An AI investor treats hands-on model use as part of due diligence

Choudhury argues that a startup built around a temporary model limitation can lose its reason to exist after the next release. Understanding what models can do, and which constraints are likely to change, therefore becomes part of evaluating a company’s durability.

In this interview, he describes reading papers, talking to researchers, and implementing technologies himself. His reported spending and token consumption are unusually large and partly supported by subsidized access. They are his account of his practice, not an independently audited measure of investment skill.

The more transferable distinction is between a persistent customer problem and a passing capability gap. A useful product may retain value through workflow knowledge, distribution, integration, or service even as the underlying model improves.

Choudhury also argues for stronger Indian access to compute and data. Readers should keep his position in view: he manages an AI-focused fund and has a commercial stake in the ecosystem he is describing.