Stanford MS&E435 Economics of the AI Supercycle | Spring 2026 | Economics of Generative AI

The AI sector currently follows an inverted value structure where the majority of capital flows into foundational hardware and infrastructure, while application software remains in an early stage, costly to run, and less profitable than historical software models.

• Infrastructure spending — capital flows are heavily concentrated in chips and data centers, creating a pyramid where value resides at the bottom rather than the user-facing top [27:03].

• Profitability gap — unlike traditional software where adding a new user costs almost nothing, AI applications require expensive computing power for every interaction, resulting in lower margins [08:51].

• Compute utilization — roughly 60% of high-end processor usage is currently dedicated to training new models, while 40% is used for running them (inference), though the inference share is expected to rise [18:55].

• Market maturity — history suggests a decade-long lag between initial infrastructure construction and full value realization, indicating the current AI sector remains in its early phase [10:15].

• Usage volume — leading AI tools currently fall into a middle ground of adoption, sitting between niche applications and mandatory communication utilities [31:32].

• Future revenue — advertising models may supplement or replace current subscription plans because AI interactions provide high-quality data regarding user intent [33:03].

• How does the predictability of training workloads differ from inference workloads?

• What criteria are used to determine if a new business is a standalone platform or a feature of existing infrastructure?