Can an Indian startup build an AI model that competes with the likes of OpenAI, Google and Anthropic? A couple of years ago most people would have laughed. Today it is a serious national project, backed by government compute, venture money and some very determined founders. The goal is not always to beat the biggest models on every benchmark. It is to build AI that understands India: its languages, its accents, its documents and its price points.
Here are the Indian players working on their own large language models (LLMs) and foundation models, and why their approach is different from Silicon Valley’s.
Why India wants its own AI models
- Language. India has 22 scheduled languages and hundreds of dialects. Global models still stumble on many of them, especially in speech.
- Cost. Paying per token in dollars adds up fast for apps serving hundreds of millions of users. Smaller, efficient models can be much cheaper to run.
- Data and sovereignty. Government services, banks and hospitals are cautious about sending sensitive data abroad.
- Strategic independence. Like chips and energy, AI is now seen as core infrastructure.
To support this, the government launched the IndiaAI Mission, which includes shared GPU capacity for startups and researchers and support for building Indian foundation models.
The Indian companies building foundation models
1. Sarvam AI
Bengaluru-based Sarvam AI focuses on Indian languages, both text and voice. It has released models and tools for translation, speech recognition and text-to-speech across major Indian languages, and was selected under the IndiaAI Mission to build a sovereign Indian model. Its pitch is practical: AI that a farmer, a shopkeeper or a government office can actually talk to in their own language.
2. Krutrim
Founded by Ola’s Bhavish Aggarwal, Krutrim has built its own family of AI models and an AI cloud. It aims to own more of the stack, from models to infrastructure, and to offer Indian businesses an alternative to global AI platforms.
3. BharatGen
BharatGen is a government-supported initiative led by a consortium of academic institutions, including IIT Bombay. It is working on multilingual and multimodal models designed for Indian use cases, with an emphasis on open research.
4. Soket AI Labs
Soket AI is among the startups chosen under the IndiaAI Mission to develop foundation models. Its focus is on building capable models designed with Indian languages and efficiency in mind.
5. Gnani.ai
Gnani.ai has spent years on voice AI for Indian languages, powering customer support for banks and other enterprises. It is now working on voice-first foundation models, which makes sense in a country where many people prefer speaking to typing.
6. CoRover (BharatGPT)
CoRover, known for conversational AI used by large public-sector clients, has been building BharatGPT-branded models aimed at enterprise and government chatbots in Indian languages.
7. AI4Bharat and the open-source community
Not a startup, but impossible to leave out. AI4Bharat, based at IIT Madras, has released open datasets and models for Indian languages that many startups build on. Without this kind of shared groundwork, the whole ecosystem would move much slower.
Can they really compete with GPT-class models?
Honestly, matching the very largest global models across the board is extremely hard. Those models are trained with budgets running into billions of dollars and enormous amounts of compute. Indian startups are working with a fraction of that.
But “competing” does not have to mean being bigger. Indian teams can win by being:
- Better at Indian languages and code-mixed speech like Hinglish
- Much cheaper to run for high-volume consumer and government apps
- Easier to deploy on Indian servers for regulated sectors
- Tuned for specific jobs like customer support, legal documents or agriculture advice
What this means for businesses and builders
If you run a business in India, it is worth testing Indian models alongside global ones, especially for voice bots, regional-language support and anything involving sensitive customer data. If you are a developer, open Indian-language datasets and models are a great place to start building products global players are ignoring.
The bottom line
India’s AI story is not about one startup beating OpenAI. It is about a group of companies, labs and public programmes building AI that works for India first. If they get language, cost and trust right, they do not need to win every global benchmark to build very large businesses.
