OpenAI Filed to Go Public — at an $852 Billion Question

On June 8, OpenAI did the thing it has spent two years insisting it was in no hurry to do: it told the world it had confidentially filed to go public. The company confirmed a draft S-1 registration statement was already sitting with the SEC, and it confirmed it the most OpenAI way possible — in a blog post, because it expected the filing to leak and decided it would rather break its own news. The reported valuation is roughly $852 billion. Several analysts think the public number could clear a trillion. Either way, it would be one of the largest IPOs in the history of capital markets.

And it arrives in a two-week window that has started to look less like a coincidence and more like a stampede. Anthropic filed days earlier at a reported $965 billion. SpaceX is mid-roadshow on a listing that could value the company near $1.75 trillion and raise as much as $75 billion. Three of the most valuable private companies ever built are all sprinting for the public markets at once, after years of telling everyone the private markets were serving them just fine. When the smartest money in the room all decides to sell at the same moment, it is worth asking what they can see that the buyers can't.

The number under the number

The valuation is the headline. The income statement is the story. OpenAI's annualized revenue has reportedly crossed $25 billion — a staggering figure for a company that was a research lab with a chatbot three years ago, and genuinely one of the fastest revenue ramps any software company has ever posted. But the same financials reportedly show the company losing about $1.22 for every dollar it brings in. It is not a business that is profitable and growing. It is a business that is buying growth at a loss, at enormous scale, and betting that the loss inverts before the patience does.

That is not, by itself, disqualifying. Amazon ran thin or negative margins for years on purpose. The difference is the size of the bet and the cost of the inputs. OpenAI's marginal cost is not warehouses and trucks; it is compute, and compute is getting more expensive at the frontier, not less, because every capability jump demands a bigger cluster than the last. The bull case is that inference costs collapse and the $25 billion top line keeps compounding until the math flips. The bear case is that the company is structurally short on the one resource — cheap, abundant compute — that its entire valuation assumes it will eventually have.

Why the rush, and why now

The cynical read is the simplest: private valuations have run so far ahead of any plausible near-term cash flow that the only buyer left big enough to absorb the next markup is the public. Secondary sales and late-stage rounds can only recycle so much paper among so many funds. An IPO opens the float to pensions, index funds, and retail — a far deeper pool, and one that has historically been willing to pay for a story when the story is good enough.

The less cynical read is that the AI labs genuinely need the capital and the IPO is the cleanest way to raise it at this scale. Training runs are now budgeted in the tens of billions. Meta alone has guided to $115–135 billion in AI capital expenditure for 2026, roughly double the prior year. In an arms race where the table stakes are measured in data-center build-outs, going public is less an exit than a refueling stop. You do not list a company like this to cash out. You list it to keep paying for the next model.

The real test

What makes this moment matter beyond OpenAI is that it is the first honest referendum on AI valuations. For three years, the prices have been set by a closed loop of venture funds, sovereign wealth, and strategic investors marking each other's books upward. None of those prices were ever tested against an open market of skeptics who could short the stock and vote with their dollars the other way. The public markets are exactly that test. They are where narrative meets a bid-ask spread.

If OpenAI, Anthropic, and SpaceX all price at or above their private marks and trade up, it validates the entire AI capital structure and pulls even more money into the sector — possibly at the expense of other risk assets, which is why crypto desks are already nervously modeling the rotation. If any of the three prices below its private valuation or breaks on debut, the repricing ripples backward through every late-stage AI round still sitting on a fund's books at a number nobody has had to defend in public.

The listing window points to September through November, with September the earliest plausible date, and the usual suspects — Goldman Sachs, Morgan Stanley, reportedly JPMorgan — lined up to underwrite. That gives the market a summer to make up its mind about a company that is, by its own filing, growing faster than almost anything in history and losing money on every unit it sells.

The question on the cover of the prospectus is whether $852 billion is too much. The question underneath it is more interesting: in an industry where the cost of staying at the frontier rises every year, is there a price at which a frontier lab is actually a good business yet — or is the IPO itself the admission that nobody knows? OpenAI is about to find out in the most public way there is.