The honest gross margin of a typical AI-native SaaS product in 2026 is in the 40s, not the 80s. That is not a fundable number on classical software multiples, which is why so many of these companies are quietly being repositioned.
Where the next 30 points come from
Three places, in order of leverage: aggressive caching of expensive completions, routing between model tiers so the cheapest model handles the boring 80% of traffic, and — most importantly — owning more of the workflow so the price-anchor stops being the API and starts being the labor it replaces.
Model swapping is the least durable lever. Caching is more durable. Workflow ownership is the only one that compounds.
"Pricing power follows from the workflow you replace, not from the model you ship."
Founders who get this right stop benchmarking against other AI startups and start benchmarking against the cost of the human function they are removing. That is the only comparison the customer actually runs.
Dispatch Newsletter
One operator-grade breakdown every Tuesday. SEA tech, AI economics, execution frameworks. No fluff.
