E371: Midas List VC: Why AI Models Will NOT Become Commodities

Episode Summary

Executive Summary: The discussion centers on why Anthropic appealed as an investment versus OpenAI: a developer-led, feedback-loop-driven model that could compound through APIs, enterprise use, and network effects. The conversation broadens into durable investing principles—values as business strategy, first-principles thinking, immigrant and founder otherness, ownership versus valuation, and where AI may create value next, especially in B2B, physical AI, and the broader AI stack.

Main Topics: Why Anthropic was favored over OpenAI (Priority: 5/5): The investor explains choosing Anthropic because its developer/API usage suggested a compounding feedback loop and durable moat, especially for enterprise and B2B use cases. Feedback loops, network effects, and compounding product quality (Priority: 5/5): The thesis is that repeated usage by developers and enterprises improves the model, similar to consumer winners like TikTok, Airbnb, Google, and Intel Inside-style embedding. Values as a practical business advantage (Priority: 4/5): Examples like Airbnb’s COVID response are used to argue that culture and values translate into dollars, trust, loyalty, and long-term enterprise resilience. Founder psychology, immigrant otherness, and contrarian investing (Priority: 5/5): The speaker argues that successful founders and investors often come from being different, learning to think independently, tolerate loneliness, and act with intellectual honesty. Ownership, valuation, and fund construction (Priority: 4/5): The conversation explores why VC returns depend on owning enough of a few winners while still maintaining diversity, and why fund size and ownership targets matter. AI-native opportunities and defensibility (Priority: 5/5): The speaker outlines ways AI apps can defend against advancing LLMs: superior UX, owning system-of-record data, taking balance-sheet risk, specialized domain/compliance expertise, and physical AI. Long-term industry and geopolitical outlook (Priority: 3/5): The dialogue touches on AI stack layers, energy, chip infrastructure, global manufacturing partnerships, and historical perspective as inputs to investing and strategy.

Key Arguments: Anthropic’s developer/API activity signaled a compounding feedback loop, where model improvement comes from repeated use by developers and enterprise workflows. A business embedded into workflows can become more valuable over time, similar to Google inside Yahoo or Intel Inside. Enterprise/B2B is the highest-value path when model tokens are expensive, because businesses can monetize AI more directly than consumers can. Airbnb’s crisis response during COVID showed that values are not abstract—they create trust, loyalty, and concrete financial outcomes. Public-market quarterly reporting can distort long-term value creation by forcing short-term thinking and managerial distraction. Great founders often have the mental fortitude to stay right for a long time while being lonely and unpopular. Immigrant and outsider experiences can create an edge because they force people to develop independent thinking and adaptability. VC success depends on making enough bets, owning enough of the true winners, and staying flexible on valuation and portfolio construction. AI-native companies must defend themselves against model improvement by owning UX, data, regulated workflows, or economic risk. Physical AI is attractive because most of the world’s economy is physical, and specialized models may emerge in many industries rather than one universal model.

Data Points: Anthropic API business share: Smallest business by revenue in 2024 - Used to argue the most strategic signal was not consumer scale, but developer adoption and feedback loops. Airbnb employee reduction: 25% - Mentioned in the COVID-era response after cancelations and support measures. Immigrant representation in venture funding: 50% - Speaker cites a statistic that first- or second-generation immigrants make up about half of venture-funded founders. AI stack layers: 4 - Described as LLM stack, app stack, chip stack, and energy stack. US economy tied to services: 70% - Used to explain why B2B software and developer tools are especially valuable in the US. Non-US GDP tied to physical industries: 90%+ - Used to support the case for physical AI globally. Spotify/consumer era timeline reference: 2013–2018 - Speaker references riding consumer internet investments during this period in the US. China visit year: 2005 - Used as an example of contrarian geographic investing before China became obvious. Return to China with wife: 2013 - Referenced as the period when China was hot and consumer internet investing was strong. Podcast episodes: 300+ - Host references having asked similar ownership/valuation questions over more than 300 episodes. Founder/venture statistic: 30 other VC firms - Referenced in launching a prosumer AI-40 list with other firms and partners. Prosumer adoption example: 3 years ago - Used to note early users of ChatGPT/Claude-like products were the first adopters of innovation.

Pivotal Quotes: "The API business for Anthropic in 2024 was their smallest business in terms of revenue size. But what is interesting about that business is that you can see a lot of developers using Claude for code." — Pons: Explaining the original Anthropic investment thesis. "When you talk about why values matter, we're talking about dollars and cents and zeros and ones. That's why it matters." — Pons: Using Airbnb’s COVID response to show culture has economic value. "To be different, give you an edge, but also are you intellectually honest? It’s not about pride, it’s about figuring out, look at everything from first principles." — Pons: On founder psychology and contrarian investing.

Implications: For investors and founders, the message is to seek compounding feedback loops, durable B2B/enterprise value, and real moats beyond hype. AI winners may come from workflow embedding, data ownership, or specialized physical industries—not just model scale.