Most investors are still tracking revenue and user growth to evaluate AI companies. That worked for SaaS. It doesn't work for AI.

Here are the 3 metrics I actually care about:

1. Inference cost per query — This is the real arms race. The company that gets cost-per-query lowest wins the margin game. Watch how fast it's declining quarter over quarter.

2. Data flywheel velocity — How quickly does the product generate proprietary training data from usage? Companies with tight feedback loops between deployment and training are compounding faster than anyone realizes.

3. Integration surface area — How many external systems does the AI plug into? An AI tool that touches 50 enterprise workflows is 10x stickier than one that lives in a chatbox.

Revenue matters, but these leading indicators tell you which companies will still be growing 3 years from now.

I break down specific companies through this lens every week inside Frontier AI.