Overview
- Goldman Sachs compares today’s AI upswing to 1997–1998’s internet buildout, saying the cycle still has room to run if funding holds steady.
- Hyperscalers are set to spend about $375 billion on AI infrastructure and chips this year, rising toward $450 billion next year, concentrating gains in a small group of firms.
- Financing is shifting from internal cash and equity toward corporate bonds, off‑balance‑sheet vehicles and securitised data‑centre debt, with some borrowers tied to chips that can age out within a few years.
- Analysts detail circular relationships that inflate reported revenues and valuations, citing investor‑supplier‑customer loops across Microsoft–OpenAI, Amazon–Anthropic, Nvidia–CoreWeave and long‑dated OpenAI infrastructure contracts.
- Policy and risk voices warn of broader stability threats if AI assets reprice, pointing to opaque private‑credit links, power and supply bottlenecks, and the potential for a correction to spill over beyond equities; some investors say a shakeout could shift opportunities to markets like India.