Martin Casado on Frontier Model Economics, AI's Slowdown, and the Cursor Deal - [Invest Like the Best, EP.498]

What It Covers

Martin Casado, a16z general partner, breaks down four distinct investor archetypes, explains why private markets are efficient and defensibility is overrated in growing markets, analyzes frontier model economics and the three-to-six-month model lifespan problem, and shares lessons from the Cursor acquisition — the largest private software deal ever completed.

Key Questions Answered

• •Four Investor Archetypes: Casado identifies four distinct investing mindsets: financial (SaaS metrics, market size as proxies), productive (technical approach creates value and market), frontier (founder-led YOLO bets on science-fiction-scale ideas), and speculative (pure market timing, no asset conviction). The critical failure mode is mixing frameworks — a disruptive tech investor demanding strong NRR is internally incoherent and will systematically avoid the most disruptive opportunities.
• •Frontier Model Lifespan Risk: The lifespan of a state-of-the-art AI model is three to six months before competitors erode its market position. Unlike traditional software with long retention tails, a frontier lab that slips its release schedule by even three months risks losing its entire customer base. This creates a structural treadmill where labs must continuously ship or face near-total revenue collapse — a risk not captured in standard software valuation models.
• •Competitive Equilibrium Prevents Model Commoditization: Even when 95% of coding tasks are saturated, companies in competitive markets will not downgrade to cheaper, older models. Because rivals are also using frontier models, switching to a cheaper option is unilateral disarmament. This dynamic sustains frontier model pricing power even as token consumption plateaus, and forces revenue growth to come from broadening use cases — knowledge work, automation — rather than volume increases.
• •Moat Shape Matters More Than Mechanism: Investors conflate the mechanism creating a moat (network effects, scale, two-sided marketplace) with its actual shape. The shape — superlinear, linear, or sublinear with scale — determines real defensibility. A superlinear moat like Metcalfe's Law (LinkedIn, GitHub, telcos) produces natural monopolies. A sublinear moat means challengers with equal capital will always catch up. Casado argues most "network effects" cited in pitch decks are actually sublinear and provide minimal long-term protection.
• •Strategic M&A Requires Trust, Not Poker: The Cursor acquisition — the largest private software deal ever — succeeded because founders engaged buyers in good faith without leaking to press or running competitive auctions against sophisticated counterparties like Elon Musk or Mark Zuckerberg. Casado's framework: show the buyer a base case (existing revenue) plus an inductive step (how their assets multiply your market), treating the conversation as a high-stakes enterprise sale built on trust rather than a commodity auction.
• •Capital Now Solves Engineering Problems: The Mythical Man Month principle — that throwing engineers at software problems makes them slower — is being overturned by AI. A team of 20 people can now productively deploy a billion dollars where previously that capital would sit idle or detonate the organization. This transforms venture capital from talent-network arbitrage into capital-allocation strategy, and Casado argues private markets have unlimited TAM — more capital deployed creates proportionally more value, not diminishing returns.