Three weeks in Bangalore: observations on India's AI startup ecosystem (keldenddorji.medium.com)

After three weeks in Bangalore, the author reports a bifurcated AI startup scene: a broad base of companies serving Tier‑2/3 India while a smaller set of Tier‑1 teams aim to go global and attract VC checks. Notable plays include Sarvam AI — positioning itself as a full‑stack generative platform with backing from Tata, Aadhar, Infosys and government links — NimbleEdge building on‑device AI (a deep‑tech, long‑horizon bet), and SarvMAI, which shipped a voice‑ERP in 12 languages and 500+ dialects at ~90%+ accuracy for ~$300K and is eyeing Southeast Asia and Eastern Europe. The piece flags Groq as a likely winner in inference infrastructure as it scales in India, and observes many GenAI startups carry frothy ~33x valuations, suggesting more realistic exit paths might be M&A into SMB or enterprise verticals rather than perpetual VC raises. Ecosystem builders matter: community hubs like Draper Startup House and South Park Commons are helping early-stage founders and deep‑tech experimentation, but VCs, corporates and community groups remain siloed. Key implications: India’s deep‑tech base is still small versus the US, Bangalore punches above its reported rank (14th in one index) but needs a global “home run” (a Postman‑scale success) in the next five years or investor appetite could wane. Government partnerships and a $1B+ deep‑tech alliance offer upside but will require careful navigation to balance policy, market dynamics and private innovation.