Google is adding more Indian languages to its services and developing an AI model that would be able to handle 100+ Indian languages across speech and text
Sankalp Phartiyal / Bloomberg :
Context & Ripple Effects
Google's India language push is a decade-long build-out reaching a new stage: after bringing neural machine translation to nine Indian languages in 2017 and open-sourcing a multilingual model covering 16 Indian languages in 2020, the company is now training a dedicated model for 100+ Indian languages spanning both speech and text. The technical foundation came from its 400+ language speech model and the Translate expansion that added 110 languages via PaLM 2.
First-order effects
- Indian-language speakers get services that work natively in speech and text rather than through English-first interfaces, and developers building for India inherit a far richer base model than the 16-language open source release offered.
Second-order effects
- With Google claiming 700M to 800M users in India and ramping AI across its products there — including Gemini Live in nine Indian languages and an experimental Hindi rollout — rivals like Amazon and Microsoft, which have pledged part of the combined $67.5 billion India investment surge, face pressure to match language coverage rather than just data-center spend.
Third-order effects
- If the 100+ Indian language model performs, it validates a two-track approach — one global foundation model plus region-specific models — and extends Google's stated ambition toward the 1,000 most spoken languages, making language coverage a structural moat in emerging markets.
The trend: AI platforms are competing for the next several hundred million users by building language-specific models, turning localization depth in markets like India into a distribution advantage.