$852 billion and a guess
OpenAI just closed a $122 billion round. Amazon put in $50 billion. Nvidia, $30 billion. SoftBank, $30 billion. The implied valuation: $852 billion. Monthly revenue: $2.6 billion. Weekly active users: 900 million.
Those are the numbers. Now here’s the question nobody at the cap table is answering publicly: what, specifically, are they buying?
Not the product. The product is a chat interface and an API. You don’t pay $852 billion for a chat interface. You don’t pay it for $2.6 billion a month in revenue, either — that’s a 27x revenue multiple on a company that hasn’t demonstrated it can hold margins when the subsidy era ends.
What they’re buying is optionality on a thing that doesn’t exist yet.
The bet isn’t “AI can do X.” AI can already do X. The bet is “AI will become Y” — where Y is undefined, unproven, and carrying the entire valuation on its back. That’s not investment in the traditional sense. That’s faith with a term sheet.
Pascal’s Wager works like this: you can’t prove God exists, but the asymmetry of outcomes means belief is the rational bet. Infinite upside if right, finite cost if wrong.
Infrastructure capital has discovered the same structure. You can’t prove AGI arrives, or that it arrives through this particular company, or that the moat holds when it does. But if it does — if the thing these models are becoming is as transformative as electricity or the internet — then $852 billion will look like a rounding error. The asymmetry of belief does the rest.
The logic is clean. The problem is that clean logic and correct logic aren’t the same thing.
Every infrastructure cycle produces this moment. The moment where capital commits before theory does. Before the economics are proven. Before anyone can draw a straight line from the current product to the implied future.
Railroads in the 1840s. Fiber optic in 1999. The math always looks the same: massive upfront capital, a growth curve that hasn’t yet inflected to profit, and a story compelling enough that “what if we’re wrong” gets drowned out by “what if we’re right and we weren’t in.”
Sometimes the infrastructure pays off. The railroads connected the continent. The fiber carries the internet you’re reading this on.
Sometimes it doesn’t — or at least, not for the people who wrote the checks. The fiber was real. The companies that laid it went bankrupt. The asset survived; the cap table didn’t.
What makes this round different from a generic bubble is that the product works. ChatGPT isn’t vaporware. 900 million weekly users aren’t imaginary. The question isn’t whether AI is real — it is. The question is whether “real and useful” justifies $852 billion, or whether the gap between “useful now” and “transformative later” is being filled entirely by narrative.
Revenue grows. Users grow. Costs grow faster. The path from here to a margin structure that justifies the valuation requires either a cost miracle or a capability miracle — a jump from “very good tool” to “something qualitatively different.” The round is priced for the miracle.
That’s the bet. Not a stupid one. Not a safe one.
$852 billion and a guess that what’s coming is bigger than what’s here. The capital is already committed. The theory will either catch up or it won’t.
The term sheet doesn’t wait for proof. It never has.