Learning from DeepSeek, honing India’s AI strategy

What do these developments imply for India’s quest for AI leadership? The discussion in India in the aftermath of the shake-up caused by DeepSeek has been on ensuring the availability of AI chips and accelerating attempts to build a sovereign Large Language Model (LLM). The government has announced that India will be developing its own LLM within the next few months and that the National AI Mission has already made over 10,000 GPUs available to startups and researchers. India AI Mission also put out a call for proposals to build foundational AI models, including LLMs and Small-Language Models (SLMs).

While the quick movement on this front is laudable, these are not the only — or even the most important — action items to prioritise post DeepSeek’s success. Sure enough, DeepSeek has demonstrated its ability as a competitor to ChatGPT and others at a much lower training cost than its American counterparts. This is an encouraging development for a country like India that champions frugal innovation.

But it doesn’t mean that the first, or only, logical next step for India is to build its own LLM. DeepSeek’s main contribution is that it has come up with a differentiated approach to training an LLM vis-a-vis its US and European competitors. The lower training cost is a result of that research-backed innovative approach.

There are three major learnings for India’s tech leaders and policymakers here. First, India should prioritise putting in place all the building blocks necessary to innovate in AI. For that, plugging some fundamental gaps in India’s AI ecosystem — top-tier AI talent, unique datasets and advanced research and development (R&D) — will be critical. India today doesn’t have the top-tier AI talent that the US or China boast of. Most Indian-origin talent in AI is unfortunately still working in Silicon Valley. Aravind Srinivas of Perplexity AI is happy to contribute funds to those working in AI in India, but not willing to move back to India. That’s the logical decision for him given Perplexity AI’s leading position in the US. But it makes clear that India will need to find ways to plug this AI brain drain.

India also needs to build a clear strategy to leverage its data-rich tech ecosystem. While India’s tech platforms, including UPI, are generating large volumes of India-specific data, Indian AI startups have not been able to build atop these datasets yet. Many other such datasets unique to India remain locked within silos. India will need to find ways to unlock this for domestic AI startups to build upon.

Similarly, India’s AI R&D ecosystem will need a major boost. During PM Modi’s US visit, a key partnership was announced between India’s recently launched Anusandhan National Research Foundation (ANRF) and the US’ National Science Foundation. But India also needs to boost private and public funding of its AI research ecosystem domestically, including its recently established AI Centres of Excellence, to raise the likelihood of truly innovative AI developments (like DeepSeek) coming out of India.

India must focus on first putting together these key building blocks to achieve true innovations in AI. The outcome, then, of putting these building blocks in place could be in the form of a more efficiently trained LLM (like DeepSeek’s R1) or something else altogether that is truly innovative. That will help India truly command respect in the global AI ecosystem.

Second, India must put its weight behind open source innovation in AI. There is a big debate raging globally about open source/open weight models versus closed source/closed weight models. Open source LLMs, like DeepSeek’s R1 model, Mistral, and others, are currently locked in a battle with closed source LLMs such as ChatGPT.

Many companies, notably France’s Mistral, the US’ Meta and now, China’s DeepSeek, have put their weight behind open source. India must also. Open source advantages Indian startups and researchers attempting to compete in AI, whereas closed source AI ecosystems will further deepen India’s dependence on foreign AI systems. India will find common cause with Europe and other countries in the Global South on this, as they would similarly benefit from an open source AI ecosystem.

Third, India must shift its focus much more to building AI competitiveness for now, rather than being overly focused on shaping global AI safety rules. At the Paris AI Summit, Vance explicitly outlined the US’ worldview on AI: In the high-stakes competition with China for AI leadership, it simply wants to win. In such a highly charged global race for AI leadership, India must shift its focus to building a competitive niche in the AI race. This is not to say that India shouldn’t continue pursuing global compacts on AI safety and building domestic guardrails; it should, but without losing focus on building national competitiveness in AI.

Indian policymakers and tech leaders must realise that the US-China race over AI leadership will leave us behind if we don’t understand the fast-evolving contours of this game. Europe is also learning this lesson the hard way. To be relevant in this game, India will urgently need to devise the right strategy to boost its AI talent and AI R&D centres of excellence and leverage its unique datasets to contribute to the fast-paced AI innovation ecosystem. Only then will the world sit up and take notice of India in The Great AI Game.

Anirudh Suri is managing director, India Internet Fund, a non-resident scholar at Carnegie India and author of The Great Tech Game. The views expressed are personal