The Hater's Guide to the AI Bubble Vol. 2 (www.wheresyoured.at)

The author argues we’re deep in an overheated AI bubble driven by wildly inconsistent financials and optimistic projections. Citing internal documents, they say OpenAI spent roughly $12.4B on inference from 2024 to Sept 2025 and at least $2.469B in revenue in 2024 (based on Microsoft revenue-sharing), yet reported public figures don’t reconcile: one report put H1 2025 revenue at $4.3B with $2.5B cost of revenue while the author’s data shows $5.022B spent on inference and at least $2.2735B earned. That gap calls into question CEO claims such as “profitable on inference” and targets like $20B annualized revenue for 2025. Microsoft CEO Satya Nadella’s quip that labs sometimes inflate numbers to raise money only deepens skepticism. Anthropic’s rosy forecasts — break-even in 2027–2028, gross margins hitting 63–75% by 2026–28 — also look shaky. Reported margins have swung from ~50% to 38% to a disclosed -109% for 2024 (or -94% for paying customers), and claims that multi-vendor chip allocation (NVIDIA, Google, AWS) yields dramatic efficiency gains clash with reporting that AWS Inferentia 2 underperformed NVIDIA H100/A100 in latency and cost for many workloads. The piece’s takeaway: media-amplified vibes, opaque accounting around inference costs, and unverified margin claims are inflating investor expectations; the author will publish a company-by-company reality check to separate durable businesses from hype.