As global AI splits between closed, proprietary systems and openly released models, a notable share of India's AI effort has leaned toward the open end of the spectrum. Rather than guarding weights and datasets, several Indian labs and startups have chosen to publish them — a strategy that reflects both practical constraints and a strategic bet.
Who is doing it
AI4Bharat, the IIT Madras lab, has released open datasets, translation systems and speech and language models for Indian languages, giving developers building blocks they could not otherwise assemble. Startups building Indic-focused models have released open variants aimed at Indian languages, and the government's IndiaAI Mission has emphasised shared compute and datasets as public goods. The common thread is treating foundational AI capability as infrastructure to be spread rather than a moat to be hoarded.
Why open
The logic is partly economic and partly strategic. India lacks the enormous private capital that funds the largest closed models, so pooling data and releasing models openly lets a wider community improve them. Open models also address a sovereignty concern: relying entirely on foreign, closed systems for something as fundamental as language understanding is uncomfortable for a country that wants control over its own digital future. And because India's core AI problem — supporting dozens of under-resourced languages — is under-served by global players, open releases seed an ecosystem that commercial products can then build on.
The open approach is not a guarantee of commercial success; someone still has to build durable businesses on top. But for a country trying to build AI capability quickly and inclusively, releasing work in the open may be the fastest way to catch up.
