Survive to be right

Here is a fact that ought to be taught in every finance course as a cautionary tale, and instead mostly turns up as a footnote.

In 2021, FTX — the crypto exchange run by Sam Bankman-Fried — put $500m into a one-year-old AI lab founded by a handful of researchers who had just left OpenAI. The lab was Anthropic. The cheque bought roughly 8% of the company. At the time it looked like the eccentric side-bet of a man who made a lot of eccentric side-bets. Last Monday, Anthropic filed its S-1 at a $965bn valuation. The eccentric side-bet had quietly become one of the greatest venture positions ever assembled.

And FTX got essentially none of it.

The arithmetic, and the wince

When FTX collapsed in late 2022, the Anthropic stake passed to the bankruptcy estate. The estate did what estates do: it sold. Across 2024 it offloaded the position in tranches — two-thirds of it went for $884m in March, the rest soon after, call it $1.3bn all in, at an implied valuation somewhere around $17bn.

Now run it forward. That same stake, had it simply been left alone, would today be worth a fortune that is hard to type with a straight face. I want to be honest about the number, because the lazy version — eight per cent of $965bn, so $77bn — is wrong. The original 8% would have been diluted through six further funding rounds; the Series H alone raised $65bn. So the position would be smaller than a naïve slice of the headline valuation. But even heavily diluted, it would be worth tens of billions of dollars. Many times what the estate sold it for. Many times FTX's own peak valuation of roughly $32bn. One holding, left untouched, would have repaid every FTX creditor several times over and left a surplus large enough to start again.

It didn't, and the reason has nothing to do with whether the bet was good. The bet was superb. The reason is timing.

Being early and bankrupt looks exactly like being wrong

This is the part worth sitting with, because it generalises far beyond one disgraced exchange.

A correct thesis and a correct thesis you didn't survive to collect produce identical entries in the ledger: a realised loss. The market does not award partial credit for foresight. It does not annotate the liquidation with "but he was right." When FTX failed, the Anthropic stake became an asset held by a forced seller, and forced sellers do not get to choose the day they sell. They sell when the administrator needs liquidity, into whatever bid exists, at whatever the clock says the asset is worth that morning. In early 2024 the clock said $17bn. The clock was wrong, in the sense that it had not yet caught up to what the asset would become — but the estate had to trade against the clock it had, not the one it wanted.

SBF, whatever else is true of him, had identified the single best venture opportunity of his era before almost anyone. It bought him nothing, because conviction without survival is just a more elaborate route to the same zero. The graveyard of investing is not mostly full of people who were wrong. It is full of people who were right on the wrong schedule — early, leveraged, and gone before the payoff landed.

The steel-man: maybe the estate did the right thing

It would be too neat to say the sale was simply a blunder, so let me argue the other side. The estate had a fiduciary duty to creditors who wanted their money, not a mandate to run a concentrated venture bet on their behalf. A bankruptcy estate is not a hedge fund; holding an illiquid, volatile, single-name position for two more years on the theory that it might 50x is precisely the sort of gamble a fiduciary is supposed not to take. Judged by the rules it actually operated under, selling was defensible, maybe even correct.

But that is the whole point, not a rebuttal to it. The estate had no choice precisely because the people before it had destroyed the optionality. By the time the asset sat in the estate's hands, the freedom to be patient — the single most valuable thing an investor in a compounding asset can own — had already been spent. The blunder wasn't the sale. The blunder was everything upstream that turned a patient holder into a forced one. Survival is not a separate virtue from being right. It is the thing that converts being right into being paid.

What it means if you actually allocate capital

The lesson is not "hold your winners," which is the bumper-sticker version and mostly useless. The lesson is structural. The reason margin of safety matters is not that it improves your hit rate — it doesn't. It is that it keeps you solvent across the gap between when you are right and when the world agrees. That gap is almost always longer than your conviction expects, and the cost of crossing it is paid in survival, not in cleverness.

I think about this every time someone shows me a brilliant, aggressive, fully-committed thesis — in markets, in a startup, in an AI deployment betting the quarter on a single vendor or a single bet landing on time. The thesis can be flawless and the position can still be fatal, if it has no slack for the schedule to slip. The operators and allocators who compound over decades are rarely the ones with the best calls. They are the ones who structured their bets so that being early didn't kill them before being right could pay them.

Anthropic's IPO is about to mint a great many paper fortunes. The most instructive of them is the one that was seen first and held by the person least able to keep it. Sam Bankman-Fried found the trade of the decade and handed it to his creditors at a ninetieth of its eventual worth — not because he misjudged the asset, but because he misjudged the one variable that sits above every asset.

Be early if you can. Be right if you can. But build so that you are still standing on the morning the bet comes good. The clock is the one counterparty you can never out-argue, and it has bankrupted more good ideas than bad ones ever did.