IT-BT Budget Round Table 2025: India's foundational AI model by end of 2025, says I-T Minister

As the Chinese AI model DeepSeek created ripples globally, Union Minister for Electronics and Information Technology Ashwini Vaishnaw on Tuesday said that India's foundational AI model should be ready by the end of this year and some of the Indian firms have made really powerful models.

Speaking at the India Today-Business Today Budget Round Table 2025, Vaishnaw said the
outer limits for the Indian AI model is 10 months. "It is likely to be smaller than Open AI and others but significantly powerful in terms of features. A much bigger version to be rolled out later also so that our models can compete with the best in the world," he said.

China’s DeepSeek has dented the market valuations of major tech players globally. The Chinese player has been able to build highly advanced model at a cost much less than the US tech giants.  

The government has launched an AI Mission with an allocation of Rs 10,000 crore.

"As we speak, we have empanelled 18000 graphic processing units (GPU) and  10,000 are already available. First thing you need is a compute power and is made available to everyone.
It will be rolled out in couple of days and researchers, colleges, academicians and others will  have access to and work on it," Vaishnaw further said.

On India's competitive prowess in the sector, the minister said that AI has just begun and amount of innovation we will see is phenomenal.

"India ranks among the top 4 in terms of innovation. We are confident that we will be the leaders in this technology. Moreover, in the world of Science and Technology, a base is taken by everyone and built on the next," he explained.

The minister also called upon a change in mindset and the need for India to become a product nation from being a service provider. He announced that the first made-in-India chips will roll out this year.

On the sovereign AI model, Vaishnaw listed three things -- a foundational model with coding at the core, data sets to efficiently train model, and having its own GPU.

"We worked out on all three components of technology. All startups will have full control over the core and a large number data sets which are non personal like weather, traffic and agriculture. Entire sovereign data is humongous and will be used to train our models. For GPU semiconductors, processors and chips work, we are working on multiple options like taking form open source or licence and build on to build own own GPU. That will give own GPU in 3-5 years," he added.