The companies that later look like infrastructure rarely start as institutions. They start as a lab in the old sense: a few people, a hard problem, a loop they can run without asking.
We do not mean a wet lab by default. We mean a team that can produce evidence on a short cycle. A compiler that gets a new chip to do something it could not do last month. An assay that drops in cost. A robot that finishes a shift. A model that is better on Friday because Thursday’s customers used it.
That is the unit we underwrite.
Closed loops beat open narratives
A narrative can be about a market. A loop is about a mechanism. We want the second.
What goes in. What comes out. What you learn. How that makes the next cycle cheaper or more correct. If you cannot draw that on one page, you are still giving a tour.
This is why we are skeptical of “platform” at pre-seed and generous with “this one job, done in a way that compounds.” Platforms are what loops become after they have earned the right.
Small on purpose
State-scale comes later. Headcount is not a proxy for seriousness. A ten-person team that ships against a physical or scientific constraint will usually beat a fifty-person team that is still aligning on the story.
We partner at the size where craft still fits in one room. Studio when the room is being assembled. Fund when the room is already moving.
What we ask
Three things, in this order.
What is the loop. Not the TAM. The loop.
What have you run. A prototype, a partner, a failed experiment that taught you something expensive. Agency is the leading indicator.
What would make this a system instead of a project. The first customer, the first piece of proprietary data, the first reason a second team cannot casually copy you.
Send that. Skip the ceremony.