Full Transcript
Loading transcript…
Stanford MS&E435 Economics of the AI Supercycle | Spring 2026 | Infrastructure, Capstone Case
Watch with subtitles, summary & AI chat
Add the free Subkun extension — works directly on YouTube.
- Watch
- Subtitles
- Summary
- Ask AI
Individuals interested in the business and technical aspects of AI infrastructure, including investors, engineers, and tech strategists.
TL;DR
This video discusses the economics of the AI supercycle, focusing on compute infrastructure. It highlights the correlation between compute capacity and revenue at OpenAI, the increasing importance of inference workloads, and the complex, multi-faceted supply chain required to scale AI.
Key Takeaways
- Intel is experiencing a resurgence due to global supply constraints in manufacturing and the increasing relevance of CPUs for AI agents, positioning it as a key player in leading-edge American manufacturing.
- OpenAI's compute capacity growth is directly correlated with its revenue, with compute tripling year-over-year and revenue following suit, indicating compute as the primary driver of their business.
- OpenAI's mission as a research lab prioritizes maximizing compute for researchers to explore new ideas and push AI frontiers, rather than solely focusing on revenue maximization.
- The demand for compute is projected to continue its upward trajectory, with OpenAI's aspirational goal of 30 gigawatts split between research and products, reflecting strong future utilization.
- Scaling laws in AI now encompass the entire compute lifecycle, including pre-training, post-training inference, synthetic data generation, and actual product usage, with inference becoming increasingly dominant.
- Inference is projected to constitute a supermajority (over 80%) of future compute needs, driven by product usage, research, and the development of next-level AI intelligence.
- The entire compute supply chain, encompassing chips, memory, networking, power, cooling, data centers, and land, must be orchestrated and secured to deliver compute at a gigawatt scale.
In This Video
- 00:00Introduction to AI Supercycle Economics
Welcome to the AI Supercycle Economics course. Professor Kati joins to discuss the full spectrum of AI infrastructure.
- 00:57Intel's AI Comeback and Tailwinds
Intel's recent AI focus is driven by supply constraints and the resurgence of CPUs for AI agents.
- 02:44OpenAI's Compute Growth and Revenue
OpenAI's compute capacity has tripled yearly, correlating directly with revenue growth and user demand.
- 05:02OpenAI as a Research Lab
OpenAI prioritizes research, aiming to maximize compute for exploring new AI frontiers and ideas.
- 05:51Shift Towards Inference Compute
The majority of compute is shifting to inference, including training, synthetic data, and product usage.
- 07:18Monetization and Token Efficiency
Increasing inference use aims to boost revenue, while OpenAI also focuses on making tokens cheaper and more intelligent.
- 09:06The Compute Supply Chain Bottleneck
Securing compute involves a complex supply chain: chips, memory, networking, power, and data centers are all critical.
Questions & Answers
What is the AI supercycle?
The AI supercycle refers to the economics surrounding the rapid advancement and adoption of artificial intelligence, driven by increasing compute power and demand.
How has Intel's AI strategy evolved?
Intel spent about a decade positioning itself as an AI company, and recently saw a resurgence due to global supply constraints and the comeback of CPUs for AI agents.
What drives OpenAI's compute capacity growth?
OpenAI's compute capacity growth is closely correlated with its revenue, as revenue is a lagging indicator of compute utilization for training and inference.
What is the trend in AI compute usage between training and inference?
The trend is shifting towards inference, which already constitutes the majority of compute usage and is predicted to be over 80% in the future.
What are the main challenges in securing AI compute?
Securing AI compute involves sourcing a complex supply chain including chips, memory, networking, power, cooling, data centers, and land, all needing to align simultaneously.
How does OpenAI aim to make AI intelligence accessible?
OpenAI aims to make intelligence accessible by reducing token costs, increasing token intelligence, and minimizing tokens needed per task through hardware and software improvements.
Key Terms
- AI Supercycle — The period of rapid growth and investment in AI driven by technological advancements and increasing demand.
- Compute — The processing power required for AI, encompassing chips, memory, networking, power, and data center infrastructure.
- Scaling Laws — Principles that describe how AI model performance improves with increased compute, data, and parameters, now applied across the entire AI lifecycle.
- Inference — The process of using a trained AI model to make predictions or generate outputs, a growing component of AI compute demand.