a chance to speak to.”
• The China short he’s proudest of: you couldn’t short China directly, so he collected dominoes backwards — Chinese real estate → steel → iron ore → Australian iron-ore companies with leverage, where “if Chinese real estate is down, these guys will lose 90% of their net profit.”
• On diligence, his spectrum: governance is better in developed markets, weaker in EM, “and it basically doesn’t exist in crypto.” Avi recounts a CEO who shut the deck when questioned on a forecast; Evgeny’s response was, “if the CEO throws a temper tantrum, I’m not investing in that company.”

2. AI research replaces the junior analyst — and floods the market with false positives

• Avi’s market-structure worry: everyone now asks Claude for data-center bottleneck plays, so names surface instantly — CCOI “had a massive ripper” then “sold off 70% because nobody had actually done the real research.” His verdict: “there’s almost both more opportunity and less opportunity” in downstream research.
• Evgeny ran the experiment directly — a fully autonomous AI-agent kind of venture fund, tokenized on Base, shipped without raising or advertising, that “goes and makes venture investments on chain.” The lesson: it can replace a junior analyst doing repeatable tasks, “but we don’t feel it’s at the point where we can actually replace an expert.” Avi adds that specialists can process what Claude provides and “find alpha in the spread.”
• Evgeny points to the viral World Monitor, a rebuilt Palantir-style world-monitoring dashboard, to ask whether the underlying data is right. Palantir’s value, he says, comes from the billions spent and people hired to ensure its numbers are correct; AI can narrow the search space and accelerate research by 10x or more, but it cannot replace expert judgment. Avi’s closing warning is that people ask a question and take the answer without processing the information themselves.

3. He didn’t fall in love with crypto — he ran a systematic search for it

• Wanting to “1000x my personal growth” out of a post-Greek-crisis London opportunity set, Avi was deliberate: “I’m always hopeful to fall in love with something but it never happens… therefore you have to make it happen.” A Stanford tech course, then fintech angel checks — “what’s the fastest way to learn is to lose money. What’s the best way to lose money? Is to be an angel investor.”
• His discipline on crypto: no investments for the first year, two evenings a week at meetups — 100 events — until a 2015 Vitalik lecture: “I didn’t understand anything he was saying but it was very clear this is the future.”
• The second aha was DeFi in 2019: for the first time, decentralized apps charged fees, making them real businesses — “you can value it and you can measure it and you can monitor it.” That gave Avi confidence to dive in full-time. Separately, Evgeny says he launched 3x5 five years ago as a market-neutral fund.

4. The yield is bank economics, not a crypto risk premium

• Avi’s challenge — isn’t the yield just compensation for hack and default risk? Evgeny’s rebuttal: the 2021 model of protocols “using their equity as a marketing budget” is “obviously dead because the token market is dead”; his DeFi yield means strategies where the investor takes only smart-contract risk—the platform collapsing from software failure rather than market failure—paid for the roles TradFi reserves for prime brokers and “the Citadels of this world”: lending and market making.
• The structural claim: a bank has an interface, matches borrowers and lenders, and has its own capital; Uniswap has the interface and matching “but it has no capital.” Liquidity provision is outsourced — “we do what banks do in the real world.”
• On “vaults,” his own terminology gripe — “DeFi terminology is just terrible… a word vault doesn’t mean anything” — it’s “an investment box,” a wrapper evolution like fund-to-ETF: more accessible and transparent, but “these features can’t really make a strategy better. They just make explaining and selling that strategy better.”
• The capital base spans several very large global investment banks’ platforms, foundations, endowments, fund of funds, and professional investors; vault depositors are DAOs, whales, and other on-chain users. The business has grown meaningfully over five years “even though the space hasn’t really grown that much.”

5. AI kicks in wooden doors — and the edge is a risk framework plus survival

• The firm’s existential question: “what if DeFi is uninvestable now because Claude, Mistral, and Fable can just destroy everything.” Answer from their research: all—or 99% of hacks that actually happened—involved AI exploiting legacy OPSEC vulnerabilities, not smart-contract breaks — the metal vault holds; AI just finds “100 old banks with wooden doors.” Record hack counts in March–April masked normalized dollar losses below 2% annualized, and 50 of the roughly 60 platforms involved in those hacks weren’t even known to his team.
• Asked why he survived when peers didn’t, an honest non-answer: “I wish someone would tell me the secret sauce, but… it’s just showing up and executing” — plus a probabilistic AAA-to-CCC rating system and a DeFi-as-high-yield-credit frame where yield must exceed the cost of default. Anchor did not pass their RDD and was red-flagged; “if you lose 100% in the second after you deploy, that’s it.”
• On Saylor’s comparison of Strategy with Luna: no corollary — Luna was debt backed by worthless equity; Strategy is a Lombard loan. Evgeny’s illustrative example is an asset worth $100 against say, $30 borrowed; he does not recall the latest numbers. Forced selling in a 90% Bitcoin drawdown? “Yes, of course. But at the current levels of leverage, it doesn’t feel like it’s a risky thing.”
• He’s staying on-chain rather than chasing TradFi strategies — “I don’t think I can trade Apple better than Citadel” — because he takes the view the crypto economy is “likely to accelerate”; the freshest innovation he flags is tranching, “not very novel from a financial engineering standpoint, but definitely novel to DeFi.”