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

The AI supercycle centers on the massive scaling of compute infrastructure required to support frontier models and agentic workflows. Revenue for frontier labs remains a lagging indicator directly correlated with compute capacity, which must triple annually to meet demand. As workloads shift from simple inference to complex, multi-step agentic tasks, the compute graph evolves into a sophisticated directed acyclic graph requiring heterogeneous hardware, including GPUs, CPUs, and specialized accelerators. Building this infrastructure at a gigawatt scale involves navigating severe constraints in power, cooling, and supply chain availability, particularly regarding wafer allocation. Future efficiency gains depend on recursive improvements where AI models assist in designing the next generation of chips and software. Ultimately, the industry must transition from current simplistic systems to more resilient, multi-layered architectures to sustain the rapid expansion of intelligence.