Look, I want to get something out of the way right at the top because I do not want anyone to think I am going soft on this guy.

Sam Bankman-Fried defrauded customers out of billions of dollars. He commingled customer deposits with his hedge fund's prop trading book. He lied about it for years. He was convicted in November 2023 on seven counts including wire fraud, securities fraud and money laundering, and he is currently serving 25 years at a medium-security federal facility in California. That is what he did. That is who he is. None of what follows changes any of that.

But here is the part of the story that has not really gotten the attention it deserves, and the part that has been driving me a little nuts the more I dig into it.

The guy was an outstanding venture capitalist.

Not "decent for a crypto bro" outstanding. Not "got lucky on a few coins" outstanding. Genuinely, freakishly, top-of-the-power-law outstanding. The kind of pattern recognition on early-stage tech that, in a parallel universe where he sourced his capital legally instead of stealing it from his own customers, would have him giving keynote speeches at Sequoia retreats right now.

And the bankruptcy estate, doing exactly what bankruptcy law tells them to do, sold almost all of it. At the bottom. To make customers "whole" in nominal 2022 dollars.

Let's dig into the receipt.

The 30,000 foot view

Between roughly 2020 and the implosion in November 2022, FTX and its sister hedge fund Alameda Research deployed about $4.7 billion in venture investments across hundreds of companies and tokens. The corporate vehicles doing the deploying had names like FTX Ventures Ltd., Clifton Bay Investments LLC, and Maclaurin Investments Ltd. (which used to be called Alameda Ventures until somebody decided that name was too closely associated with the fraud).

This was not careful institutional capital deployment. There were no proper investment committees. Due diligence was minimal-to-nonexistent in most cases. The internal portfolio tracker, known to bankruptcy lawyers as "The Alameda Spreadsheet," cataloged something like 400 separate positions. A lot of them were trash. Hundreds of millions were lit on fire in low-tier altcoins, inflated mining operations, and NFT projects that have since cratered.

But buried inside the wreckage were the kind of asymmetric bets that careers in venture capital get built on:

The forensic reconstruction by crypto researchers and a couple of court-watching journalists who refused to let this story go is what we are working from here. The math is messy in places because some of these positions are private, some are tied up in still-pending litigation, and some are subject to assumptions about future IPO pricing. But the broad strokes are not in dispute.

The Anthropic story

This is the big one. Pull up a chair.

Back in 2021, before the world had really heard of large language models, FTX Ventures wrote a $500 million check to a young AI company called Anthropic for what was reported as a 13.56% stake at a roughly $2.5 to $3 billion valuation. It was the kind of bet that, if you didn't already understand what foundation model labs were going to become, would have looked completely insane. Anthropic at that point had no revenue, no public product, and a small group of researchers who had recently walked out of OpenAI.

The stake got diluted down to about 8% over subsequent rounds, which is normal as the company raised more capital. Then in November 2022, FTX blew up. The 8% stake in Anthropic became part of the bankruptcy estate, managed by John J. Ray III and the lawyers at Sullivan & Cromwell.

Their job was simple: convert assets to cash and pay creditors. They are not paid to bet on the future of the AI industry. They are paid to be liquid.

So in March 2024, the estate sold roughly two-thirds of the Anthropic stake to a consortium of buyers including Mubadala (the Abu Dhabi sovereign wealth fund), Jane Street (SBF's old employer), Fidelity, and HOF Capital, for $884 million. In June 2024, they sold the remaining 15 million shares at $30 per share for another $450 million or so. Total proceeds: roughly $1.3 billion.

On its face, that is a great trade. They put in $500 million and got back $1.3 billion. 160% return. The estate threw itself a small party.

Then the AI boom went exponential.

By early 2026, Anthropic was generating something on the order of $30 billion in annualized revenue, up from roughly $9 billion only months prior. The company was valued in private secondary markets at $380 billion, and persistent reporting suggested a late-2026 IPO that could reach $1 trillion. At a $380 billion valuation, the same 8% stake the estate sold for $1.3 billion would be worth approximately $30.4 billion today. At $1 trillion, it would be worth $80 billion. The single Anthropic position would have made the FTX creditors something between 60x and 160x their original $500 million investment.

Instead they got the $1.3 billion. Which is, again, a perfectly nice trade. It is just not the trade.

The 15,000x miss

OK, so Anthropic is the big-dollar miss. But for sheer percentage absurdity, you have to talk about Cursor.

In April 2022, somebody at Alameda Research (operating through one of the alphabet soup of LLCs, in this case Clifton Bay Investments) wrote a check for two hundred thousand dollars - $200,000 - to participate alongside Heroic Ventures in a pre-seed funding round for a startup called Anysphere. Anysphere had built a product called Cursor: a fork of VS Code with AI baked into the editing experience. Nobody had heard of it. The whole round was tiny. Alameda got 5% of the company.

$200,000. For 5% of Anysphere. April 2022.

FTX collapsed seven months later.

In April 2023, the bankruptcy estate, looking to raise fractional cash from non-core positions, sold the 5% stake back to the founders at cost. $200,000 in, $200,000 out. The estate booked a zero. The trustee moved on to the next item on the list.

Here is what happened next.

Cursor hit $100 million in annual recurring revenue in January 2025. Then $500 million by June 2025. Then $1 billion by November 2025. By February 2026 it was at $2 billion in ARR, making Anysphere the fastest-growing business software company ever measured. Slack took five years to reach its first billion. Zoom took nine. Snowflake, the prior record holder, took six. Anysphere did it in roughly three years.

The valuation tracked: $9.9 billion in early 2025, $29.3 billion in November 2025 (Series D led by Accel and Coatue), and then in April 2026 the kicker - SpaceX announced an agreement giving it the right to acquire Cursor for $60 billion or pay a $10 billion breakup fee.

At a $60 billion valuation, the 5% stake the FTX estate sold for $200,000 is worth $3 billion.

That is a 15,000x missed return on the original investment.

You can read that number a couple of ways. One is that it is the most absurd opportunity-cost number in the history of bankruptcy administration. The other is that, on a $4.7 billion portfolio, a single bet getting away from you for $3 billion is just... that's the venture business. Power law. Vinod Khosla once said that out of every ten investments, seven fail, two return their cost, and one returns the entire fund. The Cursor pre-seed was that one. It just happened to belong to a bankrupt criminal exchange, and the people unwinding the criminal exchange were not in the venture business and could not afford to wait.

SpaceX (and one thing the estate got right)

This one is actually a feel-good story. Or at least the version of a feel-good story that you can tell when the underlying capital came from misappropriated customer deposits.

Alameda's exposure to SpaceX did not come from a direct check. It came through a venture fund run by Michael Kives and Bryan Baum called K5 Global. In 2022, in one of the more questionable last-minute moves before the collapse, Alameda transferred roughly $700 million to K5, of which somewhere between $190 million and $225 million ended up in SpaceX at valuations of $80 billion to $140 billion.

The initial bankruptcy reaction was predictable. The estate sued K5 to claw back the entire $700 million, arguing the transfers were fraudulent and basically existed for SBF to buy political access through K5's deep Hollywood and Washington Rolodex. (Kives was famously close to both Tony Blair and Kim Kardashian. Make of that what you will.)

The lawsuit settled in January 2025. And here is the part that mattered: instead of forcing K5 to cough up $700 million in cash that K5 plainly did not have liquid, the estate retained its economic interest in K5's underlying portfolio, including the SpaceX exposure.

This decision is going to look like a stroke of genius any day now.

SpaceX is filing its S-1 essentially right now, targeting a Nasdaq listing on June 12, 2026 under the ticker SPCX at a valuation between $1.75 trillion and $2 trillion. That would make it the largest IPO in American history by a country mile. Depending on dilution and the specific structuring of K5's fund mechanics, the FTX estate's indirect SpaceX stake is somewhere between $2.8 billion and $15 billion at IPO pricing. Even at the low end, that is more than 10x the original investment, in cash, dropped into the creditor trust right after the lockup expires.

Robinhood: the most embarrassing 7.6% in American capital markets

In May 2022, six months before everything fell apart, Sam Bankman-Fried showed up on Robinhood's shareholder register holding a 7.6% stake. He had not bought the shares with his own money. He had borrowed roughly $546 million from Alameda Research (which is to say, from FTX customers, although nobody publicly understood that at the time) to acquire it. The full position cost $648 million at an average price of $11.52 per share.

The Robinhood shares became the focal point of one of the messier bankruptcy fights of the decade. Three different parties claimed them:

1. The FTX estate, arguing the shares had been purchased with Alameda funds that ultimately traced back to customer deposits
2. BlockFi, the bankrupt crypto lender, claiming the shares had been pledged to them as collateral
3. The US Department of Justice, claiming forfeiture as proceeds of fraud

While the lawyers fought it out, the US Marshals Service held the shares. Eventually Robinhood itself stepped in and offered to buy them back from the government, both to clean up the embarrassing legal overhang and to retire the shares so they were not sloshing around as a market-overhang risk to the company's own stock price. The deal closed in September 2023. Robinhood paid about $605.7 million for the entire 7.6% block - roughly $10.96 per share, less than SBF's original average cost.

And then, of course, Robinhood stock proceeded to go absolutely vertical.

Driven by a massive resurgence in crypto trading revenue (Robinhood pulled in $358 million in crypto fees alone in Q4 2024 and the figure has held), HOOD shares spent most of 2024 and 2025 climbing. By mid-May 2026, the stock was trading around $77 per share. If the original 7.6% block had simply been held by the estate, it would be worth somewhere in the neighborhood of $5 billion to $5.7 billion against the $648 million cost basis.

Now in fairness, the estate could not have held those shares. They were legally contested, frozen by the US Marshals, and any attempt to litigate ownership to a clean conclusion would have taken years and probably cost a fortune in fees. The buy-back from Robinhood was a clean exit from a mess that nobody fully owned. But the math is still ugly.

Solana, or how the FTX estate handed Pantera Capital a $2B gift

If Anthropic was the big private-equity miss, Solana was the big crypto miss. And the structure of how the estate handled it is one of the most interesting case studies in distressed asset sales I have ever seen.

SBF was Solana's biggest cheerleader from approximately the moment Anatoly Yakovenko stood up the network. FTX and Alameda were the biggest single holders of SOL tokens by a wide margin, accumulating somewhere between 58 million and 60 million tokens at an average price near $8 each. Total cost basis: somewhere around $60 million, although some forensic estimates put it higher due to later add-ons.

The vast majority of these tokens were subject to multi-year on-chain lock-up periods. They were not freely tradable. They were a known, public, gigantic overhang on the SOL price for the entire duration of the bankruptcy. Every crypto trader in the world knew the FTX estate was sitting on something like 10-15% of all SOL ever issued, and every crypto trader in the world was pricing in the eventual dump.

The estate, working with the bankruptcy court, came up with a clever solution. Rather than dump the tokens on the open market and crater the price (which would have hurt their own recovery and produced a worse outcome for creditors), they sold large blocks of locked tokens over-the-counter to institutional buyers at a steep discount. Between 2023 and 2024, the estate offloaded somewhere between 25 million and 30 million locked SOL tokens at a price of $64 per unit. Total proceeds: approximately $1.9 billion. The buyers were a who's-who of crypto venture: Galaxy Digital, Pantera Capital, Brevan Howard Digital, and others.

$1.9 billion is an enormous recovery on a $60 million cost basis. Roughly a 30x return. Everybody got paid. The crypto press wrote complimentary articles.

And then the buyers turned around and made multiples of their money in eighteen months.

SOL ripped from the low double-digits to a peak of $293 in early 2025. It currently trades around $85 to $86 after a multi-month correction. Even at the depressed current price, the original 58 million tokens would be worth roughly $5 billion intact. At the peak of $293, that position would have been worth $17 billion.

The OTC buyers who took the locked tokens at $64 are sitting on multi-billion-dollar gains. That is exactly the wealth transfer you would expect when a forced seller meets sophisticated buyers in a thin private market.

The crypto graveyard

The Solana story is the bright spot in the crypto book. Most of the rest is not.

SBF and the FTX team poured hundreds of millions of dollars into other Layer-1 blockchain projects that were essentially marketed as "Solana killers" or "Ethereum killers." With the benefit of hindsight, almost all of them turned out to be... worse than Solana, and not as established as Ethereum.

The Sui story is particularly painful in a kind of "wrong way around" sense. FTX led a $300 million Series B in Mysten Labs in September 2022 - literally weeks before the exchange collapsed - valuing the company at $2 billion. The estate, dealing with a deeply illiquid private position in a brand-new blockchain company, divested the entire stake back to Mysten Labs itself for under $100 million. SUI subsequently traded above $5 at peak before settling near $1.11 today. At peak token pricing, the original position would have been worth around $1.2 billion. The estate took a 67% haircut to get out fast.

Aptos has a similar shape, except that the seed round was even more aggressive and the eventual market reception even cooler. APT trades at around a dollar today and the network has not lived up to its initial pre-mainnet billion-dollar valuations.

The mining money: Genesis Digital Assets

I'm going to spend a moment on this one because it might be the single most egregious piece of capital misallocation in the entire portfolio, and it had nothing to do with the bankruptcy estate's decisions afterwards.

Between August 2021 and April 2022, SBF directed Alameda to deploy $1.15 billion into Genesis Digital Assets (GDA), an industrial-scale Bitcoin mining operation with operations concentrated in Kazakhstan. The bankruptcy estate later described the deal in court filings as having been done at "insane and off-market" valuations. Diligence, by any reasonable definition, was almost nonexistent.

Worse: of the $1.15 billion deployed, more than $550 million did not go into the company's operating balance sheet at all. It went directly to GDA co-founders Rashit Makhat and Marco Krohn as secondary share buyouts. Translation: SBF used FTX customer money to buy out existing GDA insiders, personally enriching them with hundreds of millions of dollars in liquidity, while delivering almost nothing to the underlying business.

The FTX bankruptcy estate has filed a $1.15 billion fraudulent transfer clawback lawsuit against GDA seeking the return of the entire investment. The litigation is ongoing. Genesis Digital is still operating, its enterprise value is estimated at around $3.5 billion today (a multiple on the underlying business of about 3x relative to the deal price, although nowhere near enough to justify the original valuation), and the eventual recovery to FTX creditors depends entirely on how the avoidance action shakes out in court.

Modulo Capital: the one easy recovery

You want a palate cleanser? Here's a palate cleanser.

Sometime in 2022, SBF directed Alameda to drop between $400 million and $475 million into a multi-strategy quant fund based in the Bahamas called Modulo Capital. The fund was run by Xiaoyun "Lily" Zhang (a former girlfriend of SBF's, reportedly) and Duncan Rheingans-Yoo. The fund had been operating for approximately fifteen minutes when the half-billion dollars showed up.

When FTX collapsed, Modulo's principals were... let's say, motivated... to settle quickly. Possessing what was very obviously misappropriated customer money is the kind of thing that focuses the mind. In March 2023, just four months after the collapse, Modulo agreed to return $404 million in cash to the estate and to give up any claim on an additional $56 million in assets sitting on the FTX exchanges. Total recovery: roughly $460 million on a deployment of, call it, $440 million.

The estate booked it as one of the cleanest, fastest, and most cost-efficient recoveries of the entire bankruptcy process. No protracted litigation, no clawback suit, no opportunity-cost hand-wringing. The Modulo principals walked away from the relationship before the wreckage got too close to them, and the estate got its money back essentially in full.

Dave Inc: when the convertible note got cleaned up cheap

This one is going to be confusing because the company is also named "Dave," but it's not me, I promise.

In March 2022, FTX Ventures wrote a $100 million convertible note into Dave Inc. (NASDAQ: DAVE), a digital banking app aimed at younger demographics that had just gone public via SPAC. The convertible was paired with a strategic partnership intended to integrate crypto trading into the Dave banking app via FTX US. Bold vision. Did not survive contact with reality.

When FTX collapsed, Dave Inc. very understandably wanted nothing to do with the disgraced exchange. They opened negotiations with the bankruptcy estate to clean up the note. In January 2024, Dave repurchased its own convertible note from the FTX estate for $71 million - a 33% discount to the outstanding principal balance. Dave booked a $33.4 million gain on extinguishment. The estate got immediate cash. Everybody moved on.

Three issues with this transaction, in retrospect.

First, the original convertible note had a conversion feature. If it had remained in place and converted to equity at the original terms, the eventual outcome would have been very different. Second, Dave's stock proceeded to do exactly what you would expect a profitable mobile-first banking app aimed at Gen Z to do during a fintech boom: it went absolutely vertical. The stock currently trades around $242 per share, with analyst price targets averaging around $340 (Citizens just bumped its target to $365, Keefe Bruyette to $340, Canaccord to $342). Market cap is approximately $2.3 billion.

Third, and this is the most painful piece: Dave Inc. has been reporting genuinely excellent fundamental performance. Q1 2026 revenue came in at $158.4 million, up 47% year-over-year, with non-GAAP EPS of $3.64 (crushing the $2.93 analyst consensus). Net monetization expanded to 5.1%, a multi-year high. Management subsequently raised FY2026 revenue guidance from $690-710 million to $710-720 million. The company has a fortress balance sheet, just raised $200 million in zero-coupon convertible notes, and authorized a $300 million share buyback program (of which $194.9 million was deployed in Q1 alone). The estate gave up exposure to this exactly as the inflection was kicking in.

IEX Group: the $270M ghost

In one of the more surreal moments of the FTX era, SBF wrote a $270 million check for a minority equity stake in IEX Group - the regulated US stock exchange made famous by Michael Lewis's book Flash Boys - to gain leverage in Washington DC and presumably bridge crypto market structure with traditional finance.

IEX remains a private company. The stake remains with the estate. Current valuation: opaque. Will probably end up being worth something. Probably not nothing, probably not a windfall. It is, of all things, the most quietly normal investment in the entire portfolio.

The Yuga / Web3 gaming wreckage

In March 2022, near the absolute peak of NFT mania, FTX Ventures wrote a $50 million check into a $450 million seed round for Yuga Labs, the company behind Bored Ape Yacht Club, at a $4 billion valuation. Alameda also directly held 92 high-value NFTs - including 31 Bored Apes, 26 Otherdeed for Otherside tokens, and three Gold-furred Apes (valued at 1,000 ETH each at the time). At the November 2022 collapse, the NFT collection itself was valued near $5.8 million.

Then the NFT market collapsed by approximately 90%. The Yuga Labs equity collapsed with it. The estate's NFT holdings have been winding down in various OpenSea sales over the past two years. This is essentially the rare case where the FTX estate's mandate to liquidate fast actually protected creditors from worse losses down the line.

Similar story for Sky Mavis, the studio behind Axie Infinity. FTX held a $2 million direct equity stake from the late-2021 Series B at a $3 billion valuation. The play-to-earn gaming category got destroyed in 2022-2023 and never really came back. The Sky Mavis position is heavily depreciated.

The Alameda Spreadsheet

Almost forgot to mention this. The "Alameda Spreadsheet" is real. It leaked in late 2022 and made its way around crypto Twitter. It is an Excel file maintained internally that cataloged the firm's positions, and at its largest it contained roughly 400 distinct investments. About 250 of these were token positions or token IOUs, representing a deployment of roughly $3.2 billion across crypto projects that ranged from real (Solana, Polygon, Lido) to "lol, what is this thing" (large stretches of the spreadsheet are projects most readers, including me, have never heard of).

Many of these positions had functional liquidity only because FTX's own market-making operation was creating it. When FTX disappeared, so did the liquidity. The "Sam Coins" (a market nickname for tokens heavily backed by SBF) traded off severely in late 2022 and through 2023 on the perfectly rational fear that the estate would dump them.

This created what was, in hindsight, one of the most asymmetric arbitrage opportunities in the history of crypto markets. The ecosystems that had legitimate fundamental value (Solana being the obvious one) were trading at depressed valuations specifically because of the bankruptcy overhang. The funds that bought during this window did extraordinarily well. The retail FTX creditors who had effectively financed the original investments did not.

The big number

Here is the chart that I have been thinking about ever since I started reading the forensic reconstructions.

FTX put in $4.7 billion in misappropriated customer money. The estate has recovered roughly $18 billion (most of which came from broader crypto appreciation, the Anthropic sale, the Solana OTC block sales, and the various smaller asset windings). That recovery has been enough to pay creditors back at 119% to 160% of their November 2022 dollar-denominated claim values, with the upper end of that range pushed higher in early 2026 after the Delaware Bankruptcy Court approved a reduction in the litigation reserve.

And if the portfolio had been held intact, it would currently be worth somewhere between $52.5 billion and over $100 billion.

So who did this benefit?

Let's just say it plainly. The Chapter 11 bankruptcy process for FTX has had three classes of beneficiaries.

The paradox

I am going to wrap up with the thing I cannot stop turning over in my head.

Sam Bankman-Fried was a fraud. The fraud was real, the fraud was massive, and the fraud was deliberate. He stole customer deposits. He lied about doing it. He directed an extraordinary multi-billion-dollar gambling operation through Alameda Research using funds that were not his. There is no version of this story where he is rehabilitated or exonerated by what comes next. He is going to be in prison until 2048.

And yet.

The asset allocation strategy executed with that stolen money was, by the cold, dispassionate metrics that venture capital uses to evaluate itself, one of the most successful private investment portfolios of the last decade. Anthropic at $2.5 billion. Cursor at $4 million. SpaceX before it was the largest IPO in American history. Solana when it traded for $8. These are the bets you go to your grave bragging about if you are a normal venture capitalist. He hit all four. In a roughly eighteen-month window.

The fund's terminal value, if it had been structured legally and run with anything resembling normal risk management, would have made it one of the great venture vehicles of the era. The same instincts that selected those investments also drove the gambling and the commingling and the criminal misappropriation that ultimately destroyed everything. You cannot separate the two. The risk tolerance that found Anthropic at $2.5B is the same risk tolerance that thought it was fine to gamble with customer money.

The bankruptcy estate did exactly what bankruptcy estates are supposed to do. They sold. They paid the creditors. They closed the book.

And tens of billions of dollars in upside walked out the door to the buyers of last resort.

This is what happens when an excellent investor, working for a fraudulent house, gets liquidated by an honest trustee, in a market that is about to rip.

Everybody got what they were supposed to get. It's just that none of it was anywhere close to what it could have been.