India eyes the AI pie

India’s space mission Chandrayaan-3 that landed on the moon in August 2023 grabbed global attention for cost-effectiveness. It was built and launched at a budget of around Rs 615 crore, a fraction of a cost incurred by other countries like the US, Russia and China. Russia’s unsuccessful mission Luna-25, which crashed trying to soft-land on moon’s South Pole, just two weeks before India’s successful mission, cost Rs 16,000 crore.

In January 2025, an artificial intelligence model DeepSeek-R1, developed by a Chinese firm, took the technology industry by storm for its efficiency and low cost. A little-known Chinese startup has demonstrated that cutting-edge AI models can be developed with limited computing resources. The US-based OpenAI has used around 25,000 Graphics Processing Units (GPUs) for the development of ChatGPT while the DeepSeek-R1 model has been developed with just around 2,000 GPUs. The GPU is a specialised chip that plays a critical role in the development of AI models.

Can India develop indigenous AI models like ChatGPT and DeepSeek at a fraction of their costs as it did with the moon mission? Union IT Minister Ashwini Vaishnaw looks confident and has announced that India would develop its own foundational AI model in the next six to 10 months.

“Our country sent a mission to the moon at a fraction of the cost that many other countries did. Why can’t we do a model (AI) which will be a fraction of the cost that many other countries do,” Vaishnaw said at an OpenAI event.

OpenAI co-founder and poster boy of Artificial Intelligence, Sam Altman, this week visited New Delhi to discuss collaborations. After having a closed-door meeting, Vaishnaw said that Altman had showed eagerness to work with India to build cost-effective AI solutions.

India is the second-biggest market for OpenAI. According to Altman, OpenAI’s user base has tripled in India in the last one year.

Jaspreet Bindra, Co-founder and CEO, AI&Beyond, said that the cost of intelligence had declined sharply in the past one year. Considering the huge pool of talent and history of frugal engineering, India can build its own AI model at cheaper cost, he said.

“Had you asked me this question six months back, the answer would have been very different. The conventional wisdom or the narrative was that it takes billions of dollars to build AI,” said Bindra, adding that the narrative had changed in recent months, especially after the release of the DeepSeek-R1 model.

Bindra further added that it’s not enough to develop an AI model but there is a need to have a comprehensive approach. “You need to do a model, you need to have data sets for that, you need to have applications for that, and you need to have infrastructure,” he said.

“We need to make sure that if we go ahead and build large-language models, which are indigenous to India, we need to focus on localisation properly. It has to be multilingual. Otherwise, people will not be able to apply it into different kinds of issues that they face at the ground level,” said Ashish Santhalia, Co-founder of Convin AI.

In the Union Budget 2025-26, the central government has announced a number of initiatives to advance DeepTech innovation. Allocations for all schemes related to AI have been significantly increased. A new centre of excellence in AI for education is proposed to be set up with an outlay of Rs 500 crore. Rs 2,000 crore has been allocated for IndiaAI Mission, nearly a fifth of the scheme’s total Rs 10,372 crore proposed outlay.

DeepSeek has challenged the pole position occupied by the US in the AI sector. This is likely to be a focus in the Artificial Intelligence Action Summit to be held in Paris on February 10-11. Prime Minister Narendra Modi, who is scheduled to co-chair the summit along with French President Emmanuel Macron, is likely to pitch for forging global collaborations in the sector.