TLDR: We might never get another ARR figure from Anthropic again, and that’s a good thing. This should push the market to shift focus from obsessing over Anthropic/OpenAI’s ARR numbers to hyperscale growth rates, which are currently a better indicator for AI since they capture Muse, Grok, Gemini, and open source. Anthropic achieved positive FCF in Q3 and re-accelerated in October due to Opus 5.5, according to third-party data. A roadshow a few weeks from now will broadly confirm this. With 75% of compute resources allocated to training and positive FCF, their unit economics are far better than expected. All based on public reports. According to Sarah Friar’s interview, OpenAI continues to grow at the same pace, taking share from Anthropic in Q3. This means the combined growth of Anthropic/OpenAI is accelerating from Q3 levels. OpenAI might achieve positive FCF in the coming months. Open source is growing much faster than Anthropic or OpenAI. Modal has tripled since May and, per yesterday’s announcement, is up 15x YoY. Open source tokens are generating more revenue and profit for hyperscale cloud/new cloud than frontier tokens. *Every* open-source inference cloud I know of is growing faster QoQ and YoY than Anthropic or OpenAI. And their combined compute spend is becoming increasingly significant. Third-party data shows hyperscale cloud continued to accelerate in Q3. All reports will be released in the next two weeks. The acceleration should persist into Q4 and Q1 as the installed compute base slowly reprices closer to spot rates, and new capacity comes online at higher prices. The continued acceleration of hyperscale cloud should help ease concerns about external financing. In the past 24 hours, 10-year yields have dropped sharply. There’s a risk of escalation in Iran in the coming days, but earnings reports and midterms may overshadow factors that require a super long-term perspective over the next 30 days. In my view, the biggest risk remains cyberattacks and OpenAI/Anthropic dominating the model layer. Fragmentation, competition, and lower profits at the model layer are beneficial for other players in the ecosystem and for the U.S. Slow, steady growth is far better than overbuilding and a true bubble.