Frontier AI in University:
Stanford's Two AI Economics Courses

The storyline's starting point is an industrial economics course taught by a VC. It doesn't teach "what AI is"; it teaches you how to judge where the money will ultimately settle in this supply-cycle supercycle.

"Economics of the AI Supercycle" documents Stanford's new Spring 2026 course MS&E 435: a 1-credit lightweight seminar, 9 weeks, 9 founder/CEO guests, taught by Apoorv Agrawal, a partner at Altimeter Capital — he led Altimeter's largest single investment in history, entering OpenAI at a $150B valuation. The entire course has no textbook; pre-reads consist entirely of first-hand industry memos he writes on his Substack, "Tailwind."

The course pursues a single question: in the AI supercycle, which layer will value settle into? The answer falls within a "semiconductor — infrastructure — application" three-layer stack framework:

Over two years, the Gen AI ecosystem grew 5x — from ~$90B to ~$435B — yet the stack's "shape" barely changed: the semiconductor layer still captures 79% of the entire ecosystem's gross margin. The course homepage also features an interactive bottleneck calculator: applying Liebig's law of the minimum, current compute demand growth is 310%/yr while energy supply growth is only 15%/yr — whoever is scarcest captures the margin. On Christmas Eve 2025, Jensen Huang — who had spent a lifetime mocking ASICs for failing to beat him — turned around and spent $20B to acquire Groq, providing the most dramatic footnote to this framework.

The storyline's second stop shifts to a different department, a different instructor, and a different posture — from "a VC's live memos" to "a scholar-hosted guest lecture series."

"The AI Awakening" is Stanford's ECON295/CS323, taught by Erik Brynjolfsson, Stanford HAI Senior Fellow and Director of the Stanford Digital Economy Lab, cross-listed between the Economics Department and the Computer Science Department, with admission by application. One sentence on the course page reveals its pedigree: "Over the past three years, guests have included Eric Schmidt, Mira Murati, Jeff Dean, David Autor, Condoleezza Rice, Reid Hoffman" — this course has run for at least three iterations.

The only publicly available complete syllabus is the Spring 2025 edition: 10 Tuesday lectures + Thursday discussion sections, with guests including Anthropic co-founder Jack Clark, OpenAI's first chief economist Ronnie Chatterji, MIT's David Autor and Daniela Rus, plus a special discussion guest Larry Summers. The final deliverable is a team (max 3 people) presentation of an AI startup/policy/research proposal, graded in five equal parts: weekly assignments, project progress, final presentation, final document, and discussion participation — each worth 20%.

Worth noting is one module in the Thursday discussion sections — "How do you pitch an AI startup to VCs?" — a VC pricing workshop citing data that VCs invested roughly $130 billion in AI startups in 2024. This shows that "teaching students how to sell AI concepts to investors" didn't emerge out of nowhere in 2026 — at the time, it was merely one discussion module, not yet the course's core deliverable.

A lecture course built over three iterations with top-tier guests迎来 its biggest pivot in Spring 2026.

The Stanford Bulletin's official course description contains an explicit pivot signal:

In plain language: for the past three years, this course was essentially a lecture series "inviting luminaries to discuss how AI changes the economy"; in Spring 2026, the deliverable becomes a real AI startup prototype — interdisciplinary teams develop, refine, and ultimately present a "runnable prototype + venture strategy" as their final showcase. Spring 2026 applications closed on 2026-03-16 (priority deadline 2026-03-09) — this is not a paper plan but an arrangement already implemented at the admissions stage.

A necessary disclosure: at the time of writing this storyline, the detailed weekly schedule and guest list for Spring 2026 have not been made public. This storyline, like the original articles it cites, does not fabricate any missing details — what can be confirmed rests on two layers of evidence: the pivot phrasing "emphasize venture creation" in the Bulletin, and the Spring 2025 syllabus serving as a baseline for "how this course previously ran."

Placing the two courses side by side in the same table reveals a signal larger than a course-selection guide.

Both courses are tilting toward "hands-on" — MS&E 435 has students use a computable framework to decompose the industry, while CS323 from 2026 onward directly requires students to turn the framework into something that runs. The Economics Department is no longer satisfied with teaching students "how to analyze the AI economy"; it is now teaching students "how to enter the AI economy." From "understanding the AI industry" to "building an AI company" — that is the single arrow direction between the two stops on this storyline.

For readers concerned with China's AI education ecosystem, the most worthwhile question is not "who can replicate the guest list as-is" — that's not hard. What's hard are the most hardcore parts of each course: behind MS&E 435's bottleneck calculator is first-hand industry data from sources like Epoch AI that are continuously updated; behind CS323's "runnable prototype" is the closed loop of "student prototype → real investor feedback → iteration," which requires investors genuinely willing to spend time reviewing student projects, not judges going through the motions. The activity level and pace of AI startup financing in China are also not in the same order of magnitude as the ~$130 billion U.S. AI venture market in 2024 — whether these two loops can be transplanted as-is, or require a different coordinate system, is a question everyone wanting to replicate such courses must answer for themselves, not a conclusion they can simply take away from this guide.