Every few months an Indian founder announces a plan to build "India's answer to OpenAI," and every few months the arithmetic stays the same: frontier model training is a game measured in billions of dollars per run.
The honest opportunity is one layer up. India has proprietary problem spaces — vernacular commerce, smallholder agriculture, public health logistics — where domain data matters more than parameter count.
And India has a proven playbook for technology public goods. Aadhaar and UPI did not out-spend Visa; they out-designed it for Indian constraints. AI-as-DPI deserves the same ambition.
The founders who internalise this will build durable companies. The ones who cosplay San Francisco will build expensive demos.
