The number circulating through trading desks this week is $117.6 billion. That is the theoretical value of the 7.84% stake in Anthropic that the FTX bankruptcy estate once held, marked against a rumored $1.5 trillion IPO valuation for a listing as early as mid-November. The estate actually recovered approximately $1.3 billion through a staged 2024 sale, netting roughly $800 million on a $500 million outlay made in 2021. The gap โ $116.3 billion โ is being retold across the industry as the greatest missed trade in crypto history, a morality play about conviction and cowardice.
It is neither. It is a documentation failure dressed as a tragedy, and it depends on an assumption nobody in the retelling has bothered to audit. Code compiles, but context reveals the exploit.
The setup deserves precision, because the numbers being quoted are almost universally stale. Anthropic's rumored listing price implies roughly $1.5 trillion, about 55% above the $965 billion mark attached to its May financing โ a step-up of more than half a trillion dollars in under six months. The cap table behind that figure has been assembled in layers. Spark Capital put in $75 million in 2023 at a company valuation near $5 billion and secured the only known board seat held by a venture firm. Sound Ventures, the Guy Oseary and Ashton Kutcher vehicle, entered with $35 million and later added $400 million. Menlo Ventures has deployed over $1 billion in total, including a $500 million Series D in May 2023, with its position now carried at roughly $21.8 billion. Iconiq arrived in 2025 at an initial valuation of $183 billion and has accumulated more than $6 billion, marking its stake near $26.4 billion. Sequoia deployed about $3 billion this year for approximately 1% of shares, a position carried near $15 billion; Singapore's GIC committed over $4.5 billion across 2025 for a similar 1%. Eric Schmidt bought in for tens of millions in 2021 at about 1%. Lightspeed holds under 2%.
FTX's entry was the largest single-position anomaly in that table. $500 million in 2021 bought 13.56% of the company. By 2024, dilution had reduced that to 7.84% โ a decline of 42.2% in share count over roughly three years, or about 12.5% annually. The estate sold the position in batches: a first tranche of approximately $884 million distributed across 24 buyers, with Abu Dhabi sovereign fund Mubadala's affiliate ATIC taking around $500 million of it, Jane Street roughly $100 million, and a Fidelity fund $45 million. The remainder is opaque, and the names with the largest disclosed exposure โ Google, Amazon, Nvidia, Microsoft, Founders Fund, Coatue, Altimeter โ have not published position sizes sufficient to compute anything.
Start with the dilution, because it silently corrupts the headline figure everyone is quoting. The $117.6 billion number treats 7.84% as a fixed constant and applies it to a peak mark. That is not how a cap table behaves. The arithmetic clears. The assumption does not. If a single additional round lands between now and the roadshow โ and every company in this position raises one more time โ that 7.84% compresses toward 6.8%, and the theoretical value falls to roughly $102 billion. Apply a standard pre-IPO lockup discount and secondary-market illiquidity haircut, both of which any due diligence desk applies to private marks, and the gap narrows by another 20% to 30%. The claimed loss does not shrink from $116 billion to pocket change, but the number that gets screenshotted is a static snapshot of a moving denominator, and nobody is disclosing their margin of error.
Then examine how the sale actually cleared. Twenty-four counterparties absorbed $884 million, and a single sovereign-adjacent buyer took 57% of the tranche. In the Wash Trading Index methodology I have run for years on NFT and low-float token volumes, any print where one counterparty exceeds 40% of the distribution is flagged as non-representative of price discovery. A bankruptcy-court-approved transfer is the least price-discovering trade that exists: the seller operates under a fiduciary clock and cannot wait, while the buyer has information asymmetry and can. That is not a market clearing. It is a negotiated transfer with a judicial signature attached, and treating its clearing price as evidence about the underlying asset's value โ in either direction โ is a category error.
The valuation step-up itself is where I would normally spend a compliance week. In 2020, verifying Aave's v1 liquidity mining incentives, I built a SQL dashboard tracking daily APY against actual treasury reserves. The signal was never the yield; it was the widening gap between promised returns and the balance sheet backing them. The same instrument applies here. $965 billion to $1.5 trillion in six months, with no disclosed revenue multiple, no audited compute obligations, no contracted backlog figure. Private marks carry no liquidation test, which is their entire appeal to the people setting them. A mark is not a price; a price is not a market. Until an S-1 forces audited concentration and commitment disclosure, every figure in the $965 billion to $1.5 trillion range is an opinion with a decimal point.
Expect the tokenization lobby to claim this as vindication. They will argue that a tokenized pre-IPO registry would have let FTX's creditors hold the position through to listing rather than force a distressed dump. Review the mechanics of what actually happened. The transfer required a bankruptcy court order, an SPV structure, KYC on 24 institutional counterparties, and securities counsel in multiple jurisdictions. No public chain was consulted at any step, and no chain could have substituted for the court. During the 2025 MiCA compliance build I led for a Portuguese CASP, the binding constraints on every transaction flow were always KYC/AML algorithm gaps and jurisdictional liability โ never settlement speed. Tokenization solves a problem that was not the constraint. The institutions holding Anthropic paper did not need a public ledger; they needed a signature on a court document.
Where the bulls are correct, and this deserves stating plainly, is that the FTX trustees did their job. A bankruptcy estate cannot warehouse illiquid private equity across an IPO window and an 18-month post-listing lockup while creditors wait. The fiduciary duty was to maximize present recovery with certainty, not to speculate on a listing that had not been filed. Roughly 160% gross return on a $500 million outlay, realized inside a liquidation timeline, is a defensible outcome โ arguably a good one. The $116 billion figure is a hindsight artifact, not a breach. And the skeptics who called the $965 billion mark absurd were, on direction, simply wrong; enterprise AI revenue concentration is real, contract-backed, and growing. Cold skepticism is a method, not a mood, and a method that cannot register when the other side is right is just pessimism with better vocabulary.
The actionable item is the S-1, not the rumor mill. The first audited disclosure of customer concentration, compute commitments, and gross margin will settle more arguments than a decade of private marks. The second item is the claims market: watch whether the next large estate in this position prices its illiquid equity differently, or sells into a 24-buyer auction again.
If a $117.6 billion position can be liquidated for $1.3 billion and no party is liable, what precisely is the thing due diligence claims to price?