How to Back or Build The New Thing
AI and Beyond
Whatever the new hotness is, there’s always three ways to back it or build it and the same general pattern tends to play out with investor and founder interest:
Foundational production: making the thing (possible). This is hard, capital intensive, and takes patience but obviously hits big sometimes. Think foundation models, semis, crypto L1s, etc.
Monetizing activity: clipping a rake/building infrastructure that solves generalizable problems once there’s enough usage for those problems to generalize. Think ML ops, AI observability, crypto infra, BaaS, software for hardware, etc. This is the “easiest” bet to feel smart making because you’re calling a direction, not a specific outcome.
Applications: using the new thing. What’s newly possible now that THE THING is better, cheaper, faster. What new services, businesses, products, or business models become viable? This is hard and sometimes capital intensive.
Applying this to AI specifically:
Production (“making” AI): These are principally the foundation models and the chips. We can debate over whether or not foundation models are good long term bets with moats or commodities (I’m in the middle here: there opportunities to de-commodify them through usability and features) but they are obviously not appropriate bets for early stage funds and only make sense for founders who can somehow bend the capital markets.
Middleware (monetizing AI): this is all the rage is clearly like building on quicksand - the technology and the use cases aren’t solved/solid enough yet. Michael Dempsey says that early in the cycle “the pickaxe viewpoint is for those with lower conviction and poor ability to reason through near to mid-term futures over time.”
Application (using AI): There’s AI products (think CRM powered by AI) and then there’s AI-based-businesses (companies using AI internally). AI will transform/power big companies products and small companies businesses. The AI CRM will probably be Salesforce BUT the small companies can punch well above their weight with AI even if the existing products have the data and distribution to win their markets. The caveat will be entirely new products and services that weren’t previously possible (not just enterprise apps with a chat interface).
TL;DR - there’s good and bat bets in AI
Good bets: Early bets in small companies because they should generally be more profitable faster as AI gets better, and new businesses/models that wouldn’t have previously been possible/economical.
Bad bets: foundation models, middleware, and AI versions of existing services.