Anthropic Ascends While Doubts Linger

written

The center of gravity in AI tilted toward Anthropic this week, and the numbers, the drama and the doubts all arrived together. Internal figures reported by Bloomberg showed Anthropic's second-quarter revenue overtaking OpenAI's for the first time — the Claude maker posting more than $11 billion against OpenAI's roughly $6.7 billion, and booking its first positive operating income while OpenAI's losses widened to about $12.3 billion. By late July its annualized revenue run-rate had reached roughly $65 billion, up from about $47 billion in May. At the same time the company is preparing a public offering that, per Bloomberg, it expects to match or beat SpaceX's record June listing, which raised roughly $86 billion at a $1.77 trillion valuation; co-founder Dario Amodei is reportedly in line for super-voting shares, and the filing could come within weeks.

And yet the same week delivered a countertext worth pinning to the wall. The Financial Times reported that Anthropic's most powerful model is struggling to hold users, with customers trading down to cheaper tools for routine work — a price-performance signal that matters more to a buyer than any revenue headline. So the ascendant lab is simultaneously the most commercially dominant and the most exposed to the "cheaper is good enough" dynamic reshaping every AI budget.

The founders themselves are cooling on the index even as their company climbs, and the reason is legitimacy, not performance. Dario and Daniela Amodei both registered as falling amid IPO scrutiny and a wave of culture and governance stories. Dario spent the week pushing back on the claim that his own risk warnings fueled public backlash; he argued the problem runs deeper — a "crisis of trust" rather than a messaging failure — and insisted his writing has been roughly balanced between AI's risks and its benefits, pointing to his optimistic essay "Machines of Loving Grace."

That trust theme is what's actually pulling the field. Fei-Fei Li rose on the back of a Bloomberg interview on "The Circuit" in which she warned that growing U.S. opposition to AI — much of it aimed at the water and energy demands of data centers — could carry global consequences. "If we are not showing a positive attitude and positive path toward AI, everybody loses," she told Emily Chang, arguing the U.S. should be a "role model" for AI's economic value and that technologists like her have done a poor job communicating with the public. Her World Labs, focused on world models, raised $1 billion in February.

Geoffrey Hinton also rose, but from the opposite pole. The Nobel laureate spent the week warning that big tech is effectively betting on AI to do "the work of workers much cheaper," and that mass unemployment is a real possibility as roughly a trillion dollars flows into data centers and chips. Where Li worries the backlash is overheated, Hinton is arguing the labor disruption is underpriced — the same anxiety, read two ways.

The other mover story is OpenAI's quiet reshuffle. Fidji Simo, Sam Altman's No. 2 as head of applications, stepped down from her full-time role after a three-month medical leave following a severe flare of a chronic illness she's lived with for seven years, transitioning to a part-time advisory seat; she is now focused on ChronicleBio, her startup collecting biological data — reportedly some 153 terabytes from thousands of blood samples — to help AI models understand chronic disease, with the caveat that smarter models alone aren't enough. Her exit fed a broader narrative of departures that has left Greg Brockman consolidating control over OpenAI's product and scaling teams ahead of its own IPO push. And Andrej Karpathy continued his climb, still the field's most-watched teacher, now inside Anthropic's pre-training effort and shaping this week's conversation about moving from "vibe coding" to disciplined agentic engineering.

The throughline: this was the week AI's commercial peak collided head-on with its legitimacy problem. Revenue records, a blockbuster IPO and a genuine trust crisis are now the same story — and for CTOs, the tell is buyers quietly choosing cheaper models while the market debates whether to trust the expensive ones at all.

read-aloud

The biggest story in AI this week is Anthropic — and what makes it interesting is that the triumph and the trouble showed up at the exact same time.

Start with the triumph. Internal figures reported by Bloomberg show that, for the first time, Anthropic's quarterly revenue passed OpenAI's. In the second quarter, the maker of Claude pulled in more than eleven billion dollars, against roughly six-point-seven billion for OpenAI. And here's the kicker: Anthropic booked its first positive operating income, while OpenAI's losses widened to around twelve-point-three billion dollars. By late July, Anthropic's annualized revenue run-rate had climbed to about sixty-five billion, up from roughly forty-seven billion in May. On top of that, the company is preparing an IPO it expects to match or beat SpaceX's record listing from June — the one that raised about eighty-six billion dollars at a one-point-seven-seven trillion dollar valuation. Co-founder Dario Amodei is reportedly set to get super-voting shares, and the paperwork could land within weeks.

Now the trouble. That same week, the Financial Times reported that Anthropic's most powerful model is actually struggling to keep users — because customers are trading down to cheaper tools for everyday work. So think about that. The company posting record revenue is also the company most exposed to the "cheaper is good enough" problem. If you run a technology budget, that's the detail to remember, because it's the tension you're living every day: do you pay up for the best model, or does the cheaper one clear the bar?

Here's the twist. Even as Anthropic climbs, its founders — Dario and Daniela Amodei — are actually cooling on our index. And it's not about performance. It's about trust and scrutiny, sharpened by the IPO and by a run of stories about the company's culture and governance. Dario spent the week pushing back on the idea that his own doom-y warnings helped turn the public against AI. He said the real issue runs deeper — he called it a "crisis of trust," not a messaging problem — and argued his writing has been roughly balanced between the risks and the benefits, pointing to his hopeful essay "Machines of Loving Grace."

And that trust question is really what's pulling the whole field this week. Take Fei-Fei Li, who's rising. In a Bloomberg interview, she warned that the growing backlash against AI in the United States — a lot of it aimed at the water and energy that data centers guzzle — could have global consequences. Her line was blunt: if the U.S. doesn't show a positive path forward on AI, in her words, "everybody loses." She thinks technologists have done a bad job talking to the public, and that America should be a role model for AI's economic upside. Her startup, World Labs, raised a billion dollars back in February.

Then there's Geoffrey Hinton, also rising — but coming at it from the opposite direction. The Nobel winner spent the week warning that big tech is basically betting on AI to do the work of employees much more cheaply, and that mass unemployment is a real risk as something like a trillion dollars pours into chips and data centers. So notice the split: Li worries the backlash is overheated; Hinton worries the job losses are underestimated. Same anxiety, two very different readings.

Over at OpenAI, there's a quieter but important shakeup. Fidji Simo — Sam Altman's number two, who ran applications — stepped down from her full-time job after a three-month medical leave, following a severe flare of a chronic illness she's lived with for seven years. She's moving to a part-time advisory role, and she's now focused on her startup, ChronicleBio, which is collecting biological data — reportedly around a hundred and fifty terabytes from thousands of blood samples — to help AI understand chronic disease. Her point, notably: smarter models alone won't cure anything. Her exit is part of a bigger wave of departures that's left Greg Brockman consolidating control over OpenAI's product and scaling teams, right as that company gears up for its own IPO. And Andrej Karpathy kept rising too — still the field's favorite teacher, now inside Anthropic's pre-training work, and driving this week's conversation about graduating from "vibe coding" to real, disciplined agentic engineering.

So here's the throughline. This was the week AI's commercial peak ran straight into its legitimacy problem. Record revenue, a giant IPO, and a genuine crisis of trust are now the same story. And if you're making buying decisions, the quiet tell underneath all of it is this: customers are already reaching for the cheaper models — while everyone else argues about whether to trust the expensive ones at all.