Sep 17, 2026 · 27 min
Unused GPUs expose the AI boom’s biggest assumption
AI Bubble: ‘This could humiliate the largest companies in the world’ | Ed Zitron
If companies bought vast computing capacity before demand materialized, the resulting inventory glut could force write-downs and reset AI valuations.
- 1Microsoft’s reported gap between expected and operating AI capacity raises questions about how much infrastructure is actually in use.
- 2Premature GPU purchases could leave companies holding obsolete hardware while capital spending outruns real-world demand.
- 3A future GPU glut would challenge valuations built on promised growth rather than deployed capacity and productive usage.
Don't miss
Ed Zitron frames unused AI hardware as tangible evidence that the industry may be valuing future promises over productive demand.
The brief
Ed Zitron and Isaac Pound begin with reports that Microsoft has brought roughly two gigawatts of AI compute online against expectations of twelve, exposing a striking capacity gap.
The discussion shifts from data centers to warehouses, where purchased GPUs could sit unused, lose value as newer generations arrive, and turn enormous capital spending into inventory risk.
The hosts question whether NVIDIA’s sales reflect deployed demand or pre-orders made years before facilities are ready, leaving the industry dependent on growth that has not arrived.
The standout tension is tangible hardware versus abstract promise: unused GPUs offer a physical test of whether AI valuations rest on productive demand or future expectations.
They close on a less risky procurement model—buying GPUs shortly before data centers finish—while warning that immediate purchasing may have spread the inventory problem.