$65bn vs $40bn: How Anthropic Overtook OpenAI on Revenue
Trending Topics Jakob Steinschaden
Anthropic's revenue run rate just topped OpenAI's, $65bn versus $40bn, per Bloomberg. The kicker: it's the coding-agent bet, not model quality, that flipped the script.
Based on reporting by Trending Topics, Jakob Steinschaden — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Numbers like these deserve a squint before a nod. Bloomberg reports Anthropic hit an annualised revenue run rate north of $65 billion at the end of July, based on unnamed sources, while OpenAI's equivalent figure, disclosed last Thursday, sits above $40 billion. Both companies are gearing up for public offerings, and both run rates are extrapolations from a short window of earnings rather than audited totals — Bloomberg itself flags that the two firms may not even calculate the metric the same way. So the absolute gap matters less than the shape of the climb.
And that climb traces back to a very personal split. Anthropic's founders, including Dario and Daniela Amodei along with Tom Brown, Jared Kaplan, Sam McCandlish, Chris Olah, Ben Mann and Jack Clark, all left OpenAI over disagreements about direction after Microsoft's investment and how to approach safety. What looks like a corporate rivalry today started as a falling-out among people who used to work in the same room.
The two companies then chose almost opposite audiences to chase. Anthropic went after developers first, launching Claude Code in early 2025 as an autonomous terminal agent — at a time when the company was running around $1.4 billion. Developers could put a number on what the agent saved them, so they bought seats for entire teams without haggling over price. Anthropic then repeated the trick for non-coders with Claude Cowork in early 2026, followed two weeks later by eleven industry-specific plugins covering sales, finance, legal, marketing and more. The market's reaction was brutal: software stocks shed roughly $285 billion in market cap within days, the North American software index fell 15 percent that January — its worst month since 2008 — and Figma dropped as much as 40 percent year-to-date. A Jefferies strategist even labeled it the
My take — AI-written commentary, not fact-checked reporting
, is that the real story here isn't model supremacy but go-to-market discipline. OpenAI, meanwhile, chased consumer reach — GPT-4o's free multimodality, then Sora as a public video product and later a standalone app modeled on a vertical-feed social network. But the Sora app's user base reportedly peaked around one million and then halved, and OpenAI shut it down this spring, ending a billion-dollar Disney deal in the process along with an advertising pilot that started small and lost its minimum-budget floor by spring. The freed-up compute went into coding and enterprise tools instead, with Codex arriving four months behind Claude Code and ChatGPT Work only landing this summer. Board chair Bret Taylor has admitted the company was playing catch-up in coding, and CFO Sarah Friar now names just three growth drivers — the GPT-5.6 models, ChatGPT Work and Codex — with nothing consumer-facing among them. None of this erases OpenAI's own progress; doubling its run rate since the end of last year is not nothing. But the comparison looks lopsided once costs enter the picture: OpenAI's actual 2025 revenue was $13.07 billion against a net loss of $38.5 billion, and Anthropic's losses, while not disclosed in comparable audited form, are presumably billions too. A new hierarchy is forming regardless — Anthropic's potential IPO valuation of $2 trillion would eclipse even SpaceX, while whether OpenAI goes public in 2026 at all remains an open question. "SaaSpocalypse." Between the Cowork launch and February's low point, the S&P Software & Services Index lost a quarter of its value, and Claude Design's release cost Figma another seven percent in a single day. The fear underneath all of it is simple: once an agent does the job a subscription used to gate, the money follows the outcome, not the tool. ents value 20 through developers first was patient and expensive; it's paying off now."}analysis wait I need to produce valid JSON only, not this garbled mess. Let me redo properly.this is broken, restart cleanly.<br/>Also opinion field must be separate. I clearly mangled formatting badly. Let me write clean final answer.},
Read more about this at: Trending Topics