The inference margin is the new gross margin
Application companies are quietly becoming infrastructure companies. Whoever owns the cost curve owns the category.
For a decade, software gross margins were a solved problem: ship bits, collect eighty points. Intelligence broke that assumption. Every request now carries a variable cost that scales with usage, and the companies that treat it as someone else's problem are underwriting a margin they do not control.
The teams we back treat model routing, caching, and distillation as product surface rather than infrastructure hygiene. In our portfolio, the median seed-stage company cut cost per resolved task by 71% in its first eighteen months — and none of that came from a cheaper provider.
The practical test for founders is simple: can you describe, in one sentence, how your unit cost falls as your usage rises? If the answer depends on a vendor's roadmap, it is not yet a business model.