India's foundation artificial intelligence model in 8-10 months: Vaishnaw

The foundational models are being developed under the Rs 10,372-crore India AI mission, which was approved by the Union Cabinet in March 2024

Against the backdrop of Chinese startup DeepSeek recently stunning the world with its artificial intelligence (AI) model, Union Minister for Information Technology Ashwini Vaishnaw on Thursday showcased India’s ambition in the AI space. India will have multiple sovereign foundation AI models developed and ready for deployment over the next eight to 10 months, Vaishnaw said. 

Promising to put India on the AI map, he told the media: “We believe there are at least six developers who will be able to create foundation models in the next eight to 10 months on the outer limit and four to six months on an aggressive, optimistic estimate.” 

The Ministry of Electronics and Information Technology (MeitY) has been working with experts in the field of large language model (LLM) and small language model (SLM) over the past 18 months to develop the framework for India’s foundation AI model, the minister said.   

India’s AI mission, approved by the Union Cabinet in March 2024, had sought to procure at least 10,000 graphics processing units (GPUs) for the country. Almost a year later, the government has empanelled 10 companies to procure and supply 18,693 GPUs across a range of latest versions such as the H100, H200, MI325, and MI300X. Reliance Industries-owned Jio Platforms, Tata Communications, and the Hiranandani Group-owned Yotta Data Services are among the key players that will procure and supply GPUs to startups, academics, researchers, and other users in the country. CMS Computers India, Ctrls Datacenters, E2E Networks, and Locuz Enterprise Solutions are some of the other companies to qualify.

 

Based on the bids placed by the 10 companies, the average rate per AI compute unit has been discovered at Rs 115.85 per GPU hour for the low-end units and Rs 150 per hour for the high-end computer processing units against a global benchmark of roughly $2.5 to $3 per GPU hour.

 

The foundation models will be able to compete globally by using algorithmic efficiency to cut development costs and deployment time, Vaishnaw, who also holds charge of ministries of railways and information and broadcasting, said at a media interaction. This comes days after China’s DeepSeek revealed to the world that it had built an AI foundation model at a fraction of the cost incurred by global AI leaders, mostly American such as OpenAI, the maker of ChatGPT.    

 

Addressing concerns linked to DeepSeek sending user data to China, the minister said the government was studying the open-source codes and technical reports of the foundation model. DeepSeek would be hosted on Indian servers, he added without elaborating.

 

The companies selected to develop the sovereign foundation models will be evaluated on the capability of their team, the approach and strategy they take, the use-case they work on, their ethical considerations, scalability and sustainability of the models for use on India-specific problems, the timelines and milestones they promise and the financial viability of the foundation model.

 

The government has outlined a two-dimensional approach to fund the companies, which will be selected to develop these foundation models. Under the direct funding approach, there will be central funding to the companies over and above any other financial support from states. The government has also planned a second equity-based funding for building these models. Private investors will be allowed to take part in this investment process.

 

“Overall, we have nearly 15,000 high-end GPUs. Just to give you some context, DeepSeek was trained on 2000 GPUs while the Chat GPT version 4 was trained on 25,000 GPUs. This (procurement of 18,693 GPUs) gives us a very robust compute facility, which will give a boost for creating AI applications, models, distillation and training processes and creating new algorithms,” Vaishnaw said.

 

The government will offer a 40 per cent discount on the market rate for startups, researchers, academics, educational institutes and students willing to use GPUs for various purposes. The government is also likely to offer additional discounts to users who sign up for long-term use of six months or a year and above, an official said. These incentives and discounts will be provided for the next four years, the official said.

 

In December 2024, the government had received 19 bids, of which 14 were eligible to procure and supply GPUs under the India AI mission. Of the 14, only 10 companies were technically qualified to procure and supply the GPUs.

 

The government is working parallelly with stakeholders to develop an indigenous GPU, Vaishnaw said. Also, the government has selected 18 applications, which will receive funding for the development of AI tools in the field of agriculture, learning disability, and mitigating climate change, he said.