Pre-seed AI diligence: look at inference cost, not just the team

At pre-seed, most AI companies are an application layer sitting on someone else's model. That changes the nature of the risk: the question is no longer whether they can ship, but how much gross margin survives inference costs — and who controls pricing. Editor's note: source data feeds were unavailable in this session, so this piece covers diligence method only and cites no new figures.

• Inference cost per monthly active user, and how sensitive it is to a model vendor's price change.
• Share of revenue that is real cash, excluding promotional cloud credits.
• What proprietary data the company accumulates that a competitor cannot simply buy.
• How long a migration to an alternative model would take if the current vendor changes terms.