Something Is Changing in the Unit Economics of Software

Software's defining superpower was zero marginal cost: build once, serve millions. That created the 75-85% gross margins and "grow now, margins later" playbook that defined SaaS. AI breaks this. Every LLM call costs real money, making per-user compute costs that scale directly with usage. AI turns software's cost of goods sold from negligible to large, sticky, and inseparable from the core experience.

This introduces a new tradeoff: model quality vs. margin. Use a cheaper model and protect economics but risk losing users to competitors with better experiences. Use a frontier model and win on product but bleed on cost. ICONIQ survey data puts average AI product gross margins at ~52% in 2026, up from 41% in 2024, but still well below the 75-85% SaaS baseline.

AI also increases variability in cost-to-serve: a power user making hundreds of inference calls costs dramatically more than a casual user. Flat-rate subscriptions that worked for traditional SaaS break when heavy users can blow up margins. Usage-based pricing is the rational response — the pricing model is changing because the cost model changed first.

Falling inference costs don't fully solve this. Jevons paradox predicts cheaper inference gets consumed by deeper integration, more calls per interaction, more autonomous agents — cost per call drops but calls per user multiply. The playbook that emerges looks less like SaaS-era "burn on growth" and more like capital-efficient businesses where unit economics matter from day one.

The zero marginal cost era produced extraordinary companies. It also produced businesses that mistook a favourable cost structure for genuine competitive advantage. AI is forcing a harder conversation about what it actually takes to build something sustainable.