米湯
世界真小. It's a small world!
https://x.com/aakashgupta/status/2053845623397552439
"OpenAI authorized $10.3 billion in employee stock sales last October. Employees only sold $6.6 billion. They left $4 billion on the table because they think the next round prints higher.
The tender priced at a $500 billion valuation. 600 current and former employees cleared an average of $11 million in personal liquidity. That single transaction accounted for 6.2% of all 2025 US secondary trading value by itself, per PitchBook.
The trajectory is what's powering the holdout behavior. January 2023, the internal mark was $29 billion. February 2024, $86 billion. October 2024, $157 billion. Early 2025, $300 billion. October 2025, $500 billion. March 2026, OpenAI raised at $852 billion post-money.
17x in 32 months. The default at OpenAI is now to hold.
That is the actual problem the tender exists to solve. OpenAI cannot let its top engineers sit on tens of millions of dollars in illiquid PPUs, watch Mark Zuckerberg dangle nine-figure signing bonuses across the bay, and have no cash on hand to ground them. Meta's Superintelligence Labs has been recruiting from the OpenAI cap table at packages reported above $100 million for senior researchers.
The tender is a counter-bid. Sam is keeping people too rich to defect and too invested in the next mark to leave.
The retention math gets clearer when you stack the alternatives. Anthropic ran the same play last year at $350 billion and saw the same holdout behavior. SpaceX ran one in December at $800 billion. Stripe, Databricks, Fanatics, all of them. Tender offers in US startups hit $18.4 billion in 2025.
Facebook ran tenders in 2009 and 2011 before going public in 2012. Uber ran one in 2017 before going public in 2019. The pattern is compulsory at any company staying private past year 7.
600 people. Eleven million dollars apiece. Four billion turned down because the curve is still going up.
The IPO is now the last tender."
