Who Owns the Generative AI Platform?

Author: Matt Bornstein, Guido Appenzeller, Martin Casado · Publication: Andreessen Horowitz (a16z)
yes
Summary: Analysis of value distribution across the generative AI stack — applications, models, and infrastructure — with data on margins, cost structures, and where economic value accrues.

Editorial Notes

The essential read for anyone building AI cost models or unit economics projections. a16z analyzes where value accrues across the generative AI stack and finds a counterintuitive pattern: infrastructure vendors (especially Nvidia) capture the most value, while application companies face 50-60% gross margins because inference costs consume 20-40% of revenue. For technical leaders, this reframes AI product economics fundamentally — your AI feature isn’t a software feature with near-zero marginal cost, it’s a service with real per-query costs that scale with usage. The article provides concrete benchmarks for cost structure planning that you can hand to your CFO with context. Compare these macro-level economics with the per-token pricing from Anthropic and OpenAI (covered in other paths) to build a complete picture from unit cost to gross margin.

Source: AI Strategy & Leadership

Appears in Learning Paths

Link last verified: 2026-04-08