India is preparing to enter a new phase of artificial intelligence development with the much anticipated launch of PARAM-2, a 17-billion-parameter multilingual model, at the upcoming India AI Impact Summit 2026. The model has been developed under BharatGen, the country’s national generative AI initiative supported by the Department of Science and Technology.
With this release, a strategic bet is being placed on public digital infrastructure as the backbone of India’s AI ambitions. PARAM-2 has been trained on India-centric datasets under Bharat Data Sagar and supports all 22 Scheduled Indian languages. A Mixture-of-Experts architecture has been adopted, allowing complex multilingual tasks to be handled efficiently.
According to Professor Ganesh Ramakrishnan of IIT Bombay, one of the key leaders behind the initiative, the milestone represents the coming together of researchers, institutions, government bodies, and industry partners. He said the effort ensures India can shape its own AI trajectory rather than depend entirely on foreign systems.
A Sovereign Model Built for Indian Realities
As artificial intelligence reshapes global power structures, foundational models have largely been dominated by the United States and China. However, through BharatGen, India has quietly built domestic capability at the foundational layer.
Unlike consumer-facing platforms such as ChatGPT or Gemini, BharatGen has been structured differently. Its models are released as national public digital goods. As a result, government departments, banks, hospitals, courts, and educational institutions are allowed to deploy them locally. In some cases, deployment can even occur in secure, air-gapped environments without internet connectivity.
This decentralized framework has been designed to strengthen data sovereignty and institutional trust. Transparency has been emphasized in model development, and training processes have been documented. Furthermore, oversight remains within national systems.
Professor Ramakrishnan has described sovereignty as a guiding principle behind the project. He noted that transparency, sustainability, and trust are central concerns when AI systems are embedded in governance structures. Consequently, BharatGen’s architecture reflects those priorities.
Backed by Public Funding and Institutional Collaboration
The backbone of BharatGen is the National Mission on Interdisciplinary Cyber-Physical Systems under the Department of Science and Technology. An initial allocation of Rs 235 crore was approved to seed the project. Subsequently, it has been scaled further with Rs 900 crore under the IndiaAI Mission led by the Ministry of Electronics and Information Technology.
The programme has been implemented through a consortium model. Engineering institutions, AI researchers, management experts, and industry stakeholders have been brought together. Through this approach, development has been distributed across sectors rather than concentrated in a single entity.
Earlier iterations of the project laid the groundwork. The journey began with PARAM, a 2.9-billion-parameter bilingual model in Hindi and English. Later, a 7-billion-parameter system supporting 16 Indian languages was introduced. Along the way, speech-to-text, text-to-speech, and document vision-language models were also released.
Meanwhile, domain-specific versions have been fine-tuned for agriculture, Ayurveda, and the legal system. These models are currently being tested for integration into governance workflows and public service delivery.
Addressing Scale and Compute Challenges
Training large AI models requires significant computational power. Access to GPUs has long been seen as a constraint for India. However, thousands of GPUs have already been used to train BharatGen models, and additional capacity is being added.
At the same time, decentralization has been prioritized to address energy and scalability concerns. If inference were centralized for a population of over 1.4 billion, energy demand would increase sharply. Therefore, a scale-out design has been encouraged, allowing models to run closer to the point of use.
With the formal launch of PARAM-2, India is signaling its arrival at the foundational AI layer. Although global competition remains intense, domestic capability has now been demonstrated. Larger ambitions, including future trillion-parameter models, have also been outlined under the IndiaAI Mission.
