Vaishnaw outlines India’s five-layer AI stack; says 38,000 GPUs opened up as ‘public good’

New Delhi: Union Minister Ashwini Vaishnaw on Thursday laid out India’s AI roadmap at the India AI Impact Summit, framing it around five layers of the AI stack, applications, models, compute, infrastructure and energy, and pitching “democratising technology” as the organising principle behind the approach.

Addressing delegates with Prime Minister Narendra Modi in attendance, Vaishnaw described the event as the “first AI summit in the Global South” and “the biggest AI summit so far”, and said it had participation from 118 countries.

Calling AI a “foundational technology”, he said it is already changing how people work, learn and make decisions. 

He said Prime Minister Narendra Modi believes the true value of technology lies in ensuring benefits reach the masses, and that India’s approach is to deploy AI at scale and make it accessible to all.

Vaishnaw said the first layer of India’s AI stack is the application layer, which he described as the source of real returns and mass impact. He said India is building real-world solutions across sectors including healthcare, agriculture, education, logistics, design and financial services.

The second layer, he said, is the model layer, which he linked to sovereignty. While acknowledging that very large models will continue to push the frontier, Vaishnaw said India believes over 90 per cent of use-cases can be served through smaller, focused models. 

He said specialised models can deliver specific use-cases at lower cost, including lower cost per token and overall access costs.

Vaishnaw said India has launched a “bouquet of sovereign models”, many of them during the summit, with a focus on multimodal and multilingual capabilities. 

He cited India’s cultural and linguistic diversity as the reason for that focus. He also said these sovereign models have “stood the ground” against frontier models on points that matter most to India.

On the third layer, compute, Vaishnaw said India is treating compute as a public good. He said a public-private common compute platform has been created to provide access to 38,000 GPUs at an “affordable rate” for startups, academia, researchers and students. He added that another 20,000 GPUs will be added to the platform.

Vaishnaw said the fourth layer is infrastructure, anchored in India’s talent pool and what he called trusted policies. 

He pointed to a recent policy shift announced by the Prime Minister that seeks to bring the world’s data to India to reside here, be processed here, and support the delivery of high-value services globally. 

He said the government expects large investments in data centres in the coming months.

The fifth layer, he said, is energy, while linking it to the government’s push on clean energy and said more than 50 per cent of India’s power generation capacity now comes from clean, renewable sources. 

He also cited recent reforms in the nuclear energy sector, describing it as a way to provide a baseload of clean power. He said the national grid has been “practically rebuilt” over the last decade.

Vaishnaw also referenced the Stanford University AI Index, saying it places India among leading countries in AI adoption, AI talent and AI diffusion. 

At the same time, he said the government is conscious of challenges facing the IT industry and is working with industry and academia to upskill and reskill, and to build a new talent pipeline for what he called the “new intelligence age”.

He said the gains from AI will need to be balanced with collective solutions to mitigate risks, and argued that human safety and dignity must remain at the heart of AI adoption.

“We must also find collective solutions for mitigating the risks by placing human safety and dignity at the heart of AI,” he said, urging participants to shape an AI future “by the humans and for the humans”.