Is India Keeping Up?

Ever wondered why every AI company in India is becoming a voice AI company?

See who's building it
  1. 1 in 3Google searches in India are spoken

    India would rather talk to its phone

    Google put spoken search at about a third of Indian queries, against 3–4% in the US. Typing in English, or in a Devanagari or Tamil script on a cramped keyboard, is the hard way to use a phone. Code-mixed Hinglish is how people actually speak, and voice takes it as it comes.

    Prabhakar Raghavan, Google, via Inc42

  2. ₹5–8per connected minute for a human telecaller

    Call length 10:00

    Telecaller₹50–80

    AI agent₹20–30

    The call centre is the customer

    India runs the world's call centres, and Indian banks, lenders and brands still sell, collect and support by phone. An AI agent at volume costs ₹2–3 a minute. That gives a voice startup a buyer with a budget and a saving it can count. A general-purpose text model comes with no such sum.

    The Hundred-Buyer ProblemMarket Bites

  3. 22scheduled languages covered by free speech models

    The parts come free

    AI4Bharat and BHASHINI give any startup working recognition and synthesis in every scheduled language. Fine-tune those, rent a language model for the middle, and a calling product can ship in months. Pretraining a competitive text model takes clusters of GPUs and years of research, and India has very few teams who have done it.

    AI4Bharat IndicConformerBHASHINI

  4. 0.21%of the web's pages are in Hindi

    Speech can be recorded. Text has to already exist.

    Hindi has over half a billion speakers but makes up 0.21% of Common Crawl, the web scrape language models learn from; English is 41.9%. Every other Indian language is smaller still. Speech is a different problem: you can go out and record it. IndicVoices alone holds 23,700 hours from 51,000 people across 400 districts.

    Common Crawl, CC-MAIN-2026-39IndicVoices

  5. $329Mraised by Indian voice AI companies, January to August 2026

    2025$12.9M

    2026$329M, 25×

    Early winners made it fundable

    That's up from $12.9M in the same months of 2025, and Sarvam's unicorn round is most of it. Sarvam and Gnani showed founders and investors that Indian voice could be a business. The copies followed, and dozens of calling platforms now undercut each other at ₹2 a minute.

    TracxnTechCrunch

6text models from Indian labs since 1 January 2026

95Indian companies and labs building voice AI

there are so many voice ai companies in india... it is getting to a point where it is just ridiculous. people saw sarvam and the few others, and were like - this is it. the holy grail.
Samrith Shankar @samrithshankar ·

What people are saying

Who's building it

See the full directory