On Monday, November 10, our Lab’s seminar series welcomed Andrey Fradkin, Principal Economist at Amazon, and his talk “The Emerging Market for Intelligence: The Supply, Demand, and Usage of LLMs.”

We document new facts about the supply and demand of LLMs as measured by API usage on Microsoft’s Azure platform and on OpenRouter. We first document the rapid proliferation in the number of LLMs available, and the entry of inference providers such as DeepInfra, Fireworks, Cerebras, and Groq for open-source models.

Second, we document trends in the pricing of LLMs, showing that usage-weighted prices remain relatively flat even as the price per unit of intelligence has fallen. We also document differences in pricing between open and closed-source models. Third, we consider the adoption of LLMs. There is substantial heterogeneity in the speed of adoption across industries, firm sizes, and models. We also document the degree of multi-homing by firms and apps across models.

Finally, we study the response of demand to the entry of new and better models into the industry.

Professor Fradkin is an economist who studies digitization and search and matching markets. He’s written papers on topics such as the design of Airbnb’s search and matching algorithm, reputation systems, online job search, and 401(k) contribution choices by workers. Before joining Boston University, he was a Postdoctoral Fellow at the Initiative on the Digital Economy at MIT. His research has been published in both economics journals (American Economic Review, The Review of Economics and Statistics) and computer science conferences (ACM-EC). He’s provided expert input about the digital economy at the President’s Council on Science and Technology and the Federal Trade Commission. He worked as a data scientist at Airbnb while completing a Ph.D. in Economics at Stanford University.