Model gross margin

Formula

(AI revenue - inference COGS) / AI revenue

Unit

%

Models

SaaS, Usage-based

What it is

Model gross margin measures the gross margin earned specifically on AI-powered revenue, net of inference costs. The formula: (AI revenue minus inference COGS) divided by AI revenue.

How to calculate it

Isolate revenue attributable to AI features — this may be a dedicated AI tier, a usage-based charge per action, or an allocated portion of a blended subscription if the AI capability commands a premium. From that revenue, subtract the direct inference costs: model API fees, compute provisioned specifically for model serving, and any other cost that would disappear if the AI feature were turned off. Divide the remainder by AI revenue. For SaaS products where AI features are bundled into a flat subscription, the allocation requires a cost-to-serve estimate per user based on observed token consumption — imprecise but necessary for understanding true unit economics.

Why it matters

This metric determines whether AI revenue actually funds growth or consumes it. It is the AI-era analog of the compute-heavy gross margin debate from the early cloud era. Traditional SaaS targets 70–85% gross margins; AI-native products often run materially below that in early stages because inference costs are high relative to price. The ICONIQ data cited in the notes puts median free-cash-flow margin for sub-$100M AI-native companies at approximately negative 126%, which reflects this structural margin compression at scale. Whether and how quickly model gross margin converges toward SaaS-like levels determines the business's long-run capital efficiency and fundraising profile.

How to read it

There is no published benchmark for model gross margin as a discrete metric. AI-native gross margin figures that do exist in the public record tend to be blended across a company's full revenue mix rather than isolated to AI revenue specifically, which makes them difficult to apply as a direct comparison.

As an Omega Point estimate, early-stage AI-native companies (pre-scale, with inference costs still above the long-run secular trend) typically report model gross margins in the range of 30–65% — well below the 70–85% SaaS norm. This estimate is reasoned by analogy: inference costs are currently a significant fraction of revenue for products priced at competitive SaaS rates, and the floor is set by the cost structure of foundation-model APIs; the ceiling is bounded by early companies that have achieved meaningful usage-based pricing or dedicated infrastructure efficiency. This is a directional figure — an Omega Point estimate — not a measured one.

The relevant trajectory question is whether model gross margin is improving as the business scales: efficiency gains from batching, caching, model optimization, or negotiated API pricing should push the metric upward over time. A company not seeing improvement in this metric as revenue grows is likely facing a structural unit-economics problem, not just a scale-of-operations problem.