Anthropic Turns Profitable While OpenAI Preps a $1T IPO in the Red

Two numbers landed within days of each other this month and they tell almost opposite stories about the same industry. Anthropic reported more than $11.5 billion in second-quarter revenue and, for the first time, a positive operating income. OpenAI, meanwhile, is moving toward a public listing that could value it above $1 trillion, while it is on track to lose around $14 billion this year. Both companies make large language models. Both are burning enormous sums on compute. Only one of them, so far, has a quarter that adds up.

What Anthropic actually reported

CNBC, Bloomberg, and Fortune each cited figures putting Anthropic’s second-quarter revenue at more than $11.5 billion, up from $787 million in the same period a year earlier and up from $4.73 billion in the first quarter of 2026. That is not a typo: revenue grew more than 14-fold year over year. Alongside that growth, Anthropic posted positive adjusted operating income for the quarter, which the company and multiple outlets described as its first profitable quarter.

The company had told investors back in May to expect roughly this outcome, so the number itself is not a surprise. What is notable is the shape of the improvement. Reports put Anthropic’s gross margin moving from around negative 94% in 2024 to roughly 60% now, and attribute most of that shift to inference efficiency gains rather than to price increases. In plain terms: Anthropic is spending less per unit of usage to serve the same requests, not simply charging customers more for the same product. Whether that margin holds once the company’s next round of data center buildout lands on the books, including the multi-billion-dollar joint venture with Theseus we covered earlier this month, is a separate question the quarterly number does not answer.

What OpenAI actually filed

OpenAI’s picture runs the other direction. The company filed a confidential S-1 registration with the SEC in June, formally starting the process toward a public listing. Reporting puts the target valuation above $1 trillion, with a listing window that could open as early as September 2026, though some more recent coverage suggests the timeline may slip toward 2027. OpenAI has reportedly brought in Goldman Sachs and Morgan Stanley to manage the offering.

The financial backdrop for that filing is a company generating roughly $2 billion a month in revenue, or about $25 billion annualized, against a projected loss near $14 billion for the year. Because the S-1 is confidential, the full breakdown of costs and margins is not yet public and likely won’t be until closer to the actual listing. What is publicly known lines up with a pattern OpenAI has shown for years: revenue growing fast, losses growing alongside it, and a business model that depends heavily on continued external financing to bridge the gap. We wrote in July about one piece of that financing, the arrangement giving the U.S. government an equity stake, and about Nvidia’s role backstopping OpenAI’s data center leases. Both are pieces of the same underlying fact: OpenAI’s spending is running well ahead of what its subscription and API revenue currently covers.

Why the same industry produced two different quarters

Part of the gap is business mix. Reporting attributes over 65% of OpenAI’s revenue to consumer subscriptions, with roughly 900 million free ChatGPT users that cost money to serve but generate no direct revenue, a combination that reportedly drags OpenAI’s gross margin down by 20 to 30 percentage points compared to a pure enterprise API business. Anthropic’s revenue leans more heavily on enterprise and API customers, a smaller and more expensive-to-acquire base, but one where usage is more reliably paid for. Neither structure is objectively better. They are different bets on where the money in AI actually is, and this quarter Anthropic’s bet produced a profit while OpenAI’s produced a bigger loss.

It’s worth being careful about what “first profitable quarter” does and does not mean. One quarter of positive operating income is not the same as a durable business. Anthropic itself has reportedly told investors it may not stay profitable for the rest of the year, given the scale of compute commitments still ahead of it, including the chip-speed deal with Decart we covered in mid-August. Compute costs for frontier labs do not move in a straight line; a single large infrastructure commitment landing in the wrong quarter can erase a margin gain built up over several prior ones.

The bigger question this sets up

Both companies are heading toward public markets on different clocks, and public markets will ask a blunter question than either has faced as a private company: not “how fast are you growing” but “when, concretely, does this become a normal business.” Anthropic now has one data point suggesting an answer sooner rather than later. OpenAI, at a valuation more than double Anthropic’s roughly $850 billion mark from earlier this year, is betting that scale and market position matter more to investors than near-term profitability, the same bet many large tech companies made successfully in past cycles, and the same bet a fair number of pre-profit companies have made unsuccessfully.

Neither trajectory is resolved by this month’s numbers. Anthropic’s margin gains depend on further inference efficiency improvements holding up against ballooning infrastructure spend. OpenAI’s IPO will live or die on whether public investors are willing to fund a $14-billion-a-year loss at trillion-dollar scale the way private investors have been. Both are testable claims, not settled ones, and the coming two or three quarters from each company will say a lot more than this one snapshot does.

Sources: CNBC, Bloomberg, Fortune, TechCrunch, Bloomberg on OpenAI’s $1T IPO target.