The Stanford MS&E435 course, led by Apur, explores the unique economic dynamics of the AI supercycle, focusing on the challenges of profitability across the AI ecosystem, from semiconductors to applications, and the evolving roles of major tech players. The session highlights the high costs of AI inference, the competitive landscape, and the potential for consumer AI monetization, setting the foundation for deeper analysis through guest speakers and interactive discussions.
The session begins with an introduction by Apur, the instructor for the Stanford MS&E435 course on the Economics of the AI Supercycle. He shares his background, including his career journey from India to Singapore and eventually Stanford, and his current role at Altimeter, an investment firm focused on AI. Apur outlines the course logistics, emphasizing its conversational nature, the inclusion of guest speakers from leading AI companies, and the grading system based on attendance and a final assignment. He encourages active participation and highlights the course’s goal to equip students with the knowledge to either start or fund AI companies effectively.
A central theme of the course is explored through a discussion of the AI economic ecosystem, represented as an inverted triangle, contrasting it with previous technology revolutions like the internet, mobile, and cloud. Apur explains that unlike software businesses with high gross margins, AI applications currently face high incremental costs due to expensive GPU usage for inference. This results in a different economic structure where profitability is challenging, especially at the application layer. The course will delve into the dynamics of this ecosystem, including the dominance of companies like Nvidia in semiconductors and the competitive nature of the inference layer.
The conversation includes student questions about the integration of incumbent platforms like Salesforce and Google into the AI ecosystem. Apur clarifies that these companies’ AI-related revenues are accounted for within the model and inference layers, reflecting their hybrid roles. He also discusses the cyclical nature of semiconductor investments and the timing mismatch between capital expenditure and application revenue generation. The discussion touches on the vertical integration of major players like Google, Apple, and Meta in past technology cycles and how this might inform the AI supercycle’s development.
Apur addresses the challenges of profitability and market structure in AI, noting that the semiconductor layer, dominated by Nvidia, is currently the most profitable segment. He highlights the ongoing competition in the infrastructure layer and the uncertainty about which startups will emerge as dominant platforms versus features within larger ecosystems. The course will feature speakers from various layers of the AI stack to provide insights into these dynamics. Apur also emphasizes the importance of understanding hyperscaler capex guidance as an indicator of the industry’s health and future trajectory.
The session concludes with a discussion on consumer AI applications like ChatGPT and Google’s Gemini, comparing their user bases and monetization models to established consumer tech products. Apur notes that while ChatGPT has reached about a billion users, its monetization per user is significantly lower than giants like Alphabet and Meta. He raises questions about how AI applications can expand their user base and increase revenue, suggesting that advertising models leveraging better user intent and attribution may be key. The lecture ends with a quiz to engage students and reinforce the concepts discussed, setting the stage for deeper exploration in subsequent classes.