Overview
- Hyperscalers are on pace for roughly $400–$500 billion of AI infrastructure spending in 2025, underpinned by multi‑year GPU and data‑center deals that include OpenAI’s arrangements with Nvidia, Oracle and AMD.
- Economists and investors are split on crowding‑out risks, with Ed Yardeni saying cash‑rich cloud firms can largely self‑fund buildouts while others warn heavy AI investment could divert capital from non‑tech sectors.
- Visible Alpha estimates point to capex growth slowing to about 19% in 2026 from roughly 54% this year, putting a premium on Q3 calls next week from Meta, Amazon, Microsoft, Google and Apple for updated spending plans.
- Financing structures are getting more complex, as Oracle and CoreWeave add debt and The Information reports Nvidia has discussed guaranteeing loans for OpenAI to build its own data centers.
- Bubble warnings draw parallels to the late‑1990s overbuild, with CityAM estimating today’s pace would require $2–$3 trillion in annual revenues by 2030 and cautioning that a pullback could weigh on GDP and equities.