On February 18, India unveiled three sovereign artificial intelligence (AI) models at the India AI Impact Summit in New Delhi, marking a major push to develop indigenous alternatives to global AI platforms largely dominated by Big Tech firms.
The announcement comes amid the rollout of the IndiaAI Mission, cleared by the Union Cabinet in March 2024 with an initial allocation of Rs 10,000 crore. The mission aims to strengthen India’s sovereign AI ecosystem by building a domestic foundational model, expanding large-scale computing infrastructure, and promoting AI applications for public services.
The mission gained traction in late 2024 and early 2025 after the government rolled out GPU subsidies and invited startups to seek compute support. Since then, more than Rs 100 crore has been disbursed to support access to high-performance GPUs, with over a dozen companies selected under the initiative.
Bengaluru-based Sarvam AI, conversational AI firm Gnani.ai, and BharatGen, a sovereign AI initiative led by IIT Bombay, have unveiled a large language model (LLM) and voice models designed to be trained, deployed, and governed entirely within India.
“The developer energy I find in India is second to none. Recently, the work Sarvam has done developing local AI models, what you’re talking about is actually happening. India is very well positioned,” said Sundar Pichai at the India AI Impact Summit.
Two LLMs by Sarvam
In a key announcement, Sarvam AI unveiled two large language models, a 30-billion-parameter model and a 105-billion-parameter model, both built and trained from the ground up in India. The company said its larger model outperforms global systems such as DeepSeek’s R1 and Google’s Gemini Flash on multiple benchmarks, while adopting a mixture-of-experts architecture to lower inference costs.
The models are designed to handle advanced reasoning, programming tasks, and agentic AI applications, with the company emphasising efficiency as critical to delivering AI at population scale.
“Sovereignty matters much more in AI than building the biggest models,” said Sarvam co-founder Vivek Raghavan at the Summit.
Gnani.ai’s text-to-speech model
Gnani.ai introduced Vachana TTS, a text-to-speech model capable of cloning human voices across 12 Indian languages using less than 10 seconds of reference audio.
The company said the system retains key voice attributes such as tone, pitch and speaking style, while enabling the same voice to function seamlessly across multiple languages.
Designed for low-bandwidth environments and high-volume deployment, Vachana TTS is aimed at government services, customer support platforms and large-scale enterprise applications. The company added that both the data and models are hosted entirely within India.
BharatGen’s 17-billion-parameter model
The IIT Bombay-led consortium BharatGen launched BharatGen Param2 17B MoE, a 17-billion-parameter multilingual foundational model. Param2 17B is a Mixture-of-Experts (MoE) model optimised for multiple Indic languages, with the goal of accelerating AI adoption across governance, education, healthcare, agriculture and enterprise use cases.
BharatGen plans to release the open-source model, along with documentation and post-training workflows, via its Hugging Face repository, enabling developers, startups, and enterprises to build, fine-tune and deploy India-focused AI applications.
With Rs 900 crore in funding under the IndiaAI Mission, BharatGen has emerged as the largest beneficiary so far of the government’s sovereign LLM initiative.
