Dan Sundheim - The Art of Public and Private Market Investing - [Invest Like the Best, EP.460]

What It Covers

D1 Capital founder Dan Sundheim covers his simultaneous public and private market investing approach, with major positions in SpaceX, OpenAI, and Anthropic. He analyzes LLM business models through Netflix/Spotify frameworks, explains why hyperscalers face structural deterioration, addresses the GameStop crisis of early 2021, and identifies Taiwan semiconductor concentration as the single largest tail risk facing the global economy.

Key Questions Answered

• •LLM Business Model Framework: Evaluate AI companies as a hybrid of Netflix and Spotify. Netflix analogy applies to upfront model training costs amortized over growing users, creating flywheel economics. Spotify analogy applies to differentiation: since models are broadly similar in capability, personalization and accumulated user data history create switching costs and pricing power — not raw model performance superiority over competitors.
• •Private vs. Public Market Competition: Private markets have fewer participants but all pursue fundamental long-term value analysis. Public markets have vastly more participants but most operate on different time horizons and objectives — passive funds, quants, retail. This means public market inefficiency exists beyond short-term events: once analysis extends to three-plus year intrinsic value, the competitive set thins dramatically, creating exploitable opportunity for patient fundamental investors.
• •CEO Evaluation via Written Communication: Weight a CEO's written clarity heavily when making early-stage investments. Sundheim missed Amazon partly because he didn't read Bezos' 1997 shareholder letter, which demonstrated exceptional strategic clarity. He backed Anthropic's Dario Amodei after reading his essays, recognizing the same signal — a founder who can articulate precisely what they want to achieve and how, in writing, before results materialize.
• •Hyperscaler Structural Risk: AWS and Azure face a deteriorating long-term business model despite near-term acceleration. Their customer base is concentrating from thousands of enterprises to four or five LLM companies. When those LLMs become cash flow positive — likely within five to ten years — they will insource compute, as Meta already has. GPU cluster management also favors specialized neo-clouds over traditional CPU-oriented hyperscalers.
• •Focus vs. Breadth in Platform Companies: LLM companies face a structural tension: spreading fixed training costs across more end markets improves unit economics, but history shows few companies successfully pursue consumer, enterprise, science, hardware, and robotics simultaneously. Anthropic's concentrated focus on enterprise coding produced market-leading positioning. OpenAI's multi-front strategy carries execution risk despite exceptional talent density, and few historical precedents exist for that approach succeeding.
• •Drawdown Management and Investor Communication: During D1's GameStop-driven drawdown in early 2021, Sundheim held scheduled LP dinners in June 2022 at the trough rather than canceling them. The key message: shift to singles-and-doubles portfolio construction, accepting a slower path to high watermark recovery in exchange for reduced tail risk. Communicating a concrete plan — even during maximum adversity — reframes the narrative from collapse to controlled turnaround.