This moment fuelled a frenzy around the world, which suddenly discovered two things: 1) You don’t need piles of cash for good AI; 2) With the right minds, resources, and intent, anyone can develop a good AI model.
This sentiment was perfectly captured by Zoho Founder Sridhar Vembu, who said, “One immediate consequence… is that AI-enabled capabilities cannot be priced at a huge premium… The pace of innovation in every industry will accelerate as we learn how to use this freely available technology.”
This presented a golden opportunity for India which, from the dawn of the AI era, had pushed for sovereign AI models. According to a survey conducted by venture capitalist firm SenseAI, India dominates 65% of the global AI applications market, yet when it comes to foundational AI infrastructure and early-stage startups, it barely registers at 3%.
India’s missed moment
Infosys was one of the founding donors of OpenAI. Infosys, Elon Musk and other individuals/entities together donated $1 billion. However, due to leadership changes, Infosys missed the opportunity to invest in OpenAI when it restructured into a for-profit entity in 2019.
In its recent Indus Valley Annual Report 2025, Blume Ventures said, “Infosys was one of the founding donors of OpenAI, then structured as a non-profit. Subsequently, due to Vishal Sikka’s exit, there was no link with OpenAI, and it missed the opportunity to be an investor when OpenAI restructured itself and spun off a for-profit entity in 2019.”
Catching up
In DeepSeek’s wake, Union IT Minister Ashwini Vaishnaw was quick to strike while the iron was hot — he said on January 30 that the country will build an indigenous AI model by October or November, using over 18,000 graphical processing units (GPUs) housed in a common compute facility.
“Our focus is on building AI models that maintain the Indian context and culture… India now has a robust compute facility that will support our AI ambitions… With algorithmic efficiency, we can create these models in a much shorter time frame… in just a few months,” Vaishnaw said.
The facility was established under the IndiaAI Mission. Jio Platforms, Tata Communications, Yotta, and NextGen Data Centre are among the GPU providers.
India’s data centre in numbers
| GPU Type | Quantity |
|---|---|
| NVIDIA H100 | 12,896 |
| NVIDIA H200 | 1,480 |
| AMD MI325 | 742 |
| AMD MI325X | 742 |
Source: IndiaAI Mission
In the Union Budget 2025, the government allocated ₹2,000 crore to IndiaAI Mission — a near 300% jump from last year’s allocation.
Money matters
Tech giants like Google, Amazon, and Meta pumped $250 billion in 2024 alone. Meanwhile, Indian AI startups secured around $750 million in funding last year, with Bengaluru claiming roughly 40% of this pie.
And DeepSeek came along and completely threw this viewpoint out of the window, developing its model at a fraction of the cost and on a cluster of 2,048 NVIDIA H800 GPUs — a weaker and less expensive version of the H100 GPU. Depending on the configuration, one H100 GPU costs between $17,500 and $28,000. The H100 costs about 30% more, with each GPU priced at $25,000-40,000. One H200 GPU costs an average of $32,000.
Further, OpenAI is said to use more than 700,000 H100s, which puts just the cost of the compute power in the range of $17.5-28 billion. For context, OpenAI used 25,000 GPUs to train its model.
While this might seem like an expensive proposition, the Indian government has sweetened the pot — Vaishnaw said the government will subsidise 40% of the cost of each GPU, which he said will result in an operational cost of less than ₹100 per hour — global models charge between $2.50-3 per hour.
Frugal Innovation: India’s answer?
As per Blume, India has a history of frugal innovation, exemplified by the success of the Digital Public Infrastructure (DPI) and ISRO’s space missions.
These initiatives have shown that India can achieve significant technological advancements with limited resources, Blume noted, adding that the same approach could be applied to developing foundational AI models, leveraging public-private partnerships and cost-effective strategies. The report draws a parallel, “The success of Digital Public Infra and ISRO / Space Missions are two great examples of frugal innovation. Could AI models be the third?”
India boasts a wealth of talent in the AI and tech sectors, Blume said, with many of its best minds working in leading tech companies globally. A recent NASSCOM-Deloitte report projects that Indian AI talent pool will grow from 6,00,000-6,50,000 engineers to over 1.25 million by 2027.
Conclusion
India’s journey towards developing a foundational AI model is just beginning, but the signs are promising. With the right support and continued focus on innovation, India could soon join the ranks of countries leading the AI revolution. The success of DPI and ISRO provides a blueprint for how India can achieve this, making the dream of a homegrown foundational AI model a likely reality in the near future.
