Stanford MS&E435 Economics of the AI Supercycle | Spring 2026 | Infrastructure, Capstone Case

The AI Supercycle is fundamentally driven by the scaling of compute capacity, which serves as a primary indicator for revenue growth in frontier labs. Modern agentic workloads, characterized by reasoning and iterative tool use, have shifted compute demand toward complex, directed acyclic graphs rather than simple, one-off inference calls. Consequently, inference now constitutes over 80% of total compute usage, necessitating a heterogeneous infrastructure that balances specialized accelerators with traditional CPUs. The primary bottleneck for this growth remains the physical supply chain, including power generation, cooling, and wafer allocation. To sustain progress, the industry must transition toward more resilient infrastructure and leverage AI-driven recursion to accelerate the hardware design cycle. Ultimately, the shift from application-based interfaces to outcome-oriented AI agents, as described by OpenAI industrial compute lead Sachin Katti, will redefine both the underlying compute requirements and the value distribution across the technology stack.