Sarvam AI, a relatively young artificial intelligence startup based in Bengaluru, has taken center stage in India's tech ambitions after being selected by the Ministry of Electronics and Information Technology to develop India’s first homegrown large language model (LLM). This decision is part of the government’s IndiaAI Mission, a national effort to ensure that India has its own digital infrastructure for AI, one that is tailored to the country’s multilingual and multicultural context.
Unlike dominant global AI models, which are predominantly trained on English and other Western languages, Sarvam AI aims to create models that can understand and generate text in Indian languages such as Hindi, Tamil, Bengali, Telugu, and Marathi, among others. This move is crucial in a country where less than 10% of the population speaks English fluently.
To support Sarvam AI, the Indian government is providing access to high-performance computing resources—including 4,000 GPUs for six months—and financial grants. Sarvam’s approach includes training models from scratch, rather than fine-tuning existing foreign models, which allows for deeper cultural alignment and performance optimization in Indian contexts.
The company has already made significant progress. It previously released OpenHathi-Hi-v0.1, a state-of-the-art Hindi language model built on Meta’s LLaMA-2 architecture. The model is open-source and has shown promising results in benchmarks for translation, summarization, and question-answering tasks in Hindi.
Sarvam AI has also collaborated with the Unique Identification Authority of India (UIDAI) to develop voice-based services for Aadhaar users, enhancing access to government services for the non-literate or digitally underserved populations.
The government’s backing of Sarvam AI highlights a shift in India’s AI policy—from consumer to creator. With rising concerns about data sovereignty and dependence on Western AI tools, there is a strong push to build domestic AI capabilities, both to preserve linguistic diversity and to control critical tech infrastructure.
The success of Sarvam AI could set a precedent for other countries in the Global South looking to build culturally and linguistically relevant AI tools. If successful, India may not only cater to its massive internal market but also become an exporter of language models to other multilingual nations across Asia and Africa.