Yotta Data Services has joined forces with artificial intelligence (AI) startup Sarvam AI to build Sarvam-1, India’s first large open-source language model (LLM) focused on Indian languages.
This model, designed as a foundational tool for AI in India, was trained on Yotta’s Shakti Cloud infrastructure using a dataset of approximately four trillion tokens.
The partnership aims to support local innovation in AI, offering Indian businesses access to advanced AI technologies without reliance on foreign cloud services.
Sarvam-1: A model for multilingual AI
The Sarvam-1 model distinguishes itself as India’s first native multilingual LLM, trained from scratch to work with 10 Indian languages—Hindi, Tamil, Telugu, Malayalam, Punjabi, Odia, Gujarati, Marathi, Kannada, and Bengali—alongside English.
Built using NVIDIA H100 GPUs on Yotta’s infrastructure, Sarvam-1 leverages around two billion parameters to ensure fast, high-quality language representation and inference efficiency. The focus on Indian languages aligns with Sarvam AI’s mission to make AI accessible and practical for local industries and communities across India.
AI-powered tools for businesses
Sarvam AI has already rolled out Sarvam Agents, a suite of AI-powered tools designed for conversational applications. These AI agents can interact with users via phone, WhatsApp, and other apps in natural, voice-based conversations, supporting various industries with services such as customer support and employee engagement.
This suite, available in all 10 supported Indian languages, empowers businesses to provide better service in their preferred languages, which is critical for user engagement in India’s linguistically diverse market.
Harnessing NVIDIA technology for advanced performance
Yotta’s Shakti Cloud infrastructure, enhanced by NVIDIA's advanced computing solutions, provides Sarvam-1 with the capability to process large datasets with impressive speed and accuracy.
The infrastructure includes NVIDIA Hopper GPUs and the NVIDIA AI Enterprise software platform, facilitating efficient model training and data handling.
This combination has allowed Sarvam-1 to achieve up to four to six times faster inference speeds than its international counterparts, positioning it as a practical and competitive choice for businesses needing reliable and scalable AI solutions.
Building India’s digital future with AI
The partnership between Yotta Data Services and Sarvam AI is positioned as a foundational step in advancing India’s AI landscape.
By developing Sarvam-1 on local infrastructure and creating a model optimized for Indian languages, the partnership not only enhances data sovereignty but also ensures a cost-effective solution for Indian businesses.
Leadership comments
“Partnering with Sarvam to create the Sarvam 1 foundational model, leveraging Yotta’s Shakti Cloud infrastructure, marks a significant milestone in advancing India’s AI landscape. Powered by cutting-edge chips and state-of-the-art infrastructure, Shakti Cloud accelerates AI model training, offering businesses the resources they need to drive innovation. This collaboration is focused on equipping Indian companies with world-class AI capabilities, enabling them to compete and excel in the global digital economy,” said Darshan Hiranandani, Co-Founder and Chairman of Yotta.
Kari Briski, Vice President of AI Software at NVIDIA, said, “India is emerging as a global leader in AI innovation. Working with local technology experts like Yotta, NVIDIA supports the development of sovereign large language models and applications to help India’s businesses and people use AI to drive economic and social growth.”