Ben Horowitz on AI Infrastructure, Economics and The New Laws of Software

What It Covers

Ben Horowitz, cofounder of a16z, speaks with general partner Alex Rampell at Fintech Connect in Deer Valley about how AI has invalidated two foundational rules of software — that money cannot solve engineering problems and that customer lock-in creates durable moats — and what this means for CEOs, investors, and infrastructure.

Key Questions Answered

• •Mythical Man Month Reversal: For 50 years, Fred Brooks' principle held that throwing money at software problems never worked — hiring 1,000 engineers couldn't close a two-year competitive gap. AI has invalidated this. With sufficient GPU compute and proprietary data, companies can now compress years of development into weeks, fundamentally changing competitive dynamics for both incumbents and challengers.
• •Moat Erosion Framework: Three traditional software moats — migration pain, proprietary data lock-in, and UI switching costs — are simultaneously collapsing. AI agents interact flexibly with any interface, data portability has increased, and code replication is faster. CEOs must identify value that exists entirely outside these three categories or face severe pricing pressure within a compressed timeline.
• •Infrastructure Bottleneck Sequencing: The US faces cascading AI infrastructure shortages in a specific sequence: chips arrive first, then memory becomes the constraint, then electricity becomes the binding limit. a16z has invested in physical transformer manufacturing — unchanged since electricity's invention — because grid capacity, not GPU supply, represents the most durable near-term ceiling on AI deployment.
• •Crypto as AI Authentication Layer: AI-generated deepfakes, personalized phishing, and synthetic identities make three verification problems urgent: proving human presence, proving individual identity, and cryptographically signing content. Horowitz argues blockchain infrastructure — not Google, Meta, or government databases — provides the game-theoretic trust properties needed for these verification systems, and also enables AI agents to function as independent economic actors.
• •Product Lifecycle Compression: The window for a differentiated software product has shrunk from a potential decade to potentially five weeks. Companies should evaluate whether revenue is actively shifting to competitors — requiring deep cuts and pivots — versus whether valuation has dropped while underlying customer relationships remain structurally defensible, as with Horowitz's board example Navan in corporate travel.