India To Develop Own Foundational AI Model: Ashwini Vaishnaw

In a move to boost India’s artificial intelligence capabilities, Union Minister Ashwini Vaishnaw revealed that the country is set to develop its own foundational AI model.

During a press briefing, Vaishnaw highlighted the progress of the IndiaAI Mission, which has already far surpassed its initial GPU (Graphics Processing Unit) targets. With 18,693 GPUs now available, the initiative aims to provide resources to researchers, startups, and businesses across the country. Notably, 15,000 high-end GPUs have been procured, including 1,480 H200 GPUs.

For context, models like DeepSeek and ChatGPT were trained using 2,000 and 25,000 GPUs, respectively.

Roughly 10,000 GPUs are now available for use starting today, said Vaishnaw, adding that a common compute facility for researchers will be operational within two days. This facility will serve as a vital resource for various AI projects in India.

In addition to strengthening the AI infrastructure, the minister also announced the launch of a new initiative to create India’s own foundational AI model. A framework has been established, calling for proposals from developers to build an AI model that caters specifically to the Indian context, languages, and cultures. The datasets used for training this model will be India-centric, focusing on local needs and challenges.

At least six developers, startups, and teams are expected to start building this model over the next four to 10 months, Vaishnaw said. India's foundational AI model will be able to compete with the world's best, he emphasised, while assuring that data privacy will be a priority throughout the development process.

In another major development, the IndiaAI Mission has announced funding for 18 AI applications across various sectors such as agriculture, learning disabilities, and climate change. These projects were selected for the first round of funding, marking a significant step in leveraging AI for societal benefit.

Additionally, as part of the global AI partnership discussions, the government has approved the establishment of an AI Safety Institution. The institution will be set up on a hub-and-spoke model, bringing together teams from IITs and other universities, and will focus on techno-legal approaches to AI development.

The compute capacity will be made available to companies at a 40% discount to the actual market rate, with the subsidies being funded by the government. These subsidies will run for a period of four years.

Compared to global models that cost $2.5-3 per hour of usage, India’s AI compute capacity will cost less than Rs 100 per hour after the subsidy. Further, attractively priced half yearly and annual plans will make it more affordable.

With cheaper compute capacity available, India will also seek to roll out its own set of foundational models in the next 10 months. "There will be multiple models -- large language models, small language models as well as large multimodal models -- that will be created as part of this program," Vaishnaw said.