India is planning a GenAI project tailored to the nation’s unique cultural and linguistic diversity.
India’s Union IT Minister, Ashwini Vaishnaw, has announced plans to develop a generative AI model powered by the India AI Compute Facility specifically designed for the unique needs and contexts of India.
The project has secured 18,693 high-end Graphics Processing Units (GPUs), including Nvidia H100s and H200s, to support the development of the India-focused large language model (LLM). The minister stressed that even the most novel AI models, like the recent Chinese sensation – DeepSeek, were trained on a modest number of 2,000 GPUs.
The substantial investment in computing infrastructure underscores India’s commitment to advancing AI research, providing startups, researchers, and academic institutions with the necessary resources to innovate.
Besides, the country aims to eliminate possible bias which can be present in existing LLMs predominantly developed by U.S. companies by incorporating diverse local languages and cultures into a future AI tool.
If finalised successfully, this initiative might significantly enhance digital inclusivity and accessibility for India’s vast and diverse population. It could also potentially drive economic growth and enhance the country’s competitiveness in the technology sector.
Additionally, this move reflects India’s strategic focus on leveraging AI for local challenges while also contributing to global innovation. The country’s national AI strategy presupposes incubating AI projects that can apply to the rest of the developing world, investing in AI research and specific sectors that can maximise the country’s economic growth and social impact, and trying to tackle some of the global challenges from AI’s perspective.
To reduce the cost of GPU access, the government is providing a 40% subsidy for researchers and academic institutions, lowering the price to under Rs 100 per GPU hour. This is significantly cheaper than the global average of $2.5–$3 (around Rs 200–240), while the standard compute power rate stands at Rs 115.85 per GPU hour.
Vaishnaw is confident that India can build its own AI models. He said that at least six major developers could create these models within six to eight months, or possibly as fast as four to six months if things go well. Per rough estimations, the country is home to about 6,700 AI startups, so there is a wide selection of indigenous teams who might participate in the new initiative. “We have great software capability, we have a good innovation ecosystem. AI models will be distributed, that’s for sure. What would be important is how we use AI,” said the minister.
The minister emphasised the importance of “distillation” in AI development and confirmed that India’s AI models will remain open and application-driven. As part of the IndiaAI mission, an application development program is already underway. The country selected 18 projects focused on key areas such as agriculture, climate change, and learning disabilities to drive practical AI solutions.
As for the existing popular AI models, Vaishnaw addressed public privacy concerns with an announcement that China’s DeepSeek AI model will soon be hosted on Indian servers, just like Llama. This will give India more control over data and help improve data privacy.
