OpenAI's 2025 financials LEAKED

TL;DR

• 1

  OpenAI's leaked 2025 loss is staggering: Mo highlights Ed Zitron's reported numbers showing $13.07 billion in revenue, $7.5 billion in cost of revenue, and a net loss of $38.5 billion, after already losing about $5.09 billion in 2024.

• 2

  Anthropic's weird behavior is starting to look more coherent: Citing Ben Thompson, he says more people now buy that Anthropic genuinely believes it should be the steward of safe superintelligence, rather than just using safety as PR cover.

• 3

  Fable changed how some serious observers talk about model quality: Thompson's take that Fable made GPT-5.5 and Opus 4.8 feel "small and dumb" echoes Mo's own reaction and supports the idea that Anthropic may have seen something unusually capable.

• 4

  The real business fight may be over training data, not subscriptions: He points to heavily subsidized $200 plans, massive token allowances, and Anthropic's new 30-day enterprise retention policy as evidence that usage data is strategically valuable.

• 5

  AI comedy is drying up because the systems are less blatantly deranged: Mo says the GPT-4o era produced endless "AI psychosis" and goofy failures, but newer models are tamer, so the conversation is shifting from slapstick to serious tradeoffs.

• 6

  Agentic coding still breaks on the last 20%: Using Mitchell Hashimoto's comments, he argues that AI can build impressive demos and decent unit tests, but still struggles with underspecified product decisions, performance tradeoffs, compatibility, and coherent human-facing design.