India AI stack: Why Sam Altman's visit to India matters

Just when an artificial intelligence (AI) war is raging, Microsoft-backed OpenAI's chief Sam Altman is visiting India. Days after China's low-cost DeepSeek artificial intelligence tool rattled the tech world a week ago by surpassing ChatGPT in popularity on Apple's App Store in the US, OpenAI launched o3-mini, the most cost-efficient of its reasoning agents, which “advances the boundaries of what small models can achieve” as well as a 'Deep Research' tool with which ChatGPT can now carry out complex and multi-step research by parsing large amounts of online data. This is Altman’s first visit after 2023 when he met with Prime Minister Narendra Modi. Altman's visit to India, along with South Korea, Japan, the UAE, Germany and France, amid the turmoil in the AI world indicates OpenAI's plan to partner with a large market where AI has yet to take root.

Altman's visit comes at a time when India is upping the ante on its AI play with the launch of the IndiaAI Mission. “Altman’s visit is to push OpenAI’s case for becoming a part of India’s national AI agenda. India was shown to be a laggard in the AI race after China's DeepSeek shook the world with its low-cost model while OpenAI had to contend with a much smaller rival. In India, OpenAI is facing a copyright lawsuit from a group of book publishers and media companies.

India is the second-largest market for OpenAI, Altman said in New Delhi on Wednesday, adding that the company tripled its number of users here in the past year.

Union minister for electronics and IT Ashwini Vaishnaw today met Altman to discuss India’s strategy for developing a comprehensive AI ecosystem. Following the meeting, Vaishnaw posted on X: "Had super cool discussion with Sam Altman on our strategy of creating the entire AI stack – GPUs, model, and apps. Willing to collaborate with India on all three."

During the discussion, Vaishnaw emphasised India’s ability to develop cost-effective AI models, drawing comparisons to the country’s low-cost space missions. "Our country sent mission to the mood at a fraction of the cost, that many other countries dis right? why cant we do a model which will be a fraction of the cost, that many other countries do. So yes Innovation will bring in that cost down within application in healthcare and education.in agriculture, in weather forecasting, in disaster management, transport multiple different things we are working on," he said.

Altman backs India's quest for foundational AI models

During his meeting with Vaishnaw, Altman also clarified that his remarks at The Economic Times Conversations in 2023, where he stated that Indian startups may struggle to develop foundational models due to high computing costs and that competing with OpenAI was "hopeless," were taken out of context.

Altman explained: “That was a very specific time when there was a certain scaling thing where I thought and I still think, just staying on that frontier of creature models is expensive. But one of the most exciting things that's happened to me since the bad things happen in the industry. Since we're now in a world where we made incredible drugs with distillation, we learned a lot about doing small models, and this reasoning in models, in particular, can be that it's not cheap. It's still expensive to train them, but it's doable, and I think that's going to lead to an explosion of really great creativity. And you know, India should be the leader there.

"There are two different ways you can look at the costs of models, to stay at the frontier. We believe those costs will continue to rise on this exponential curve, but also the returns to increase the intelligence are exponential in terms of the economic value of the scientific value that you would create.

"So, we're doing this big Stargate project, and that's going to go like this. On the other side of it, the cost for a given unit of intelligence one year later seems to fall by about 10x. Was a 2X every 18 months for the number of transitions on a chip and that changed the world if you waited a few decades.

"But, what's happening with the reduction in cost in AI models, is extraordinary. Now, I don't know what it means that the world's gonna need any less AI hardware because you bring the cost down and just the people are going to use it for a lot more things. The total number of dollars will go up, but that's a really exciting thing happening.”

Can India do in AI what China has done?

Despite India traditionally being home to a large tech talent, China's AI advances with the release of DeepSeek have raised a question over India's technological capabilities when innovation is happening at a lightening speed. Building its own cutting-edge AI technology is critical for India not just for social, economic and governance aspects but it is also needed for strategic reasons.

India has been divided into two factions over the past year. One group advocates building indigenous foundational models from scratch, while the other supports the development of smaller language models with fewer parameters, focusing on specific applications. India is now setting aside scepticism about its capacity to develop its own foundational models.

India will offer the cheapest compute in the world at less than $1 per hour for high-end chips that power generative artificial intelligence (GenAI) as part of the government's Rs 10,000 crore IndiaAI Mission which came into play last week.

The government will also incentivise the development of local language models built by academia and industry with investment capital and other support, Union minister Vaishnaw has said. The move is aimed at building up Indian language foundational model muscle. Proposals for model development will be invited soon. At least six startups and developers that can do it within the next 10 months have been identified, he said. ET was the first to report on January 23 that India will back indigenous foundational models.

“The real value will come from two things--algorithmic efficiency and the quality of training datasets,” Vaishnaw said, adding that Chinese AI company DeepSeek has proven to the world that a cost-efficient model can be developed.

“DeepSeek was trained on 2000 GPUs (graphics processing units),” Vaishnaw said. “We have now 15,000 high-end GPUs. (OpenAI’s) ChatGPT version 1 was trained on about 25,000 GPUs. So this gives us a huge compute facility, something which will really give a boost to our ecosystem.” GPUs are high-capacity chips needed to run complex AI development tasks. Vaishnaw said India is not late to the AI party and will play a big role in the innovation taking place in the field globally. Since India has now incentivised compute, the models will follow, he said.

When India has shun the hesitancy to develop its own full AI stack, which would include not just GPUs and apps but also foundational models, a collaboration with Altman's OpenAI, the pioneer in the field, can help the country achieve its goal faster.

(With inputs from TOI)