Holy sh*t I think Anthropic is profitable now

TL;DR

• 1

  Anthropic’s numbers suddenly look IPO-serious — Theo highlights reports that Anthropic could more than double revenue to about $10.9 billion in Q2 and post its first operating profit, a huge shift for a lab category that has mostly been losing billions per quarter.

• 2

  AWS may be the real distribution engine behind Claude — Because Anthropic models are available across AWS, GCP, and effectively Azure routing, Theo argues enterprise buyers had a practical default while OpenAI stayed trapped on Azure, giving Anthropic the best channel into Fortune 500 AI spend.

• 3

  Anthropic built a profit machine by taking cloud revenue share without supplying the compute — His key point is that Anthropic can hand weights to AWS and Google, let them host inference, and still collect roughly half the token revenue while preserving its own GPU capacity for research.

• 4

  Opus got cheaper on paper but more expensive in practice — Theo says Opus 4.5 effectively replaced Sonnet for many real workloads, moving users from roughly $15 per million output tokens to $25, while tokenizer changes in Claude 4.7 may have increased token counts by 30-50% or more.

• 5

  The real unlock was Claude Code and Opus 4.5 making coding agents feel 10x better — He describes Claude Code as a dramatic leap over Copilot-era tools, good enough that individual engineers bring it into work and trigger much larger enterprise usage and spend.

• 6

  This looks a lot like product-market fit, not just accounting luck — Citing Simon Willison and examples like Uber blowing through its annual AI budget in three months, Theo argues companies aren’t backing off after price hikes; they’re using these tools more because the value is finally obvious.