Anthropic Made 5 Cents on the Dollar. OpenAI Lost $1.22 — and It’s the One Asking Wall Street for $1 Trillion.

Two artificial intelligence companies filed confidentially to go public this summer, within weeks of each other. One of them makes money.

It is not the one you would guess from the valuation.

Anthropic’s second-quarter figures, reported in mid-August, show the company clearing roughly $10.9 billion in revenue and posting an operating profit of $559 million — the first in its history, and by its own earlier projections about two years ahead of schedule. CNBC subsequently reported preliminary revenue of more than $11.5 billion for the same quarter. Either figure is a company that has stopped burning cash to serve its customers.

OpenAI, in the most recent quarter with public detail, brought in $5.7 billion and ran a non-GAAP operating margin of negative 122 percent. That is a loss of $1.22 for every dollar of revenue, and it is the flattering version — the figure already has stock-based compensation stripped out of it.

OpenAI is the one going to market at up to a trillion dollars.

The Two Sets of Numbers

Both companies are private, both sets of figures come from reporting rather than audited statements, and both should be read with that caveat attached. With that said, the gap is not subtle.

Read the second row before anything else. Anthropic did not grow into profitability gradually. Revenue went from $787 million in the year-ago quarter to more than ten times that twelve months later, and the margin flipped inside a single three-month window.

What Actually Changed Was 15 Cents

The temptation is to explain this with model quality, or enterprise sales, or some strategic masterstroke. The reported cause is duller and much more important.

In the first quarter of 2026, Anthropic spent 71 cents on compute for every dollar of revenue it booked. In the second quarter, that fell to 56 cents. Nothing else in the business moved anywhere near as much. Fifteen cents, in ninety days, is the entire distance between a loss-making AI company and a profitable one.

This is also why OpenAI’s numbers look the way they do. Its losses are, by the company’s own accounting, driven almost entirely by compute. Revenue of roughly $2 billion a month — about $25 billion annualized, up from $20 billion at the end of 2025 — is a genuinely enormous business by any standard except the one it is being measured against, which is its own cost of goods.

Growing Faster Is Not the Same as Growing Better

For three years the industry’s defence of AI economics has been that scale fixes it: get big enough, and the unit costs come down. Anthropic’s quarter is the first real evidence for that argument. OpenAI’s is the first serious evidence against a related one — that being biggest is what gets you there first.

OpenAI has more users, more consumer brand, more revenue than almost anyone predicted, and a projected loss of around $14 billion for 2026. Volume did not rescue the margin. Something about the cost of serving each request did not fall the way the thesis said it would.

The Case That the Profit Isn’t Real

This deserves airing, because the counter-argument is specific rather than sour.

Writing at Where’s Your Ed At, technology critic Ed Zitron argues the profit is an artifact of timing, not of a better business. His claim is that Anthropic negotiated a reduced compute rate covering exactly the months in which it declared profitability, and that its costs have historically scaled linearly with revenue — a relationship he says has not actually changed. The conclusion he draws is blunt: the result is “a result of accountancy rather than any improvements to its business model.”

Two things should be held alongside that. The $559 million is an adjusted operating profit and excludes stock-based compensation — though so does OpenAI’s $1.22, which keeps the comparison honest. And Anthropic has committed more than $130 billion to future cloud compute, reportedly over $100 billion to AWS across ten years and $30 billion to Azure. A margin built on introductory pricing has an expiry date somewhere inside those contracts.

The test is simply the next quarter. If the compute line holds near 56 cents when no discount window is open, the turn is real. If it snaps back toward 71, Zitron was right and one very well-timed quarter went into an IPO prospectus.

Two Filings, One Autumn

Both companies filed confidentially with the SEC in June.

OpenAI is targeting a listing as early as September, with Goldman Sachs and Morgan Stanley leading, at a valuation reported between $852 billion and more than $1 trillion. Anthropic’s investors are said to expect an October listing, with backers modelling valuations in the $2 trillion to $3 trillion range — Gavin Baker of Atreides Management said publicly he thinks the company is “worth $3 trillion today.”

Set those side by side and the surprise is which company is which. OpenAI has the household name, the consumer app and the larger share of public attention — and roughly half the revenue: about $25 billion annualized against Anthropic’s $44 billion. It is also the one with the negative margin, and the lower valuation.

Strip the branding away and the market is doing something unusual for an IPO window: pricing the quieter enterprise business above the famous consumer one, on the strength of a single quarter of positive margin. Previous technology cycles priced attention first and asked about unit economics afterward.

It is also a bet that public investors will hold that line once the lockups pass, which they have historically not done.

What This Means If You Just Use the Products

Prices stop being arbitrary.

Every model subscription and API rate you pay today was set by a private company answering to investors who tolerate losses. Once these are public companies with quarterly margin disclosures, the free tier, the rate limit and the per-token price all become things a chief financial officer must defend to shareholders every ninety days. The 80 percent price collapse the industry saw in July was a private-market move. Public markets do not reward those the same way.

What To Watch

• Anthropic’s next compute line. 56 cents per revenue dollar holding through a quarter with no discount window is the single fact that settles whether this profit was structural. Anything near 71 confirms the skeptics.
• Whether OpenAI publishes a path, not a promise. A public filing has to state when losses end and on what assumptions. That sentence will be the most scrutinised in the prospectus.
• The order of the listings. September then October means OpenAI prices first and sets the comparable. If it prices well despite the margin, the market is buying scale after all.
• Reconciliation of the revenue figures. $10.9 billion projected against more than $11.5 billion preliminary is a $600 million spread in reporting on the same quarter. Audited numbers will pick one.
• What happens to prices after the first earnings call. The first time either company has to defend gross margin in public is the first time your bill is a governance issue.