OpenAI IPO valuation: what the confidential S-1 actually shows

Negative 122%. That is the non-GAAP operating margin sitting inside the numbers reported around OpenAI's IPO preparation, and in plain terms it means the company lost roughly $1.22 for every dollar of revenue it booked in the first quarter.

OpenAI submitted a confidential S-1 to the SEC on 8 June and is working toward a listing as early as September, with Goldman Sachs and Morgan Stanley advising. The company has been careful to say the timing is not settled, and that some of what it wants to do next is easier to do as a private company. Both things can be true at once.

The OpenAI IPO valuation range is the first tell

Reported targets run from around $730 billion to over $1 trillion. That is not a negotiation gap, it is a modelling gap. A $270 billion spread means the people pricing this cannot agree on which multiple applies, because there is no comparable company with this revenue shape, this growth rate and this cost base.

For context, a $1 trillion debut would make OpenAI worth more than almost every company that has ever listed, at a point where it has been selling a consumer product for under four years.

What the financials actually say

The figures being reported break down roughly like this:

• Around $6 billion of revenue in Q1
• An annualised run rate near $25 billion
• More than 230 million weekly ChatGPT users
• A non-GAAP operating margin of negative 122%

Read those in order and you get two different companies. The first three lines describe one of the fastest revenue ramps in software history. The fourth line describes a business where scale has not yet started paying for itself.

That is the part worth sitting with. Most software businesses get cheaper per user as they grow, because the marginal cost of serving one more customer rounds to nothing. Inference does not work that way. Every additional conversation costs real compute, and the frontier models that drive the subscriptions are the most expensive ones to run.

Why the loss is not automatically a red flag

Amazon lost money for years on purpose, and the market eventually rewarded it. The argument for OpenAI is the same shape: the losses are training runs and data centre capacity, which are investments in a future cost curve rather than an operating hole.

The argument against is that the capacity spending is not slowing down. The same buildout showing up in big tech's 2026 AI spending commitments and in Samsung's $648 billion chip investment is the cost side of OpenAI's income statement. If the compute bill keeps growing with usage, the margin does not fix itself, it just gets bigger in both directions.

Why this filing matters beyond OpenAI

Right now every AI valuation in the private market is a guess anchored to other guesses. A public listing replaces that with audited numbers and a live share price, and every subsequent fundraise gets priced against it.

It also settles an argument that has been running on vibes. The question of whether frontier AI is a viable business or an extremely well-funded science project has been unanswerable because nobody outside these companies could see the unit economics. A prospectus is where that stops being a matter of opinion.

The user numbers make the same point from the other side. More than 230 million weekly users is a genuine consumer platform, roughly in the territory of what a billion monthly Gemini users looks like on a different counting basis. Reach was never the problem. Turning that reach into a positive margin is.

What to watch next

Three things. Whether the September window holds, or slips again. Whether the final valuation lands nearer $730 billion or past $1 trillion, because the gap between those two outcomes is the market's verdict on AI as an asset class. And whether the public S-1, when it appears, shows the loss margin narrowing quarter on quarter or holding flat.

That last one is the only figure that really matters. Everything else is a headline.