Daybreak - The AI running India isn't Indian. Can that still change?

Episode Date: November 20, 2025

India is using more AI than ever. But most of that intelligence is not Indian. OpenAI, Google and others are expanding in India fast. They already shape how millions work, learn, and search. Meanwhile, India’s own sovereign AI model is only expected in 2026. Other countries like South Korea and China have already built and deployed theirs. What does sovereign AI actually mean, why does it matter for everyday users and why is India is still struggling to build the full stack. And most importantly, who will build the AI that runs India’s future?Daybreak is produced from the newsroom of The Ken, India’s first subscriber-only business news platform. Subscribe for more exclusive, deeply-reported, and analytical business stories.

Hi, this is Rohan Dharma Kumar. If you've heard any of the Ken's podcasts, you've probably heard me, my interruptions, my analogies, and my contrarian takes on most topics. And you might rightly be wondering why am I interrupting this episode too. It's for a special announcement. For the last few months, I and Sita Raman Ganeshan, my colleague and the Ken's deputy editor, have been working on an ambitious new podcast. It's called Intermission.

We want to tell the secret sauce stories of India's greatest companies. Stories of how they were born, how they fought to survive, how they build their organizations and culture, how they manage to innovate and thrive over decades, and most importantly, how they're poised today. To do that, Sita and I have been reading books, poring over reports, going through financial statements, digging up archives, and talking to dozens of people. And if that wasn't enough, we also decided to throw in video into the mix. Yes, you heard that right. Intermission has also had to find its footing in the world of multi-camera shoots in professional studios, laborious editing and extensive post-production.

Sita and I are still reeling from the intensity of our first studio recording. Intermission launches on March 23rd. To get an alert as soon as we release our first episode, please follow intermission on Spotify and Apple Podcasts or subscribe to the Ken's YouTube channel. You can find all of the links at the ken.com slash I am. With that, back to your episode. Here's the uncomfortable truth about India's AI story. We are running faster than ever before, but on somebody else's track.

Everywhere you look, FinTech, Educating. logistics, AI is suddenly the engine. But the engine itself, it's rarely Indian. At first, you might think that sounds impossible, because after all, this is a country that built UPI, it ships software to half of the world and produces some of the most sought-after AI talent globally. And yet, if you look a little bit closer at the past year

and a different story emerges, one where India is using more AI talent globally, than ever before, but building surprisingly little of the machinery underneath. Consider this. Open AI is actively expanding in India, not softly, not slowly, but with clear strategic intent. It is launching India-specific plans like Chad GPD go at $399 a month and tapping into what executives describe as the world's second biggest AI market. Google has announced plans for a $15 billion dollar AI hub in the country over five years. And many US model companies like OpenAI, Anthropic,

perplexity are setting up partnerships and teams here to absorb Indian users and Indian data and to make India a distribution base for their services. Meanwhile, the Indian government says that the country's first sovereign foundational model trained on Indian data sets for Indian use cases will be ready by February 26. It is an ambitious goal, yes, but it is also uncomfortably late. Because in the time that it will take India to launch its first big model, other countries,

with far smaller populations and far fewer engineers have already put sovereign AI systems into the field. South Korea did it last year. China has been doing it for years. And they're not talking about sovereign AI. They are shipping it. So the question today is simple. Why has India, a country with all the ingredients, not yet managed a foothold in sovereign AI? And what exactly have South Korea and China figured out that we haven't? Welcome to Daybreak, a business podcast from the Ken.

I'm your host, Nick Da Sharma, and I don't chase the news cycle. Instead, every day of the week, my colleague, Rachel Vargheese and I will come to you with one business story that is worth understanding. understanding and worth your time. Today is Friday, the 21st of November. You see, sovereign AI is not a buzzword. It is a control problem. And it asks a very simple question. Who should decide what the AI in your country knows, how it behaves, and whose interests it serves? And to answer that, it helps to understand what countries are trying to protect themselves from. A rest of the world article laid this out really clearly. It said, even China, which is arguably the world's most protectionist

tech environment, relies heavily on US foundational models because training frontier AI is so compute and data intensive and because American firms still dominate the highest performing models. Chinese researchers describe this dependency as a strategic vulnerability. And their biggest concern, if the US decides to tighten export controls, cap API access, or restrict foundational model licensing, Chinese industries, from finance to health to manufacturing, could be left scrambling. That is the heart of sovereign AI, reducing that vulnerability. So for a country to claim sovereignty, it needs three intertwined layers. First, compute and infrastructure, the data center.

chips and power supply required to train and serve large models. India still lives in a world where industry leaders acknowledge the country is predominantly software-driven and hardware must catch up if it wants to play in this league. Second is foundational models that are not just copies of global ones but are truly built for local languages and cultural nuances. China has dozens of domestic LLMs. but even those are trained using architecture patterns set by US labs, which is a subtle form of dependency baked right into the technical layer.

And the third is large-scale national use cases, clear use cases. Korea's HyperClova X-Think isn't just another LLM. It directly powers Korean commerce, search, logistics and content platforms, which in turn ensures its improvement loop stays domestic. And this is the recipe. Compute, models and use cases. You cannot skip one and still call the system sovereign. For more on what Korea and China have built so far, stay tuned.

Let's start with South Korea because it is a story that is surprisingly instructive. In June 2025, Nevo launched Hyperclova X-Think, a Korean foundation. model that not only understands Korean better than any foreign model, but outperforms domestic and regional competitors on Korea's core benchmarks. Its Cobalt 700 score is 48.9, which is far ahead of its rivals, which are between 32 to 33. But what makes Korea's approach work is not just the model, it is the buildout underneath it. Through partnerships with Nvidia and domestic industrial giants, Korea is deploying roughly 260,000 new GPUs, taking total national compute to nearly 300,000 units in two years.

That is not ambition, that is capacity. China's case, on the other hand, is more complex. Beijing champions its sovereign AI stack. Domestically produced models proliferate, its tech giants deliver impressive demos. But in the article by rest of the world, that, I mentioned earlier, a quieter reality is revealed. Many Chinese companies still license or depend on U.S. Frontier models for high-value tasks because domestic models lags several years behind. Chinese researchers point towards two constraints. One is insufficient access to high-end chips

and second, the enormous cost of training frontier scale models. Because of U.S. chip restrictions, China runs many models on older, less-capable GPUs, which in turn limits performance and over time widens the gap. The takeaway is sobering. Even with scale, funding and industrial alignment, sovereign AI is hard. China has the strongest industrial machinery in the world and it still lacks full autonomy. Korea has speed and focus, but not the global reach. And yet, Both of them are far ahead of India. So why can India still not get a foot in the sovereign AI door? Stay tuned to find out.

See, India has the ambition. It has the talent and it has the user base. And it also has the political will. But it lacks something painfully simple, which is the full stack. Let's start with compute. The government has given. committed to increasing GPU availability for Indian companies, and it is working on domestic data centers. But as multiple analysis have pointed out, India's hardware ecosystem is nowhere

near what sovereign AI demands. Even training a mid-sized LLM requires tens of thousands of high-performance GPUs. And as of now, India does not have them. And this is the compute paradox that analysts describe. India is a global AI consumer, but not an AI training hub. Then there is also the issue of foundational models. The government's target of delivering a sovereign model by February 26 is significant, but 18 months later than Korea and nearly five years after China's first wave of domestic models. India's Bharat-GPT initiative supporting 22 Indian languages is promising, yes, but still operates at a fraction of the scale required to compete with global models. And finally, there is the market dynamic.

The vacuum created by India's slow foundational model development has been filled rapidly by foreign AI companies. Open AI and Anthropic are expanding. Google is committing billions. Global models already dominate Indian use cases. Experts warn that when foreign models become too entrenched, domestic ecosystems find it almost impossible to catch up, which is a phenomenon that China is now experiencing

where even its own companies quietly prefer US models for high-value tasks because they are simply better. India risks the same trap, being a world-leading market for AI, but without being a world-leading builder of it. And when that happens, sovereignty becomes a slogan, not an outcome. India does have the right narrative, the right talent, the right digital infrastructure, but sovereign AI is not a narrative, it is a capacity.

And building it requires the kind of long-term industrial-scale investment that Korea understood early on and China also has been forced into. So the real question for India is whether it can build fast enough before foreign models become the default operating system for everything that Indians do. online. Because once that happens, the country's AI future may be built for India, but not by India. And the gap between those two is the difference between sovereignty and dependence. Daybreak is produced from the newsroom of the Ken, India's first subscriber-focused business news platform. What you're listening to is just a small sample of a subscriber-only offerings

and a full subscription offers daily, long-form feature stories, newsletters and a whole bunch of premium podcasts. head to the ken.com and click on the red subscribe button on the top of the website. Today's episode was hosted and produced by my colleague, Niktha Sharma and edited by Rajiv Sien.