OpenAI Q2 sales hit $6.7B as margins sink
The booked quarter is in, and it does not look like the annualized story OpenAI has been selling.
OpenAI's second-quarter revenue grew 18 percent from the first quarter to $6.7 billion, according to people familiar with the figures cited by the Wall Street Journal — and its operating margin got worse, though the paper did not say by how much. Some investors were disappointed. The company told them growth picked up again in the third quarter. That is a different object from the $40 billion annualized run rate we tracked last week — OpenAI's revenue run rate tops $40B ahead of IPO — and the gap is the news. A run rate annualizes a hot stretch. A quarter is what actually hit the books. Eighteen percent sequential growth on $6.7 billion is still a huge software business. It is not the slope implied by a $40 billion run rate, and it is not the story a company wants circulating while it is still in confidential IPO paperwork.
The Journal's reporters framed the print as a check on vague ARR talk, and they are right to. OpenAI has spent the past week telling shareholders that enterprise has overtaken consumer and that July run rate was still climbing 20 percent month over month — OpenAI's enterprise revenue overtakes consumer, CFO tells investors. Those claims can all be true at once and still leave a tepid booked quarter. Compute bills do not care about annualized headlines. If the margin is still sliding after a $6.7 billion quarter, the cost of serving the product is winning the race against the price of selling it. The missing number is the one that matters: how negative, and how much more negative than last quarter.
The comparison the Journal chose is the one public-market buyers will use. Anthropic more than doubled revenue to $11.6 billion in the same period — the same second-quarter print we already covered as more than $11.5 billion, now sitting under a $65 billion run rate — Anthropic hits $65B run rate, up $18B in two months. Two labs, two books, one league table. OpenAI still has ChatGPT and the consumer brand. Anthropic has the faster booked quarter and, on the figures now in circulation, the better operating picture. OpenAI's rebuttal is the Q3 acceleration. Until that shows up as a number, the IPO narrative just acquired a dent.
Z.ai's GLM-5.3, on maximum reasoning, scored 60 on the Artificial Analysis Intelligence Index — tied with Kimi K3, three points behind Anthropic's Opus 5 and two behind Fable 5. That is a seven-point jump from GLM-5.2, and once the weights land it will be tied as the leading open-weights model on that board. We covered the launch and its cyber tools last week — Zhipu's GLM-5.3 ships with 'emergent' cyber capabilities. The independent score is the new fact: a Chinese lab is now inside shouting distance of the two closed models that have owned the top of the index, at $1.40 per million input tokens and $4.40 per million output. It is also a chatterbox — 170 million output tokens on the index versus a 72 million median — so the sticker price is not the bill.
What to watch: whether OpenAI puts a booked quarterly figure, not just a run rate, in the S-1 — and whether GLM-5.3's open weights keep that 60 when anyone can run them.
If the booked quarter is $6.7 billion and the pitch is a $40 billion run rate, which number should price the IPO? Tell us in the comments.
Sources: Wall Street Journal · Techmeme · Bloomberg — Anthropic run rate · Artificial Analysis — GLM-5.3 · Techmeme — GLM-5.3
