DeepSeek And Ye Shall Find: India’s AI Quest

The company developed its foundation model DeepSeek — V3 using 2,048 GPUs and US$ 5.6 million — which is called a shoestring budget at least in the AI world — in the span of just two months. This is far less compute power, money and time it has taken for companies like OpenAI to create GPT models.

There is a long-held view that when it comes to all things AI, the buck stops with OpenAI, and by extension, the US. The US has been a leader in AI research, talent, infrastructure and GPUs. The numbers and the statistics back this up.

According to the Stanford AI Index Report 2024, the US leads in top AI research publications and citations, ahead of China and Europe. The Tortoise Intelligence Global AI Talent Report 2023 ranks the US as the top country for AI talent concentration, measured by the number of AI researchers and engineers.

Immigration policies such as the O-1 visa for "extraordinary" talent attract top AI researchers from around the world to the US. Moreover, the US dominates AI infrastructure via AMD and NVIDIA, with the latter supplying over 80 per cent of the world’s AI GPUs.

But with DeepSeek, things changed overnight and people starting questioning USA's unofficial claim to the throne. Countries with a considerably smaller AI prowess, like India, started wondering: If China could do it, why not us?

In a bid to catch up with the tech behemoths, Indian Minister of Electronics and Information Technology, Ashwini Vaishnaw, has doubled down on India’s AI ambitions, and announced that India will have its own foundational AI model within 10 months from now.

Experts that The Secretariat spoke to said that it is essential for India to first develop a general-purpose AI model (a foundation model), before customising it for specific industries.

The government did set up a Rs 10,000 crore AI mission in 2024, and has promised to make available 18,000 high-end GPUs for AI development. The government’s aim, as reiterated by Vaishnaw, is to democratise AI access for researchers, startups and academia.

India’s AI Dreams

At a time when DeepSeek has triggered a rethink on how much the world is spending on AI, India is allocating effectively and into the country’s pain points — like the lack of quality AI talent and infrastructure.

“The US has ChatGPT/Claude that are used by 100+ countries. China has DeepSeek, QWEN, Kimik1.5, etc. We don’t have anything,” Ananay Batra, CEO and founder, Listnr AI, told The Secretariat.

He added, “...what we really need is talent that can build on top of foundational models and further improve them, focus on reinforcement learning (an AI method where an agent learns by receiving rewards or penalties for its actions).”

This is not to say that India doesn’t produce quality researchers and scientists. In fact, India has the third highest number of elite AI talent after the US and China. But sadly, a lot of this talent resides outside India.

Back in 2019, almost every Indian AI researcher took off for opportunities abroad. It seems though that India is getting better at holding back its AI talent, because by 2022, a solid one-fifth decided to stick around and work in India.

Further reflecting the government’s seriousness in making strides in the AI space, this year’s Budget announced 10,000 tech fellowships under PMRF and Rs 20,000 crore for public-private partnerships — the latter being a key announcement to drive innovation and collaboration between government bodies and industry leaders.

In 2023, most foundation models (72.5 per cent) came from industry — meaning big tech firms like OpenAI, Google DeepMind, Meta and Anthropic. Academia contributed just 18.8 per cent. Since 2019, the industry's grip on foundation model development has only been getting stronger, noted the Stanford report.

The industry’s increasing control over foundation model creation reflects its vast resources and expertise, positioning it as the primary driver of AI progress. Indian companies like TCS, Yotta Data Services and Reliance Industries have been acquiring chips to strengthen their capabilities.

“With the cash that some of these private tech companies have on hand, our first goal should be to acquire GPUs, try to get as many H100s/H200s as possible. The next step would be to hire the right talent to get people to work with these GPUs. We can’t afford not to pay top dollar at this point to get our talent back,” said Batra.

The Race To Build The Most Intelligent Model

Mr Vaishnaw also announced that a made-in-India chip will roll out this year, in partnership with the Centre for Development of Advanced Computing (C-DAC). Although it’s being viewed as a sound move in the industry, making an AI chip in India will be a hard thing to do. 

“On a scale of 1 to 10, the difficulty of creating our own GPUs is a 10. This is a herculean task that requires US$ 20-30 billion in funding. Building and maintaining a cutting-edge fab can cost tens of billions of dollars,” added Batra.

NVIDIA CEO Jensen Huang said that it will take a country like the US 20 years to be fully self-reliant in making silicon chips. So, we can only imagine how long it would take India to reach the same level of self-sufficiency.

Nevertheless, the government’s ambitious announcement comes at a time when it’s going to get much harder for India to procure advanced AI chips like NVIDIA's and AMD's from the US, due to an executive order passed by former President Joe Biden in January.

That order puts India into a second-tier list comprising over 100 countries that have restricted access to the US’s advanced AI chips. Meaning, India can only order a paricular number of high-end AI chips before hitting US-imposed limits. This restriction places India at a disadvantage compared to nations with unrestricted access, potentially slowing its AI ambitions. 

“This is a significant setback for our AI ambitions, since there is pretty much nothing we can do without GPUs. Getting GPUs via a third country (Singapore, maybe) would be the right step. But we can’t afford not to have GPUs,” said Batra.

When asked whether an Indian company could build its own DeepSeek if it secured 2,000 advanced GPUs, similar to what DeepSeek managed, Batra was clear: That wouldn’t be enough.

"We’d need at least 10,000 GPUs to do something meaningful and build more advanced foundation models. If that does happen, then of course we’ll be able to build something like DeepSeek (or even better),” he said.

Batra explained that India’s objective shouldn't merely be to replicate DeepSeek and create a DeepSeek 2.0. To gain a competitive edge, India must innovate beyond existing models. Otherwise, it risks becoming just another entry in an already crowded field.

Yet, while hardware constraints pose a significant challenge, the broader issue may lie in how India approaches AI development.

Dilip Kumar, an entrepreneur, quipped on LinkedIn: “I'm convinced India will never be able to compete with the US and China in technology if we keep treating it as a spectacle.”

This caught many people’s attention. Kumar was speaking of the Mumbai Tech Week (MTW) 2025, which is organised by the Maharashtra government in collaboration with the Tech Entrepreneurs Association of Mumbai (TEAM). It includes a line-up of celebrities like Karan Johar, Sunil Shetty, Rahul Dravid — figures with little to no connection with tech or AI.

“These are people who haven't written a single line of code in their lives,” said Kumar, which gathered a lot of resonance on social media. “Real AI innovation doesn’t come from celebrity panels — it comes from builders. PhDs, engineers, founders — people who write code, build models and deploy systems at scale.”

India’s ambitions in AI are growing, but the challenge isn’t just about hardware, it’s about vision, strategy and staying ahead of the curve.

“Technology isn’t a spectator sport. If India wants to lead, we must put real builders at the centre of the conversation,” said Kumar.

India has girded its loins to make real waves in AI, but it needs to move past big announcements to focus on actually building things.

The playbook is clear. First, India must lock in access to high-end chips. Second, it needs to pay AI researchers and engineers what they’re worth. Because if it doesn’t, other countries will.

Finally, beyond just building a foundation model, India must create an ecosystem where AI startups, academia, and industry work together to push boundaries and develop cutting-edge applications. 

The industry experts believe that if China can pull off DeepSeek on a budget, there’s no reason India can’t do the same, or better. The world isn’t waiting, and neither should India.