Eleven point five billion dollars in ninety days. When preliminary Q2 documents surfaced this week, the figure felt almost fictional. Fourteen times the same quarter last year. Nearly two and a half times Q1’s already massive $4.73 billion. For a brief moment, the AI sector had a new revenue king. Then the footnotes showed up.
The documents also noted something called positive adjusted operating income for the first time. Adjusted. That word is doing an enormous amount of heavy lifting here, and if you’ve spent any time watching private tech companies sprint toward an IPO, you know exactly what it means. The headline is true in a specific, curated sense, and the GAAP reality is still buried under an avalanche of compute bills and partner pass-through costs.
The Fine Print in a $11.5B Quarter
Here’s what the celebratory tone misses. Anthropic’s revenue figure includes sales routed through cloud partners, which the company reportedly counts in its top line while OpenAI does not. That isn’t illegal or even unusual, but it turns the run-rate leaderboard into a comparison of two different sports. When I started digging into the May claims, a $47 billion annualized run rate looked less like a clean victory over OpenAI’s reported $40 billion and more like a category error.
And then there’s the compute. The growth driver everyone keeps circling back to is Claude Code and the broader shift toward agentic, enterprise-grade coding workloads. They’re high-value contracts that chew through infrastructure. The per-developer revenue looks beautiful on a slide deck, but the per-developer GPU burn is something else entirely.
I’ve been watching threads asking whether demand can actually outpace infrastructure spend, and the honest answer is that nobody knows yet. Not even Morgan Stanley or Goldman Sachs, who are currently packaging this story for a fall listing.
The hardware underneath these workloads isn’t getting cheaper for enterprise buyers or local builders hunting for available compute at retail. Which raises the question of whether margin compression is already baked into these contracts, even if it hasn’t shown up in the adjusted operating income line.
Dario Amodei’s recent regulatory commentary adds another layer that feels underreported. His push for frontier-model rules with revenue thresholds that conveniently exempt smaller players while constraining labs like his own is being read by some as safety advocacy, and by others as classic pre-IPO moat construction. When a company is about to test public-market patience, regulatory barriers that lock out future competitors become a lot more valuable than another research paper.
Chasing a $2 Trillion Mirage
The same investor documents that leaked the Q2 revenue also fed the $2 trillion IPO valuation rumor mill. Let that number sit for a second. Two trillion dollars. For a services business that rents intelligence by the token.
Over on Reddit and X, the pushback has been immediate and mercifully blunt. Hardware companies with fabs and fiber own trillion-dollar tags. Software margins at scale can get there, but this isn’t Microsoft in 1995. It’s a company whose quarterly numbers are still preliminary, explicitly subject to revision, and less than a year removed from a $965 billion private valuation.
Private companies approaching public listings often paint in broad strokes. We watched similar dynamics earlier this year when SpaceX pushed its own Q2 narrative around AI and Starlink revenue, and the pattern is familiar. Massive private valuations need public-market patience to survive.
What strikes me as genuinely risky is the speed at which forecasts are becoming obsolete. The May 2026 run-rate figure of $47 billion was already being treated as gospel in investor decks printed last month. Now Q2 alone suggests a higher annualized number, and every model built on the old assumptions looks instantly stale. Public markets won’t forgive that kind of whiplash the way private Series H investors might.
There’s also the export control friction. Top Anthropic models are already restricted in certain regions, which means the addressable market for the most expensive tiers is narrower than the global growth narrative implies. When you’re pricing in trillion-dollar scale, every blocked geography matters.
None of this means Anthropic isn’t a remarkable company. The revenue acceleration is real, the enterprise adoption is broadening, and the technology is clearly resonating with developers who need more than a chat interface. But if you’re expecting a clean fairy tale when this thing lists, you’re going to be disappointed. The story here is one of staggering momentum colliding with accounting ambiguity, regulatory positioning, and a market that may have already front-run its own optimism.
I’ll be watching the October window closely. Not because I think the valuation makes sense, but because I want to see how many investors pretend that it does.
