Aug 28, 2026 · 55 min
AI’s next bottleneck is the infrastructure beneath the models
The Infrastructure Behind the Machine Age
The episode frames AI’s expansion as a physical buildout constrained by chips, memory, power, cooling, networks, and data centers.
- 1AI demand is outrunning infrastructure supply as usage, reasoning, agents, and embodied applications multiply compute needs.
- 2The opportunity spans specialized chips, memory, networking, power, cooling, and data centers rather than software alone.
- 3New infrastructure companies need deep technical and manufacturing expertise because hardware requires large commitments and long production cycles.
Don't miss
The guests translate gigawatt-scale AI power demand into the physical realities of towns, homes, grids, permitting, construction, and fuel.
The brief
Ben Horowitz, Martin Casado, and Raghu Raghuram introduce a16z’s Machine Age Fund, arguing that AI’s decisive bottlenecks increasingly sit beneath the models.
Surging token use, hyperscaler spending, chip and memory prices, and sold-out supply point to a shortage that spans computing, power, cooling, and data centers.
Reasoning, agents, digital workers, and robotics could multiply inference demand, extending AI from chat and coding into broad organizational and physical work.
The infrastructure stack must be redesigned around inference, memory movement, networking, and system efficiency, creating room for specialized hardware companies.
The fund’s central bet is that expanding markets will fragment into categories measured by tokens per dollar, watt, or rack, rewarding technically experienced founders.
Ben Horowitz
a16zBooks & mentions
Listen to the full episode and explore every guest, topic, and moment on PodLume.