AI gross margin
AI gross margin is the percentage of AI-feature revenue left after subtracting the direct cost of model inference — vendor API spend, compute, and hosting — and it runs structurally thinner than traditional software margin because inference cost scales with usage rather than staying fixed.
Why AI gross margin behaves differently than SaaS gross margin
Traditional hosted-software gross margin improves with scale: fixed infrastructure costs get spread across a growing customer base, so cost of goods sold shrinks as a percentage of revenue. AI products don't get that curve for free. Every request triggers a variable, usage-linked cost — the model inference itself — so gross margin on an AI feature depends on the spread between what a customer is charged per token or per action and what the underlying model vendor charges, and that spread does not automatically widen as volume grows.
Traditional software businesses have long operated with gross margins in the 70–80% range, as tracked in SaaS Capital's annual B2B SaaS Benchmarks survey. AI-native product lines are frequently reported well below that band precisely because inference is a direct, variable cost rather than an amortized fixed one. Source: SaaS Capital, B2B SaaS Benchmarks report.
What drives AI gross margin up or down
The biggest levers are model choice (larger, more capable models cost more per token), prompt and context-window efficiency (shorter prompts and tighter output limits cost less per request), vendor discount tier or committed-use pricing, and how the customer-facing price is structured — a flat per-seat price on top of unbounded usage compresses margin at the tail of the usage distribution far faster than a metered price that tracks cost directly.
Measuring it accurately
Computing AI gross margin correctly requires attributing actual inference cost down to the customer or feature level (see cost attribution) and reconciling that cost against the vendor's actual invoice rather than list price, since committed-use discounts and negotiated rates change the real cost basis (see blended cost vs. list price). Margin figures built on list price alone systematically understate true margin.