The contradiction at the heart of the trillion-dollar AI race (www.bbc.com)

The story highlights a central contradiction in the current AI boom: hyper‑enthusiastic, trillion‑dollar investment in compute and models versus rising warnings that valuations and spending may be a bubble. Google is pouring billions into custom silicon (TPUs) it views as a strategic edge, while Nvidia, Apple, Microsoft, Meta and OpenAI have driven a concentration of market value in a handful of firms. OpenAI’s ambitious multi‑year hardware commitments and rapid user growth exemplify the race to secure chips, data and data‑centre capacity—even as regulators, investors and central banks flag stretched valuations and recent share‑price dips among AI infrastructure suppliers. Technically, the article underlines why hardware matters: TPUs are application‑specific integrated circuits (ASICs) tailored for ML workloads, differing from general‑purpose CPUs and parallel GPUs, and require massive cooling and energy to crunch trillions of operations. That creates a capital‑intensive “AI factories” dynamic where deep‑pocketed incumbents can self‑fund scale while smaller players rely on borrowed money or scarce GPU supply. Key implications include systemic market concentration risk, enormous future power demand for data centres (comparable to a large nation by 2030), pressure for bespoke chip ecosystems, and policy questions about whether governments should build or back public AI infrastructure as the industry chases AGI.