Martin Casado on Where the Value Is Going in AI

Martin Casado, a General Partner at Andreessen Horowitz leading their infrastructure practice, explores the transformative economics of AI in venture capital. He argues that the primary innovation isn't just technical sophistication but rather the unprecedented ability to turn capital directly into growth. Historically, giving a startup a billion dollars would result in hiring inefficiencies and organizational bloat (the mythical man-month problem). Today, teams as small as 20 people have productively deployed $2 billion in compute to train major models, with capital translating directly into capability, usage, and growth in ways never before seen.

Casado discusses two competing theories about AI's future: whether frontier labs (OpenAI, Anthropic) will capture the entire market, or whether value will distribute across the stack. He presents strong arguments for labs capturing 80% of dollar-weighted value due to their capital access, talent, pricing power, and supply control. However, he also argues for fragmentation: the service area is expanding beyond current model strengths (primarily code and language), open source models are maturing, capital access will rationalize, and supply constraints will ease around 2028. He predicts labs will get 80% dollar-weighted but long-tail models and open source will capture 60% token-weighted, with applications increasingly capturing margin share.

The recent acquisitions exemplify this distributed value creation. OpenRouter functions as a two-sided marketplace aggregating the long tail of models, providing visibility, analytics, and routing capabilities. While model routing (selecting the optimal model for a task) appears to be an AI-complete problem, current gains come primarily from cost optimization rather than quality optimization. Cursor's $60 billion SpaceX acquisition reflects its dominant position in software development tools combined with SpaceX's compute and resources—neither company alone would be worth that amount.

Casado emphasizes that current criticism of these deals often focuses narrowly on financial metrics (margins, churn, revenue quality) while missing their strategic value as control points in the emerging stack. He argues that in transformative waves like this, one should avoid zero-sum thinking and focus on identifying which pieces of the emerging infrastructure will command strategic importance. The distinguishing factor of Cursor's success was treating the problem as a product challenge rather than a research/model architecture challenge, maintaining focus while iterating rapidly.

On hiring and investment strategy, Casado emphasizes founder-market fit over either founder quality or market size alone. His team at A16Z Infra prioritizes people with product backgrounds who understand the interface between market evolution, product fit, and technology. He describes himself as bullish on AI overall and willing to trust founders in spaces he understands rather than second-guessing their strategic choices. The biggest unlock he's seen in his career is the current moment where value is being created across multiple layers of the AI stack simultaneously.