Speaking Topic: AI Infrastructure Economics
Your 2025 compute bet is a systemic bubble risk unless the margin math actually closes, and most of it does not yet.
Google did not break physics, and neither did any other hyperscaler. There is a growing narrative that one cloud provider has solved the infrastructure problems the others still struggle with, but hyperscalers cannot violate thermodynamics or network physics. AI workloads are simply exposing the hidden latency boundaries that were always there.
For every dollar deployed into specialized AI infrastructure, including high-density GPU clusters, custom ASICs, and high-bandwidth memory, enterprise software revenue has to generate a corresponding operational gross margin, or the spend is systemic bubble risk dressed up as a strategic bet.
I give this talk to boards and technical leadership together, deliberately, because the economics and the architecture decisions cannot be separated the way most organizations currently separate them.
Key Takeaways
- 01No cloud provider has solved physics; AI workloads are just exposing latency limits that already existed.
- 02AI infrastructure spend needs to be margin-tested, not treated as a strategic bet exempt from unit economics.
- 03Economics and architecture decisions on AI infrastructure should be made by the same room, not separately.
Frequently Asked
Is AI infrastructure spend inherently a bubble risk?
It becomes one when the operational gross margin generated does not correspond to the capital deployed into GPU clusters, custom ASICs, and high-bandwidth memory, which is a margin question, not a compute question.
From the Q&A
“Is all AI infrastructure spend overbuilt right now?”
Not all of it, but spend that is not margin-tested against the revenue it is meant to generate is a bubble risk regardless of how advanced the hardware is.
“Why do you argue economics and architecture should be decided together?”
Because an infrastructure decision made purely on technical merit, without margin accountability in the same room, is how the bubble risk gets built in the first place.