Sam Bankman-Fried Backed the Hottest AI Startup — Then Lost the Jackpot

By the time the AI gold rush reached Sam Bankman-Fried, he was already wearing prison khaki.

The former crypto billionaire had lost his exchange, his fortune, his political influence, his Manhattan trial, and, as of this month, his appeal. What he may also have lost — or more precisely, what FTX creditors may have lost — is one of the most spectacular venture-capital lottery tickets of the AI era.

Buried in the wreckage of FTX and Alameda Research was reportedly a tiny early stake in Anysphere, the company behind Cursor, the AI coding tool now at the center of a SpaceX deal valued at as much as $60 billion .

Alameda allegedly bought the position for about $200,000 in 2022. The FTX bankruptcy estate later sold it for roughly the same price. If that stake had survived and remained undiluted, it could now be worth about $3 billion .

That is the kind of math that makes bankruptcy lawyers wince, venture capitalists smile politely, and crypto Twitter reach for a flamethrower.

It also creates a deliciously awkward question: did Sam Bankman-Fried accidentally back one of the hottest AI startups in the world — only to watch the upside disappear from a prison cell?

The answer is messier than the meme. Bankman-Fried cannot credibly claim he was robbed of the asset. Prosecutors said FTX customer money was misused for venture investments, political contributions, real estate and other spending. He was sentenced to 25 years in prison, ordered to forfeit more than $11 billion, and the Second Circuit affirmed his conviction in June 2026.

So this is not the story of a genius investor cheated out of his prize. It is darker — and more interesting.

It is the story of how stolen customer money, a rushed bankruptcy cleanup, private-market opacity and the AI boom collided to create a multibillion-dollar “what if.”

Cursor was not always a $60 billion strategic prize. In April 2022, according to CoinDesk, Alameda invested about $200,000 in Anysphere at a valuation near $4 million , reportedly receiving about 5% of the company. At the time, Anysphere was an early-stage developer-tools startup. The product that made it famous, Cursor, had not yet become the default coding companion for AI-native engineers.

The estate was left with thousands of claims, assets, subsidiaries, legal disputes, crypto positions, tokens, venture investments and unresolved customer balances. The job was not to behave like a patient Silicon Valley fund. The job was to find assets, value them, sell them, litigate, recover money and return cash to creditors.

In that world, a $200,000 startup stake in an obscure coding-tool company may have looked like lint in the pocket of a burning coat.

Cursor raised a Series A in 2024 led by Andreessen Horowitz and Thrive Capital. It said it had more than 40,000 customers . In January 2025, Cursor announced a $105 million Series B , saying it had exceeded $100 million in recurring revenue . By June 2025, it raised $900 million at a $9.9 billion valuation . By November 2025, it announced a $2.3 billion Series D at a $29.3 billion post-money valuation .

SpaceX materials from its 2026 IPO roadshow described a partnership with Cursor to advance Grok and disclosed an option to acquire Cursor at an implied equity value of $60 billion . Subsequent reports said SpaceX agreed to acquire Anysphere in an all-stock transaction at roughly that value.

A stake once sold for the price of a nice SUV had become, at least on paper, the kind of asset that can change the outcome of a bankruptcy.

The viral version of the story is simple: Alameda bought 5% of Cursor for $200,000. FTX sold it for $200,000. SpaceX valued Cursor at $60 billion. Five percent of $60 billion is $3 billion.

That framing is irresistible. It is also incomplete.

Private startup stakes are not static. They dilute. Rights change. Preferred shares may convert differently. Early SAFEs and notes can be rewritten. Bankruptcy sales can include caveats. A buyer may assume risk. Documentation matters. Without the exact investment documents and sale terms, the clean “5% equals $3 billion” number should be treated as a back-of-the-envelope maximum, not a legal damages figure.

Still, even with heavy dilution, the missed upside could be enormous.

The most damning number is not necessarily $3 billion.

It is the gap between $200,000 and almost any plausible retained value if Cursor’s rise had been captured.

This is why the story has legs. It is not just another SBF anecdote. It is a stress test for the entire post-FTX recovery process.

Did the estate sell too early? Did it have enough information? Did it understand the asset? Could it have held? Were creditors better served by immediate liquidity or by optionality?

Was the buyer merely lucky, or did the buyer understand something the estate missed?

Those questions matter because FTX customers have already been told they are being made whole in dollar terms. But “made whole” is a slippery phrase in crypto bankruptcy. If your bitcoin was sold near the bottom and repaid later in dollars, you may be whole legally and still furious economically. The Cursor stake adds the venture-capital version of that same wound.

The Bankruptcy Problem: You Can’t HODL Everything

It is easy, in hindsight, to mock the estate.

Everyone is a genius after the chart goes vertical.

Bankruptcy administrators are not venture investors. Their mandate is different. They are typically rewarded for recovery, certainty, process and defensibility — not for taking moonshot risk. When a company collapses because of alleged fraud, the instinct is to turn weird assets into cash and stop the bleeding.

That instinct often makes sense. Most startup stakes die. Many never produce liquidity. A small stake in an unproven AI coding company in 2023 was not obviously a $3 billion recovery asset. It was one item on a very long, very ugly list.

But the Cursor case shows the danger of treating every illiquid asset like scrap metal.

Anysphere was not a random token. It was building in one of the fastest-growing software categories in the world. By 2024 and 2025, Cursor was becoming the workbench for AI-assisted software development. The company said its product was used by millions, and later said it had crossed $1 billion in annualized revenue .

This was not a rug-pull coin with a Telegram channel and a dead website. It was becoming infrastructure.

And infrastructure gets strategic buyers.

At first glance, SpaceX buying an AI coding tool sounds like Elon Musk alphabet soup: rockets, satellites, Grok, xAI, developer tools, Mars, maybe a flamethrower.

SpaceX has become more than a launch company. Its public-market story now sits at the intersection of orbital infrastructure, Starlink connectivity, AI compute, defense, robotics and software. Its IPO roadshow materials highlighted xAI, Colossus compute infrastructure, and a Cursor partnership to advance Grok. The same materials disclosed an option to acquire Cursor for an implied value of $60 billion.

Cursor brings three things SpaceX and xAI would want.

First, it brings distribution into the developer workflow. AI models are powerful, but the daily interface matters. Developers do not live in press releases. They live inside tools. Cursor sits where code is written.

Second, it brings enterprise AI revenue. Cursor said it had more than half of the Fortune 500 using the product by mid-2025 and later disclosed more than $1 billion in annualized revenue .

Third, it brings a strategic bridge between models and work. SpaceX and xAI can own compute and models. Cursor owns the moment when human engineers actually ask AI to change software.