The Infrastructure Behind the Machine Age
Summary
a16z partners Ben Horowitz, Ragul Ram, and Martin Casado announce the Machine Age Fund, a new investment vehicle dedicated to the physical infrastructure powering the next era of AI. They argue that the primary bottleneck has shifted from model development to the underlying hardware, including chips, memory, networking, power, and cooling, which are currently sold out through 2028. The episode explores how AI’s insatiable demand for compute is forcing a redesign of data centers and supply chains, creating massive opportunities for new founders to solve these resource limitations.
Topics discussed
Introduction to the Machine Age Fund and AI infrastructure
Why infrastructure is the new bottleneck for AI
Shift in founder focus toward hardware and physical limits
Hyperscaler CapEx and exploding demand signals
Supply constraints and historical comparisons to internet
Anecdotes on server costs and the impossibility of scaling fast
Why this investment category emerged only now
Token demand growth and physics limits of current tech
Inference scaling and the resource limitation dynamic
The Mythical Man Month vs. AI scaling dynamics
User adoption metrics and expanding demand drivers
GrokBot, computer use, and the early stage of AI journey
AI as a digital employee and organizational integration
Designing hardware and architecture specifically for AI
Power delivery challenges: AC to DC transition
Cooling, water usage, and physical data center constraints
Labor shortages and defining the fund's scope
Power grid gaps and the need for self-sufficient energy
Naming the fund: Machine Age vs. Artificial Intelligence
Innovation drivers and the role of startups vs. incumbents
Investment thesis: Hardware margins and specific categories
Founding teams: Experience, age, and ecosystem thinking
Talent pipeline, US infrastructure advantage, and closing
Listen ad-free on Castria