Is the AI Conveyor Belt of Capital About to Stop? (gizmodo.com)

A cascade of headline-grabbing, interlocking deals among AI heavyweights — Nvidia’s pledged investment in OpenAI (announced as up to $100B), OpenAI’s $300B computing commitment to Oracle, and Oracle’s $40B chip purchases from Nvidia (plus similar arrangements with AMD) — has sent valuations and stock prices soaring. Observers worry these agreements form a self-reinforcing “conveyor belt” of capital that largely shuffles money among a few firms rather than reflecting real enterprise adoption. The scale is eye-popping: analysts count more than $1 trillion in compute commitments around OpenAI this year, OpenAI projects $129B revenue by 2029 despite roughly $10B in near‑term revenue today, and Morgan Stanley estimates AI drove ~40% of US GDP growth in 2025 and ~80% of U.S. stock growth. Technically and economically, the deals are conditional and strain real-world supply: Oracle would need ~4.5 GW of additional power to meet commitments, major infrastructure spend by cloud and chipmakers has totaled ~$560B while generating only ~$35B in AI revenue so far, and surveys suggest many enterprise pilots yield no ROI. If these investments enable step‑function model improvements and widespread adoption, they could pay off; if not, the highly interdependent, metric‑contingent contracts risk a rapid unwind akin to housing‑market cascades, stressing credit markets, equity valuations, and the broader economy.