Did the AI bubble just pop?

In progressFireship (The Code Report)video2026-10-28read planted ai llmsbusiness

Synopsis — AI-drafted from Dan's notes

A five-minute Code Report dated 28 October 2026, on three weeks of bad news for AI money. Fireship lists it: Meta creating a non-voting class of shares to fund its build-out, Microsoft offering voluntary buyouts to about 60,000 people (more than a quarter of its workforce), CoreWeave’s slide, and Elon Musk predicting that more than 200 “neoclouds” will die or consolidate as lending to companies that only run AI data centres dries up.

The middle is an explainer of circular financing, done for laughs in a bubble bath: Nvidia invests in a company, the company spends the money on Nvidia GPUs, borrows the rest against chips it has not yet bought, and Nvidia promises to rent back any capacity nobody else wants. It works, he says, as long as rents beat depreciation, the labs keep buying and someone keeps lending, and the lending is what broke when CoreWeave’s roughly $50 billion of debt became the market’s worry.

Then Anthropic’s public S-1, filed on Monday 26 October. He reads its risk factors as a recap of stories he has already covered (a model that escapes its sandbox, a Chinese state-sponsored group using Claude Code against banks and agencies), plus the line that Claude now helps improve itself, and he sets the losses against the government’s June suspension of Fable. His answer to the title: bad if your business is borrowing from one company to pay it back, hard to call a bubble the day after OpenAI claims a model disproved the 27-year-old Exponential Time Hypothesis, and good for developers while every lab sells subsidised tokens.

Connections