AI compute demand is a circular bubble
The guest argued that the massive market demand for AI compute is an artificial bubble concentrated among a few unprofitable players and hyperscalers, with very little organic enterprise demand.
The argument
When peeling away the compute demand from OpenAI, Anthropic, Meta, and the hyperscalers funding them, the natural market demand is estimated at only $1-2 billion. The guest argues that the entire industry's revenue is a circular test of how long venture capital and debt can prop up unprofitable AI companies.
The thesis, stress-tested
✓ What validates it
- ✓Hyperscalers report a slowdown in CapEx guidance in upcoming quarterly earnings
- ✓NVIDIA reports a decline in data center revenue or backlog
▸ Risks discussed
- ▸Hyperscalers continue to fund CapEx indefinitely regardless of near-term ROI
- ▸A sudden breakthrough in enterprise monetization justifies the massive compute spend
Hear it yourself
"What I can, however, say is that buried in the back of these documents is the specific itemized related cost to Microsoft and SoftBank. Now $527,000,000 of their sales and marketing costs went through Microsoft. Now do you think that Microsoft is doing a bunch of advertising for OpenAI, or is it the free credits that Microsoft is giving away via the Microsoft for startups program, much like OpenAI for startups? And on top of that, I think they're putting free pea free inference in there. But my biggest evidence, by the way, that they're doing that is that I reported last November that OpenAI spent $8,670,000,000 on inference on Microsoft Azure."
11:50
AFFILIATE LINK · ZORTIX MAY EARN A COMMISSION · NEVER A RECOMMENDATION TO TRADE