The AI economy could crash on mounting chip costs — and those token costs won’t help

AI chips are insanely expensive — a single Nvidia Blackwell GPU can cost as much as a new Tesla Model 3, and modern data centers spend billions on chip clusters.
Demand has exploded and shifted from training to inference, with Goldman Sachs forecasting a 24× rise in token use by 2030; enterprises are already overspending (Microsoft cut Claude Code licenses, Uber blew its AI tools budget).
Supply can’t keep up: costly fabs, diverted production, and chip-backed financing are driving prices, hurting startups and poorer countries, and creating possible systemic financial risks.