Week of 27 October 2026
Audience. Instructors and students in venture capital, entrepreneurial finance, private equity, and related electives.
Purpose. A primary-source investigation set: students extract capital-structure facts from contemporaneous company and investor disclosures, then compare financing design across cases.
Scope rule. This pack does not restate amounts, valuations, ownership percentages, or contractual terms. Those belong only in the linked Cap Table Teardowns and issue, each of which cites company or named-investor primaries. If a claim cannot be traced to such a primary, it is out of bounds for class discussion.
1. Pedagogical design
Most VC / entrepreneurial-finance syllabi already teach preferred stock, liquidation preferences, staged financing, CVC motives, and dilution on the whiteboard, using NVCA model documents and classic cases (see, e.g., Stanford GSB FINANCE 385; Wharton FNCE 7500; MIT 15.431; Booth PE/VC Lab). What those courses often lack is a current, verifiable record of how those instruments and side arrangements appear in live frontier-AI financings.
This week’s materials close that gap with three evidentiary layers:
Layer | Material | Analytical role |
|---|---|---|
Deep case | Cap Table Teardown #2: OpenAI | Reconstruct a multi-year financing path from disclosed primaries; identify structure change, strategic packages, and disclosed vs undisclosed cells |
Comparative case | Cap Table Teardown #1: Anthropic | Hold a second mega-round pattern constant enough to contrast dilution bands, multi-cloud strategics, and disclosure choices |
Compact rights case | Pathos AI / Bristol Myers Squibb issue | Isolate a single round in which the lead is also a commercial counterparty, so students practice reading rights that travel with equity |
The research design is comparative and source-constrained. Students may not “complete” a missing valuation or lead from press. Unknowns stay unknown.
2. Evidentiary standard (required)
Primary over secondary. Accept a figure only if it appears in a company announcement or a named-investor filing/press release linked from the teardown or issue.
No inference fill-ins. Pre-money, post-money, new-money share, and step-ups may be computed only from disclosed inputs; label every derived figure as derived.
Disclose the gap. “Not disclosed” is a valid finding. Treat it as diligence output, not as a failure of the reading.
Separate equity from commercial wrap. When a financing announcement sits beside a distribution, exclusivity, or compute agreement, analyze the package as a joint contract—not as “an investment” alone.
Cite the URL. Seminar answers must name the primary link, not “I read it in the newsletter.”
3. Learning outcomes
By the end of the session, students should be able to:
Map each assigned piece to at least one core syllabus module (term sheets / governance; strategic & CVC capital; cap tables & dilution; primary-source diligence).
State, with citation, why identifying the lead investor is analytically incomplete without asking what control, commercial, or structural conditions accompanied the capital.
Produce a side-by-side comparison of OpenAI and Anthropic late-stage financing design using only disclosed figures from the teardowns, and explicitly list what cannot be compared because it is not disclosed.
Apply the same rights-reading discipline used on NVCA model docs to a live strategic-lead round (Pathos / BMS).
4. Assigned materials
# | Material | Link | Investigation focus |
|---|---|---|---|
1 | Cap Table Teardown #2: OpenAI | Structure evolution; strategic packages; disclosed timeline; computable dilution math; “Not disclosed” cells | |
2 | Cap Table Teardown #1: Anthropic | Parallel mega-round pattern; multi-cloud strategics; late-round new-money band; disclosure completeness | |
3 | Issue: Pathos AI / Bristol Myers Squibb | Strategic lead as customer; exclusivity / platform access; rights that travel with the check |
Instructor note. Assign (1) as required; assign (2) and (3) as required complements for comparison. Do not substitute secondary AI-funding blogs for these primaries.
5. Syllabus crosswalk
Investigation task | Typical course module |
|---|---|
Reconstruct OpenAI’s disclosed financing path and governance/structure milestones | Term sheets; security design; staged financing |
Compare a strategic equity+commercial package (OpenAI teardown) with Pathos/BMS | CVC / corporate venture; side letters; partnership economics |
Compute and contrast late-round new-money shares (OpenAI vs Anthropic) where disclosed | Cap tables; dilution; valuation step-ups |
Catalog “Not disclosed” fields and state diligence implications | Primary-source literacy; IC memo hygiene |
6. Suggested 50-minute seminar protocol
Time | Activity | Product |
|---|---|---|
0–10 min | OpenAI timeline + “Not disclosed” audit | Board list: disclosed vs unknown |
10–25 min | One OpenAI strategic package vs Pathos/BMS rights | Two-column comparison: equity vs commercial terms |
25–40 min | OpenAI vs Anthropic late-round dilution (disclosed only) | Derived new-money % + constraints that would justify each band |
40–50 min | Founder negotiation: conditions that require mid-commitment structure change | Negotiation checklist (what to concede / refuse) |
Secondary press and chat tools stay closed during the comparison. Park any claim that lacks a primary URL.
7. Discussion questions (cite primaries)
Conditionality. In the OpenAI teardown, identify a financing or governance condition that gates a late closing. How should a founder negotiate a commitment that requires a capital-structure or governance change before funds fully clear—and what should remain non-negotiable?
Bundled counterparty risk. Select one strategic package from the OpenAI teardown and the BMS seat in the Pathos issue. When is bundling distribution or exclusivity with equity value-creating for the startup, and when does it create irreversible dependence on a single counterparty?
Disclosure-bounded comparison. Using only figures disclosed in the OpenAI and Anthropic teardowns, how do late-round new-money shares differ? What company constraint (compute, distribution, runway, control) would rationally justify each band—and which comparisons are invalid because an input is not disclosed?
8. Source protocol
Round-level citations appear inside the OpenAI teardown (and its verification notes) and in the Anthropic and Pathos pieces. This pack is an assignment frame, not a data source. Do not cite it for amounts or terms.
Not investment advice. The AI Cap Table is published for information only and is not investment, legal, or tax advice. Nothing herein is an offer or solicitation to buy or sell any security. Figures are taken from public primary sources we cite; errors remain possible. Conduct independent diligence and consult qualified advisers before any investment decision.
The AI Cap Table · [email protected] · theaicaptable.com