The Hundreds of Decamillionaires of OpenAI: Inside Silicon Valley's Great Cash-Out Divide

The most consequential wealth event in technology this cycle did not happen on a public exchange. It happened on a cap table. Last autumn, OpenAI ran a secondary tender that let more than 600 current and former employees sell roughly $6.6 billion of stock into a company then valued around $500 billion. No IPO bell, no lockup, no S-1 to pore over. Just wire transfers, quietly, into a few hundred personal accounts.

The arithmetic is worth sitting with. Around 75 of those sellers hit the tender's $30 million individual cap. The remaining participants divided the rest, averaging something north of $8 million apiece. Fold in the earlier tenders OpenAI has run since 2021, and the cumulative picture, as SaaStr lays out, is somewhere between 300 and 500 people who have already pulled $10 million or more in realized, spendable cash out of a private company. Not paper. Not a valuation on a slide. Cash.

What 100x actually buys

Equity in OpenAI has appreciated on the order of 100 times since 2019. A mid-level engineer who joined that year, took the grant, and simply stayed is now sitting on a position plausibly worth $50 million to well over $200 million, most of it still unsold. The tender is what turns that abstraction into a house paid for in one wire, a family office, a sudden and permanent change in how a thirty-four-year-old relates to risk.

This is the part outsiders miss about the frontier-lab boom. The genuine fortunes are not accruing only to a handful of founders and the venture firms. They are being distributed, in the low nine figures on the high end and the low eight figures across a broad middle, to hundreds of employees who will never appear on a rich list and who, in many cases, have told almost no one. Silicon Valley has minted a quiet cohort of decamillionaires who still take the same standup meeting on Monday.

Two labs, two philosophies of money: sell into strength, or hold for the bell.

The Anthropic bet

Which makes the contrast down the road so telling. Anthropic ran its own tender this spring, pricing around $350 billion and targeting a $5 to $6 billion secondary. It came in under target. Some of that is mechanical, but a meaningful share is temperament: a lot of Anthropic staff chose not to sell. With an IPO widely expected before the year is out and an implied valuation approaching $1 trillion on secondary markets and in its most recent rounds, holding the stock is a wager that the public-market payday will dwarf what a private buyer will pay today. Sell now and you lock in certainty. Hold and you are betting the bell rings higher.

Neither instinct is naive. OpenAI's approach treats liquidity as a retention tool and a hedge, a way to let people de-risk their lives without waiting years for an exit that keeps receding. Anthropic's quieter posture reflects a workforce more willing to keep chips on the table and a leadership comfortable letting them. The result is two frontier labs, comparable in ambition and talent, developing genuinely different cultures of wealth. One has already made its people rich in the checking-account sense. The other has made its people rich on paper and asked them to believe.

The divide that outlasts the money

By the time both stories fully resolve, the dollar outcomes may converge; a strong Anthropic listing could vindicate every share held. But the behavioral split will linger. A person who has taken $30 million off the table negotiates, invests, and stays or leaves differently from a colleague whose net worth is still contingent on a filing. Certainty changes people. So does conviction. The two labs are, in effect, running a live experiment on which one changes them more.

What both cases confirm is older than any language model: the largest gains are captured well before the public is invited in, by the people who were simply in the room when the equity was handed out. The tender, the pre-IPO round, the cap table itself are all instruments of proximity. The returns follow access, and access follows the network. It always has.