This Time It's Different

The Voting Machine Always Pays the Weighing Machine

Scott Galloway called the dot-com crash from a faculty meeting while his colleagues were buying Pets.com. This week he put a number on AI: valuations come down 50% to 70% inside two years. Every doomer on X grabbed it and started dancing on the grave. They grabbed the wrong half.

Ben Graham gave us the only framework that matters here, the one Buffett quotes every chance he gets. In the short run the market is a voting machine. In the long run it's a weighing machine. The voting machine ran Anthropic from $300 billion to a trillion dollars in twelve months, because venture capital is a momentum business and when the momentum is good there is no price too high. That was the vote. Now the weighing machine clocks in.

So weigh it. Anthropic does about $47 billion in run-rate revenue and not a dime of real free cash flow yet. Buy it at a trillion and you have to believe it becomes a two-trillion-dollar company just to make your money back. Two trillion at a mature 20x cash flow is $100 billion in free cash flow, which at fat 25% margins is $400 billion in revenue. Eight times where they sit today. The guy who bought at $300 billion needs two-and-a-half times. Same company, same models, same week. The only thing that moved is the price the buyer has to justify, and that gap is the whole correction.

Here's where I get off Galloway's bus, or at least move to the back. He's right the price is frothy. He is not saying AI is fake, and neither am I. Tulips were flowers. A Bored Ape is a JPEG. Pets.com was dog food sold at a loss and a sock puppet. This is the first mania in history with a real industrial revolution sitting under the froth, which is exactly why it ran further and faster than tulips, the South Sea, the Nifty Fifty, Japan, or dot-com ever managed. The weighing machine reprices the stock. It does not repeal the technology. The same 2000 crash that killed Pets.com minted Amazon.

And the kicker nobody puts next to the scary number: a 50-70% haircut drops OpenAI and Anthropic to $300-500 billion. That's not the apocalypse. That's where they were priced twelve months ago. The anomaly was never the floor. It was the climb.

So no, this time isn't different. The technology is. The price never is.

At COAI today: the full Signal/Noise — the Anthropic cash-flow math, the China inference war, and why Apple quietly wins the on-device round — is live at getcoai.com.

Outsider Labs Advisory →

50–70%. That's how far Scott Galloway says AI valuations fall over the next two years. Sounds like a crater until you do the arithmetic: knock 50-70% off OpenAI and Anthropic and you land at $300-500 billion, exactly where they were priced twelve months ago. The voting machine ran them up a full year in a year. The weighing machine just wants the year back. The technology keeps compounding the entire time. This time it's different. The price isn't.

Read your AI bill the way Uber's CEO just did. Dara Khosrowshahi told investors Uber burned its entire 2026 AI budget in one quarter. His move wasn't to cut AI. It was to freeze hiring to pay for it. Read that correctly: AI just won a budget fight against payroll. Audit where your tokens go this week, then name the headcount line they replace.

Put a cheap model under your expensive one. Lindy moved 100% of its traffic off Anthropic to DeepSeek v4 and came out faster and cheaper. Fireworks ran a small GLM worker that calls Opus only when it has to — better scores at 39% of the cost. Stop paying frontier prices for median work. Route it, and start today.

Underwrite like the early buyer, never the last one. Buffett just put $10 billion into Alphabet, the toll booth, the same week Galloway said sell the cars. Both can be right. Before you buy any AI name, ask what growth the price forces you to believe. If the answer is eight times, pass. If it's two-and-a-half, look harder.

Thursday, June 4

Anthropic beats OpenAI to the IPO line and hands Wall Street a $3.5T problem. The first pure-play lab filed near $960 billion. Stack SpaceX and OpenAI behind it and you get the biggest liquidity stress test in decades, none of the three profitable. (Full analysis above.)

Apple wins the round nobody was watching. Google's Gemma 4 12B runs on a 16GB MacBook. The Deep View got Apple's Mac hardware chief on record: mini backorders out four to five months, 90% of frontier-lab staff on Macs. When the model shrinks to the device, the landlord is Cupertino.

Gary Marcus calls the death of "tokenmaxxing." On a report that Altman warned of a severe, first-of-its-kind pullback in AI spending, Marcus flagged it as a real problem for all three big IPOs. When the meter spooks the buyers, the window spooks the bankers.

Bain: the AI worked, the savings didn't show up. A new survey says deployments run exactly as designed, but the modeled savings aren't hitting the income statement on schedule. That's not a tech failure. It's a forecasting failure — tuition, not a tombstone.

Bernie Sanders wants the public to own half the labs. His bill would tax 50% of OpenAI, Anthropic, and xAI equity into a public wealth fund with board seats. It won't pass. That a senator can say it and not get laughed off the floor is the story.

— Harry and Anthony

Sources: