In a world where artificial intelligence is rapidly evolving, a quiet revolution is taking place in India—one that promises to make generative AI not just a tool of the elite but a transformational force accessible to every Indian. At the heart of this mission lies Sarvam AI, a groundbreaking startup setting out to build India-specific, voice-first foundational AI models. But Sarvam isn’t just building technology. It’s building inclusion, identity, and linguistic representation into the digital future of over a billion people.
Table of Contents
The Genesis of Sarvam AI: A Vision Rooted in Bharat
Founded in 2023 by Vivek Raghavan and Pratyush Kumar, Sarvam AI was born from a shared vision: to ensure India isn’t just a user of AI but a key contributor to its evolution. Both founders come from impressive backgrounds—Vivek was the former Chief Product Manager at UIDAI (Aadhaar), a pivotal figure behind India’s public digital infrastructure. At the same time, Pratyush is a leading researcher in AI and co-founder of AI4Bharat, an open-source initiative to build AI for Indian languages.
Their experience in deploying technology nationally laid the foundation for Sarvam AI’s core mission: to build sovereign AI infrastructure for India from the ground up, rooted in local languages and accessible through voice.
Identifying the Gap: The Language Barrier in AI
While global AI models like ChatGPT and Google Gemini have taken the world by storm, they largely cater to English-speaking audiences and high-resource languages. India, with its 22 official languages, hundreds of dialects, and a largely non-English-speaking population, has been underserved.
Sarvam AI recognised a critical gap: the lack of robust, Indian-language-focused AI models that could understand, respond, and generate content in native languages, especially through speech. This wasn’t just a technical gap—it was a cultural and economic one too. A voice-first, multilingual AI model could unlock massive utility for farmers, small business owners, students in rural areas, and millions who are digitally connected but linguistically excluded.
The Sarvam Approach: Indian Languages. Indian Voices. Indian Values.
Unlike many AI startups that build on top of existing Western models, Sarvam AI is committed to training its own foundational large language models (LLMs) tailored to Indian use cases. The startup’s approach is refreshingly India-first:
- Voice-first Interfaces: Recognising that typing in local languages remains a barrier, Sarvam focuses on speech-based interaction, which is more natural and intuitive for most Indians.
- Indian Language Training: The models are trained on Indian languages, including Hindi, Tamil, Telugu, Kannada, Bengali, and more, incorporating nuances like grammar, code-switching, and regional variations.
- Open Collaboration: Sarvam AI collaborates with open-source communities, government-backed platforms, and local institutions to ensure its models are scalable, inclusive, and safe.
A Landmark Partnership: EkStep Foundation Joins the Journey
In a major boost to its mission, Sarvam AI secured $41 million in funding in December 2023, led by Lightspeed, Peak XV (formerly Sequoia Capital India), and the EkStep Foundation, co-founded by Nandan Nilekani. EkStep’s involvement signals a powerful alignment with India’s digital public infrastructure movement.
This partnership underscores Sarvam’s role as a public-spirited innovator working to embed AI into India’s development story beyond just a start-up. By leveraging the ethos of India Stack and Digital Public Goods, Sarvam aims to build AI models that are safe, scalable, and deeply rooted in the country’s linguistic and cultural fabric.
Challenges and the Road Ahead
Building foundational AI models for India is no small feat. Data scarcity, dialect diversity, and bias mitigation are technical hurdles Sarvam must overcome. Unlike English, many Indian languages lack digitised corpora, making model training complex. Sarvam tackles this by leveraging crowd-sourced data and collaborating with AI4Bharat and other academic initiatives to create high-quality training datasets.
Then comes the challenge of inference infrastructure. Unlike Silicon Valley models that rely on expensive GPUs hosted overseas, Sarvam is working to localise AI infrastructure through edge computing and affordable cloud partnerships, ensuring its AI can serve both remote villages and urban centres.
Why Sarvam AI Matters
India is expected to become the third-largest AI market globally by 2026, with a projected value of $17 billion. Yet, for this growth to be inclusive, AI must speak the language of the masses. Sarvam AI is betting big on the future of voice and vernacular as the next frontier of digital access.
More than just a tech startup, Sarvam AI represents a powerful vision: AI that works for India, in India’s languages, and with Indian values at its core. As the world moves towards a future shaped by artificial intelligence, Sarvam ensures that the Indian voice—rich, diverse, and deeply rooted—will not only be heard but will lead the conversation.
