Why concerns about an AI bubble are bigger (www.npr.org)

The story frames an intensifying debate: industry leaders — from Nvidia’s Jensen Huang to venture capitalists and bank execs — portray AI as a lasting “super‑cycle,” while economists, analysts and some investors warn the current investment surge looks dangerously speculative. Big numbers underline both sides: Nvidia’s valuation surge, OpenAI’s reported $20B revenue and its plan to spend $1.4T on data centers over eight years, and Big Tech’s roughly $400B of AI spending this year. Analysts estimate hyperscalers added ~$121B of debt in a year and project as much as $3T in AI infrastructure spending through 2028, stoking fears of overbuilding if real customer demand doesn’t materialize. What’s technically and financially worrying for the AI/ML community is how that spending is structured. Firms are levering special‑purpose vehicles to keep debt off balance sheets, using debt and private equity to finance massive data centers, and engaging in circular deals (e.g., Nvidia financing customers like OpenAI, CoreWeave stock‑for‑capacity arrangements) that may artificially inflate chip and capacity demand. If capacity outpaces genuine revenue growth, stranded GPU farms and worthless debt could trigger a market correction reminiscent of the dot‑com bust. For practitioners and startups this means possible funding volatility, a shift from product‑market focus to infrastructure buildouts, and an elevated systemic risk that could slow deployment, hiring and sustained hardware investment.