How much of the AI boom is underpinned by Nvidia's balance sheet? Investors ask (fortune.com)

Nvidia’s blockbuster pledge — including a recent $100 billion investment in OpenAI on top of prior rounds and a web of stakes in cloud providers like CoreWeave (≈7% worth ~$3B) — has heightened investor concern that much of the AI boom is being underwritten by circular financing. Nvidia has both equity and buyback/lease deals (e.g., CoreWeave bought ~250,000 mostly H100 GPUs at ~$30k each, ~ $7.5B) and multi‑billion-dollar capacity commitments, plus agreements to rent chips from neo‑clouds (Lambda deals ~$1.3B + $200M). These arrangements lower borrowers’ funding costs (like a co‑signer), let customers avoid depreciation charges by leasing rather than buying, and often funnel cash back to Nvidia, making it hard to tell how much “real” external demand exists. NewStreet estimates every $10B Nvidia invests in OpenAI could translate into ~$35B of GPU purchases or leases — a potentially large, concentrated feedback loop. The significance is twofold: technically, Nvidia is assuming depreciation, inventory and credit risk as it levers GPU leasing and resale markets; financially, these circular flows can inflate perceived demand and echo past tech bubbles (telecom equipment roundtripping) if demand softens. Because many deals aren’t individually material for disclosure and Nvidia is systemically large and “priced for perfection,” even modest missteps—falling lease utilization, an oversupply of used GPUs, or loan losses—could disproportionately hit its valuation and broader markets. Investors should monitor lease exposure, resale markets for used accelerators, and the accounting treatment of these arrangements.