Gavin Baker: Why AI Demand Is Outrunning Compute Supply
Summary
Gavin Baker and David George analyze the AI boom, arguing that current supply constraints and sub-one-year paybacks suggest a positive-sum market where frontier labs, open source, and chipmakers all thrive. They discuss the potential for orbital compute via SpaceX to solve terrestrial bottlenecks and emphasize that AI is reindustrializing America by revitalizing small towns. The episode concludes by urging the industry to highlight tangible benefits for working-class Americans to counter bearish narratives and political resistance.
Topics discussed
The Age of Elon and Jensen: AI supply constraints and bubbles
AI acceleration trends and the 'how does it go wrong' question
Frontier model competition and public market dynamics
Training vs. inference economics and revenue volatility
Nvidia's central role and the training investment cycle
Data center payback periods and low-cost financing
Demand-side diffusion and engineer spending disparities
Personal AI usage: coding, summarization, and productivity
Historical tech bubbles and current funding structures
Physical constraints, rising rates, and AI regulation
Economic benefits of data centers for small towns
Reindustrialization and the US cost advantage in energy
The risk of massive compute undersupply and price hikes
Open source models, licensing, and the future of space internet
Orbital compute: engineering feasibility and economics
SpaceX's AI infrastructure strategy and launch economics
Futuristic SpaceX applications: asteroid mining and orbital bases
Microsoft's strategy and the rise of open source models
The 'abstraction layer' for enterprise AI and intelligence arbitrage
Competing for the AI abstraction layer: who will win?
Nvidia's ecosystem dominance and the 'Bank of AI'
Jensen Huang's supply chain control and semiconductor investing
Custom chips, RVG deals, and navigating supply constraints
Closing remarks and disclosures
Listen ad-free on Castria