Sep 25, 2026 · 48 min
AI’s infrastructure boom leaves billions in GPUs idle
Hyperscalers are hiding a massive secret about uninstalled GPUs | Ed Zitron
The episode examines whether debt, stalled construction, and collapsing AI prices are exposing structural weaknesses beneath the industry’s expansion.
- 1Data-center delays and rising financing costs are leaving vast GPU inventories uninstalled or unused.
- 2Hyperscalers keep buying chips despite idle capacity, driven by supply pressure and flawed industrial planning.
- 3Falling AI prices and unreliable agents challenge the assumption that more compute will produce durable demand.
Don't miss
Ed Zitron’s estimate that $200–$300 billion in Nvidia GPUs may be sitting unused turns the infrastructure boom into an inventory-risk story.
The brief
Ed Zitron and Isaac Pound examine the weak points beneath AI’s infrastructure boom, from stalled data centers and mounting debt to hardware that may never be powered on.
Oracle’s force majeure notice over a New Mexico facility becomes a case study in how ambitious projects can buckle under delays, contractual pressure, and rising interest costs.
Zitron estimates that $200–$300 billion in Nvidia GPUs may be sitting in warehouses or unpowered data centers, creating a potentially enormous inventory problem.
The hosts ask why hyperscalers keep reserving chips while capacity sits idle, weighing supply pressure, poor planning, and the risk that hardware becomes stranded or obsolete.
OpenAI and Anthropic face huge compute commitments as AI prices fall, while Meta’s cheap tokens intensify a race in which companies cut prices and accept losses.
The closing skepticism extends to AI agents: applications marketed as everyday assistants may be unreliable, unnecessary, and unable to justify the infrastructure built for them.