Nov 11, 2026 · 57 min
AI infrastructure may outlast the companies building it
What the Internet Boom Teaches Us About the AI Buildout | Mike Volpi
The episode tests whether AI’s infrastructure boom will repeat the fiber bust or create durable value through stronger utility and integrated systems.
- 1AI applications may generate revenue and practical value faster than many internet applications did during the 1990s boom.
- 2GPU scarcity, improving software, and demand can extend older chips’ useful lives, but abundant supply could reverse the economics.
- 3AI data centers increasingly require co-designed chips, memory, networking, energy, and software rather than standalone components.
Don't miss
Volpi argues that AI data centers must be designed as integrated systems, with the rack or facility treated as a single architecture.
The brief
Mike Volpi compares today’s AI buildout with the late-1990s fiber boom, finding familiar capital excess and infrastructure constraints—but stronger practical utility in AI applications.
The dot-com correction offers a warning: indebted builders can fail while their networks remain valuable, a pattern Volpi says could recur in AI infrastructure.
GPU scarcity and improving software have extended older chips’ useful lives, yet a supply glut could quickly weaken financing assumptions and asset values.
Volpi applies Cisco’s acquisition playbook to NVIDIA, weighing whether its financing and deals seek returns, industry control, or both as consolidation approaches.
The standout argument is architectural: AI data centers behave as integrated systems, making chips, memory, optics, networking, energy, and software inseparable.
Volpi’s broader conclusion is that this cycle demands cross-layer thinking—and may be more economically significant than the 1999 internet boom.
Weekly email
State of the Pod, every Monday
The week's most-discussed names and stories from podcasts, with the moments behind them. Free, five minutes, unsubscribe in one click.