Anthropic is gearing up for a mega funding round that could value the company at over $900 billion. But while the world's hottest AI startup is being flooded with cash, its first CFO, Krishna Rao, is doing something counterintuitive: he keeps telling investors the company isn't raising yet. According to The Wall Street Journal, Rao has repeatedly turned away potential backers, deliberately rejected the highest valuation on the table, and chosen to raise less money than expected — all while offering a deliberately conservative revenue forecast.
Rao is the man who cleans up the mess for CEO Dario Amodei — the guy who writes 10,000-word philosophical essays and publicly bickers with the White House. He joined Anthropic in May 2024, but his most famous deal was a much more desperate one: in 2020, during the pandemic, he negotiated a $1 billion emergency loan for Airbnb to keep it from collapsing. Before that, at Blackstone, he was known for pretending to be a novice at Texas Hold'em only to clean out his boss's chips — earning him the nickname "Chairman Rao." One investor says Rao always knows exactly when to step in and when to step back.
That discipline is facing its ultimate test right now. With Anthropic's Claude Code exploding in popularity, the company's April revenue run rate has already surpassed $30 billion. The staggering demand means Rao has to make a gut-wrenching call: whether to lock in hundreds of billions of dollars in compute contracts for 2027 through 2029. If he goes all-in and demand slows down, those hardware commitments become a devastating depreciating liability.
This kind of constant pressure has consumed his personal life. When he asked his mentor from Yale Law School for advice on balancing work and family, the answer was blunt: "You cannot find balance anymore. This is a company making history, and you are in the room where it's happening. Just ride it."
Meanwhile, Anthropic has been pushing for greater AI transparency, recently open-sourcing its natural language autoencoder (NLA) that translates model activations into plain English.