By Silicon Starlight
In the long history of wealth creation in Silicon Valley, no company has allowed so many employees to cash out such a staggering amount of wealth in a single event before going public.
According to a May 10 report by The Wall Street Journal, in October 2025, over 600 current and former OpenAI employees sold a total of $66 billion worth of shares on the secondary market. Among them, about 75 individuals sold up to the $30 million per-person limit, while the remaining roughly 525 people cashed out an average of about $8.3 million each.
This is the first systematic "major cash-out" of the AI era.
The old social contract in Silicon Valley was simple and long: join an early-stage company, work diligently for seven years, wait for the IPO, wait for the lock-up period to end, and then cash out.
When Google went public in 2004, it created over a thousand paper millionaires, but they could not access their wealth until after the lock-up period. Facebook followed the same pattern. Even Snowflake, Datadog, and MongoDB — some of the strongest B2B IPOs of the past decade — produced only a handful of multi-millionaires after lock-up, not hundreds.
OpenAI skipped all those steps.
The scale of this transaction already exceeded any formal IPO in the US market in 2024. The largest IPO that year, Lineage, raised only $4.8 billion. An artificial intelligence company accomplished a "shadow IPO" through a single internal secondary sale of existing shares.
The script for this liquidity is extremely simple: employees must have held their shares for two years before selling. This means a significant number of employees who joined after the release of ChatGPT participated in this transaction, transforming paper wealth into bank balances for the first time. Some of them had only worked at the company for two years before receiving a cash return that would typically require a decade of waiting for a startup founder.
For OpenAI, this is the most direct form of retention tool. Competitors are poaching talent with extremely aggressive offers. According to previous reports, Meta once offered top AI researchers compensation packages of $300 million over four years, along with signing bonuses as high as $100 million. OpenAI's response has been almost blunt: we don't make our employees wait for an IPO. Here, work for two years, and you can walk away with $30 million in cash.
This raises a question. While opening the cash-out valve can retain some people, it will inevitably wash others away.
This batch of transactions occurred when the company's valuation was approximately $400 billion. Less than six months later, by March 2026, OpenAI completed a $122 billion funding round, skyrocketing its valuation to $852 billion. Veteran employees from as early as 2019 have seen their share value appreciate over a hundredfold. Those who cashed out substantially before this valuation surge actively forwent potential fair value gains in the coming decades; those who held off, waiting for the next PE round, face the risk of a sudden shift in the company's fundamentals.
This is precisely the deep-seated dilemma that has surfaced after the first wave of cash-outs. Silicon Valley did face the issue of employee attrition after IPO-fueled wealth in the past; Google worried about "brain drain" upon its listing. But OpenAI faces a more complex proposition: a group of people achieved financial freedom before an IPO. Competitors might trigger a wave of departures with offers even slightly less than the current value of their retained equity. The only countermeasures might be an even more extreme sense of mission, or a more thorough cultural cohesion.
In stark contrast to OpenAI stands Anthropic.
Anthropic also conducted a secondary employee share sale in April 2026 at a $350 billion pre-money valuation, but its scale was far smaller than OpenAI's: investors wanted to buy more shares from employees, but they were unwilling to sell.
On one side, there's a rush to cash out; on the other, a reluctance to sell. The two AI labs have placed starkly different private bets on their own futures. These two distinct employee behaviors correspond to two different corporate valuation narratives.
Because there's another, even more glaring narrative, concerning the financial fundamentals.
OpenAI's CFO, Friar, publicly acknowledged that the company's annualized revenue in 2025 exceeded $20 billion, a more than 230% increase from $6 billion in 2024. Monthly revenue is around $2 billion, with weekly active users exceeding 900 million. However, Goldman Sachs points out its expected cash burn for 2026 is approximately $7 billion to $17 billion. Other estimates suggest full-year 2025 revenue of about $13.1 billion with a loss of around $8 billion; a 2026 loss is expected to be $14 billion, with positive cash flow potentially delayed until 2030. The company also carries long-term obligations to pay Microsoft a 20% revenue share, lasting until 2032; this expense is projected to exceed $13 billion in 2026 and 2027 combined.
Now, consider Anthropic. By the end of 2025, its ARR was approximately $9 billion, rising to $14 billion in February 2026, $19 billion in March, $30 billion in April, and $44 billion in May. Inference gross margin increased from 38% to over 70%. The number of enterprise customers spending over $1 million exceeded 1,000, growing sevenfold in the past year. Its relative share in enterprise AI spending jumped from about 10% in early 2025 to over 65% in February 2026. The company expects to achieve profitability by 2028.
OpenAI's valuation framework is anchored at one end to the rocket-like valuation from its funding narrative, and at the other end rests on the knife-edge of talent attrition risk post-massive cash-out and years of ongoing financial deficits. It's akin to simultaneously pressing the accelerator and the brake, where every meter of forward thrust consumes the internal resolve of the organization.
Greg Brockman revealed he holds shares worth about $30 billion. This $30 billion, along with OpenAI's trillion-dollar IPO ambition, has become the target in Elon Musk's lawsuit.
This is no longer just a war of AI code. This is a war of AI capital, the most expensive human experiment in San Francisco. When tens of billions of dollars flow from paper into real bank accounts, transforming from contract clauses into single-family homes on Bay Area hills and donor-advised fund charity lists, people finally see it: the most extreme algorithms are often not in the models, but in people's calculations of greed and fear.
When algorithms get too close to power and money, they cease to be purely algorithms.
