New Delhi, Oct. 1 -- "AI tokens are working capital," said Tarun Pant, Director of Customer Engineering, Google Cloud, while moderating a discussion on startup AI economics at Google Cloud Startup Summit 2026 in Bengaluru.

The line, in a way, captured a problem that kept coming up often through the event. The air was riddled with talk of smaller models, caching, model routing, observability and cost per business outcome, which are becoming central to what happens after an AI product moves beyond a demo and starts serving users at scale.

An agent may retrieve company data, reason through several steps, invoke software tools, call another model and retry an action before producing a result. Voice and video add more processing on top.

A t...