Synthesis so far
Updated Nov 19, 2026. What the data says so far; the articles below are the full record, newest first.
What we know, in five lines.
- The five biggest AI builders earned $607B in their last four quarters and kept $33B as free cash: 5 cents per dollar, down from 53.
- They borrowed the gap: +$355B of debt in a year; Alphabet's quadrupled. Can Google borrow another $100 billion?
- The labs owe the bills: Anthropic has $451.6B of computing due before 2031. Priced for 2031 or 2036?
- Nvidia's customers now owe it $81B, the number that rose before its worst falls. Why?
- The market noticed on Oct 6: Nvidia -36%, Nasdaq 100 -13%, S&P 500 -6%. Could it fall 50%?
Why. Demand for AI computing can't be stopped, but it can arrive later than the bills. In 2000, telecom builders spent more than their cash; in 2002 they cut, and their suppliers fell 89-99%.
So what. The builders carry the cash risk; the suppliers carry the pause risk (in the 2023 pause, Meta's stock rose 170%). Own both sides on purpose through the index; size any one lab or chip stock so a 30% fall is survivable.
What would change it. Builders' free cash back above 50 cents per dollar of profit; Nvidia's collection days falling; Anthropic's operating cash growing with its sales.
Open questions. Why did Anthropic win? · Do early leaders stay on top? · Where will AI's power come from? · Is AI's power bill your inflation?
About this storyline
Companies are spending hundreds of billions of dollars a year on AI chips, data centers and the power to run them. Our analyses follow the money: who is spending, who is earning from it, and how much of the payoff today's stock prices already assume.
- The spenders: cloud and platform companies, their capital spending and the returns on it.
- The suppliers: chips, memory, networking, power and data-center equipment.
- The valuations: what growth the market is pricing in, and what happens if it slows.