FTX Anthropic Investment: The $43B What-If Explained
FTX's Anthropic stake became a striking bankruptcy what-if. Examine the reported valuation, the share math, asset sales and the trade-offs for creditors.

FTX's Anthropic Stake: A $43 Billion What-If, With a Catch
FTX put early money into Anthropic, and after the company's reported IPO, a widely shared estimate claimed the exchange's stake could have been worth more than $43 billion. The bankruptcy estate ultimately sold the shares in 2024 for about $1.34 billion as part of its effort to repay creditors.
That headline comparison is attention-grabbing, but it doesn't stand on its own. The original investment was funded with customer money, the estate had to raise cash quickly enough to satisfy bankruptcy priorities, and the shares involved were not a straightforward, unrestricted public-market position. There's also a glaring arithmetic problem: the share count and closing price cited in the same account don't add up to the reported $43 billion figure.
This article lays out what the estate sold, separates that from the theoretical valuation claim, and checks what can be checked. It also looks at what other FTX investments suggest about how easily "hindsight performance" can be misread. The distinction matters: a missed upside is different from money that was actually available to customers, and a portfolio marked at optimistic future prices isn't the same thing as cash in hand.
What happened to FTX's Anthropic stake?
In October 2021, an entity linked to Alameda Research committed about $500 million to Anthropic, then an early-stage AI company. The investment later converted into preferred shares through Anthropic's Series B financing. According to the account that circulated, those shares represented about 13.6% of the company at the time, before later fundraising diluted the holding.
That investment predated FTX's collapse in 2022. After the exchange failed, the bankruptcy estate took control of assets tied to FTX and Alameda. Its job wasn't to run a long-term venture portfolio for maximum upside; it was to identify assets, preserve value, and distribute money to customers and other creditors through the bankruptcy process.
In 2024, administrators sold the Anthropic position in two batches for a reported total of about $1.34 billion, following court approval. Buyers reportedly included an Abu Dhabi-linked fund and Jane Street, the trading firm where Sam Bankman-Fried had previously worked. The proceeds went into the estate's recovery efforts, not into any personal windfall for Bankman-Fried.
The difference between the original investment and the later sale is the core point. The $500 million was not Bankman-Fried's personal capital. The same source account says customer funds were used to finance the deal. So when the shares were sold, it was property controlled by the bankruptcy estate, with proceeds intended to support creditor recovery - not a private investment outcome.
After Anthropic's reported November 2026 IPO, social media users compared the estate's sale proceeds to what the shares might have been worth if they were retained. Using a reported closing price of $97.17, the former stake was said to be worth $43.28 billion. Using an opening trade price of $111.50, the estimate rose to $49.66 billion.
Those numbers sparked a big "what if" conversation - especially after The Altruists, Netflix's drama about Bankman-Fried and the FTX collapse, drew renewed attention to the story. But a hypothetical market value doesn't prove what the estate could have realized, whether it could have held the shares, or how much customers would have received under a different path.
What the comparison leaves out
The estate's sale price is a reported transaction value. The later number is a theoretical mark based on a quoted share price. One is cash raised in a completed sale. The other relies on assumptions about share count, share rights, dilution, and whether a large position could be sold without moving the market.
That doesn't mean the comparison is pointless. It does mean it's incomplete. Before treating the difference as a realized loss, the underlying arithmetic needs to be verified.
Does the $43 billion estimate add up?
The headline valuation doesn't reconcile with the share count and prices listed in the same account. It claims FTX's position was about 44.54 million shares valued at $97.17 each at the close. Multiply those figures and you get roughly $4.33 billion - not $43.28 billion.
The opening-price calculation shows the same issue. At $111.50 per share, 44.54 million shares would be worth about $4.97 billion. The reported estimate is $49.66 billion. In other words, both headline values land exactly ten times higher than what the stated share count and prices produce.
A tenfold difference could come from something missing - an omitted conversion ratio, a different share count, a reporting error, or another adjustment. For example, if the true position were 445.4 million shares, values close to the stated estimates would be possible. But the source account doesn't include the corporate-action or conversion details needed to resolve the mismatch. Without that, the $43 billion figure should be treated as an unverified reported estimate, not as a calculation derived from the figures presented.
There's a second detail that also shouldn't be blended into the same math. The account describes the initial position as about 13.6% of Anthropic and separately references a 7.8% stake being sold. Those percentages could reflect changes in ownership, dilution, or differences in how the stake was described. But they can't verify the value unless the share class, the number of shares at sale, and any conversion terms are clearly laid out.
A valuation is not the same as realizable proceeds
Even if the share count were correct, multiplying it by an IPO price would still produce a theoretical mark, not necessarily the amount the estate could have collected. Large shareholders can face lockups, transfer restrictions, or other limits on how quickly and how easily they can sell. A fast exit can also push prices down. And the position might have terms that don't match those of ordinary publicly traded shares.
So the relevant questions aren't just "What was the stock price?" They're more specific:
- How many shares did the estate actually control after dilution and conversion?
- Were those shares freely tradable at the time the hypothetical valuation is supposed to apply?
- Did the quoted price apply to the same class of shares?
- How much of the position could have been sold, and over what time period, without meaningfully changing the price?
Until those details are available, the safest conclusion is limited: the Anthropic investment appears to have appreciated substantially. But the cited $43 billion estimate can't be reproduced from the stated 44.54 million shares and the $97.17 closing price. That gap belongs at the center of any careful retelling of the story.
Why did the bankruptcy estate sell the shares?
The sale has to be evaluated based on what the estate was actually trying to do. FTX collapsed, customers were waiting for recovery, and the estate's role included turning assets into funds that could be distributed under bankruptcy rules. Holding a private-company stake for years might have produced a higher eventual return. It also would have exposed creditors to added delay, changes in valuation, and the risk that the company's prospects deteriorated.
A private-market stake isn't the same as cash sitting in an ordinary brokerage account. Before a company lists publicly, shares can be subject to transfer restrictions and deal terms negotiated in advance. Even after a listing, insiders and large holders may still face lockups or other limits. And a publicly quoted price can reflect trades at the margin rather than the price someone would get for a full block that a bankruptcy estate holds.
The estate also had to operate through a legal process. A judge approved the Anthropic sale in February 2024, according to the source account. That approval doesn't automatically prove the sale was the best possible outcome in every hypothetical scenario. But it does show the sale was handled through bankruptcy oversight, rather than being a private, unilateral decision by Bankman-Fried after the collapse.
There's also an ethical issue that hindsight can blur. The original investment reportedly used customer funds. If those funds were misused, the later increase in the investment's value doesn't retroactively make that conduct acceptable. A profitable bet made with someone else's money still involves a breach of trust - even if, later on, it looks like it "worked."
The creditor trade-off
The estate's choice can be described as a familiar trade-off:
- Sell earlier: turn an illiquid asset into cash, reduce uncertainty, and make recovery possible sooner.
- Hold longer: keep exposure to potential growth, while accepting market, legal, liquidity, and execution risk.
In reality, it rarely comes down to "sell at the bottom" versus "hold for the top." Administrators don't know the future price, and they aren't managing assets like a single investor who can wait out years of risk. They had legal duties, incomplete information, and the obligation to treat creditors fairly.
That doesn't mean there's nothing to scrutinize. The sale price, timing, and process are reasonable targets for questions. But the later market price alone can't answer those questions. To evaluate the decision, you'd want contemporaneous bids, the stake's restrictions, the estate's liquidity needs, and the risks they expected at the time. The popular comparison can point you toward that analysis - it can't replace it.
What do FTX's other investments show?
The Anthropic stake was only a piece of a much larger portfolio. A leaked spreadsheet reportedly showed about $5.4 billion allocated across roughly 500 investments. A later reconstruction of 461 positions put the original investment total at about $5.32 billion and estimated a combined value of roughly $59.5 billion using later valuations.
On that model, the numbers imply about $54.2 billion in hypothetical gains, or roughly a 1,018% return on the reconstructed cost. The scale is striking, but the method matters: this kind of calculation assumes positions were still held and assigns values to assets that may not have been liquid or publicly traded. It's a portfolio mark, not proof that the estate could have sold every position at those prices.
Even with those limits, individual examples illustrate both upside and uncertainty:
- Anysphere: Alameda reportedly invested $200,000 in the company behind the AI coding tool. The estate later sold the stake for the same amount. A reported 2026 transaction valued the company at $60 billion. If the original stake had been around 5%, its undiluted value could approach $3 billion - but that's an upper-bound estimate before accounting for later financing and dilution.
- Solana: The estate reportedly sold about 25 million to 30 million locked SOL tokens near $64 each in 2024. At Solana's reported January 2025 high of $293.31, that tranche would have been worth roughly $7.3 billion to $8.8 billion. Again, the comparison uses a later peak, not a price available when the tokens were sold.
- Robinhood: A Bankman-Fried-linked vehicle reportedly bought about $648 million worth of Robinhood shares in 2022. Authorities seized 55.27 million shares, and Robinhood repurchased them for $605.7 million in 2023. At a reported 2025 intraday high of $153.86, that block would have marked near $8.5 billion.
Other positions in the reconstruction look more muted. A reported $1.15 billion investment in Genesis Digital would imply a value around $290 million at a 2026 transaction valuation. A $40 million investment in drone maker BRINC was estimated near $109 million, while a $20 million position in Anchorage Digital was estimated around $31 million.
Why portfolio backtests can mislead
These examples blend together completed sales, later market highs, private financing marks, and transaction valuations. They also draw on different time windows and assumptions. Adding them up can help illustrate the size of the portfolio, but it doesn't create a clean measure of realized performance.
A proper realized-return calculation would have to track when each investment was bought and sold, how much dilution occurred, whether the asset was locked, what fees or liabilities applied, and whether the estate could actually find buyers at the cited valuation. A company's headline valuation isn't automatically a guaranteed cash-out price for a large early investor.
The broader takeaway isn't that FTX's managers had a reliable system for producing exceptional returns. Some bets surged; others lost value or were sold before later peaks. The reconstruction shows how venture-style portfolios can look dramatically different in hindsight when winners are marked at later valuations. It doesn't prove the portfolio was liquid, that outcomes were predictable, or that the investments were justified by the way they were funded.
What should creditors and crypto investors take from this?
For creditors, the useful question isn't how much a portfolio might look worth at a future peak. It's how much value could reasonably be recovered, over what timeframe, and with what risk. A theoretical valuation can be high while the decision to sell still makes sense if the assets are restricted, the estate needs liquidity, or the price can't be realized for the full position.
For crypto investors, the lesson is about custody and governance as much as returns. A customer deposit isn't venture capital that an exchange can deploy at its discretion. When an intermediary mixes customer assets with its own activities or uses them without consent, a "successful" investment doesn't fix the underlying failure. It may make the lost opportunity more visible - but it doesn't change who took the risk.
For anyone evaluating bankruptcy sales, it helps to separate four numbers that often get mashed together:
- Cost basis: what was originally invested.
- Sale proceeds: what the estate actually received.
- Quoted market value: what shares might be worth at a stated price.
- Realizable value: what could have been collected after restrictions, dilution, liquidity, and transaction costs.
Then ask whether those figures refer to the same share count and the same time period. In this case, the reported $43.28 billion value and the stated 44.54 million shares at $97.17 don't match arithmetically. That isn't a minor quibble - it changes the size of the comparison by a factor of ten.
Key takeaways
- The estate reportedly sold the Anthropic position for about $1.34 billion in 2024 to support creditor recovery.
- The reported $43.28 billion mark doesn't follow from the stated share count and closing price; those figures imply about $4.33 billion.
- Any comparison between sale proceeds and a later market price has to account for dilution, share rights, restrictions, and market liquidity.
- Later highs and private-company valuations can create useful hindsight scenarios, not guaranteed proceeds.
- Investment success doesn't excuse the use of customer funds without consent.
The strongest reading isn't that the sale was obviously wrong or that the estate had no alternatives. It's that the math needs to be checked, and the decision should be judged using the information, duties, and constraints that existed when the sale was made - not just against what came later.
Conclusion
FTX's Anthropic investment is a sharp example of how bankruptcy recovery and a later what-if valuation can get mixed together. The estate reportedly collected about $1.34 billion, while a widely shared estimate claimed the shares could have been worth more than $43 billion after Anthropic's reported IPO. But the share count and price cited alongside that headline produce a figure closer to $4.33 billion.
The investment may still have generated substantial upside. That doesn't settle whether the estate should have held it, and it doesn't erase the customer-fund issue at the heart of the FTX collapse. The practical takeaway is to verify the arithmetic, distinguish market marks from realizable cash, and judge the sale based on the constraints decision-makers actually faced - not only on the price that came later.