|
The Collection Wed, 26 Feb 25 |
Multiple stories, multiple perspectives, one theme worth your time—every week. |
Good Morning [%first_name |Dear Reader%],
You are on a free plan. Your subscription has expired. Upgrade now to unlock premium newsletters, top feature stories, exclusive podcasts, and more.
Just weeks ago, one Chinese artificial-intelligence developer stunned the world with the launch of a revolutionary AI model. Deepseek-R1, massively cheaper and yet outperforming many Western rivals, made some US companies lose hundreds of billions of dollars in market cap in a matter of hours. Reports say Deepseek is now accelerating the launch of its R2, a successor model that will feature enhanced capabilities, in a month or two.
Would you like to guess what the headlines were saying about India’s own AI efforts?
India will build world-class AI model: IT minister
India’s AI mission portal’s launch soon: 10 firms to provide 14,000 GPUs
DeepSeek’s fast rise sparks debate on Indian AI models
India seeks AI breakthrough—but is it falling behind?
Big plans. Good intentions. And doubt.
Whatever you think about its future prospects, it is evident that India is playing catch-up. And by a fair mile; the losses inflicted to rivals by Deepseek might have come overnight, but its win was years in the making.
Why?
What ails, not just AI, but India’s wider deep-tech ecosystem?
The Ken has written extensively about this rapidly changing landscape—you just have to look at the work of my colleagues Abhirami G and Seema Singh. Abhirami’s latest was, in fact, about AI’s four-minute-mile moment, and a largely ignored but crucial component of real-world AI solutions.
Deepseek cracked open AI. India’s AI plumbers are loving it
Until recently, some of the best-funded Indian GenAI companies used the tech to build new apps. Now, the tide is turning in favour of startups that help other companies build with GenAI
This week’s edition of The Collection, though, is about more than just India and AI. It’s about a fundamental problem, of which the country’s AI progress (or lack thereof) is only a symptom.
Just what is wrong with India’s R&D?
Cracks in the bedrock
Besides a few like Sarvam, not many Indian companies have chosen to focus on foundational models. The more popular approach has been to build services on top of AI.
But even if more firms had prioritised foundational models, India was unlikely to have produced a Deepseek. Praveen Gopal Krishnan explained why in The Nutgraf:
The thing that prevents India from building a Deepseek is how research is viewed, and consequently practised here. Socially and professionally, research projects are neither prioritised, funded, nor lauded. Deepseek’s success is considerable, but it’s probably one of the dozens and dozens of AI research projects that are going on in China.
India has chosen to go into use-cases because researchers constantly face the question—“but what’s the point of doing this? What’s the application?”—all the time. So it’s much easier, and tempting professionally, to choose to build something that has some outcome and application than to spend time building something that may not go anywhere.
You should check out his entire piece if you haven’t already.
What China’s cheap AI model tells us about India’s future
Research is a lottery. But is India even playing?
You can lay the blame for this on many things—an absence of culture, lack of investor appetite, outcome obsession, talent bottlenecks, and torpid state funding of research projects and universities are all major contributors.
India’s share in global AI patents stood at just 0.23% in 2022, and yet, we are seeing cuts in public grants and funding for universities.
Universities go the American way as India slowly turns off the funding tap
When the government slashed funding, top Indian universities had to adapt, and they did—by starting their own private (sometimes public) companies
Private funding, too, hasn’t been very forthcoming, and deep-tech startups in general haven’t been very successful at attracting investments. They are seen as high risk.
From one of Seema’s pieces in 2021:
You don’t have to think too much to put US$50 million as part of a unicorn race, but putting US$500,000 as part of a deep-tech seed stage requires a lot of work, says Mahesh Murty. “Most people are ready to invest 100 times more by thinking 100 times less.”
This has ramifications for India, where high-risk angel funding has never been at scale. Even at the US$1–5 million funding stage, says Baskar Subramanian, you are asked if you have a go-to-market strategy; if you can show three customers. Whereas in the US, he says, if you show a proof point which is tangible but not commercial, people know how to go through that process. “This has to change, we are not there yet.”
Do we need a Tiger Global to back India’s deep-tech race, not just the horse?
But this isn’t just about money; there’s something deeper at play here. Something that Shashikanth Suryanarayanan, co-founder of IPO-bound Sedemac—which produces “world-beating tech” for the two-wheeler market—told us in a 2024 interview.
“If you ask, why is there a great cricket team in India, the answer boils down to this—a lot of kids play cricket, and you already have [role models] like a Sachin Tendulkar or a Virat Kohli. If you go to Shivaji Park [in Mumbai], you will get everybody—the next big boy or a five-year-old [playing cricket]. That entire ecosystem is present, and most importantly, a lot of people are trying. So, there’s a culture of cricket.
Now you tell me where is the technology ecosystem and the culture in India? The ecosystem in India doesn’t exist because not enough people are trying their hands at building anything.”
45M two-wheelers in India use his ‘world-beating’ tech. It’s ‘improbable’ and built on a ‘whim’
All of this put together, of course, contributes to a problem that only ends up feeding this vicious cycle—a talent shortage.
“The biggest barrier to [AI adoption] has been talent. The best talent is 5–10X better than your average talent. The gap is significant,” Srikanth Velamakanni, co-founder of Fractal, pointed out in an interview a few years ago. Fractal, a unicorn that offers AI solutions to large companies, is one of the few Indian firms to have found success in the deep-tech space. You can check out our interviews with Velamakanni here:
21 years to unicorndom: Fractal’s build-to-last map for AI treasure
At a run rate of $264 million, Fractal Analytics is eyeing the $15 billion-revenue club of global AI services companies. With acquisitions, incubations, and spinouts in tow, Fractal seems to be embodying its name in the high stakes AI game
First Principles • 17 |
|
But if Deepseek’s success and the pace at which things are changing say anything, it’s that India can’t afford to let companies like Fractal be lonely exceptions anymore.
It needs to hunker down and find a way to vitalise its R&D.
Because now, as Seema put it, “you need R&D to survive. You need your own knowledge base to figure out what to build and where to apply.”
You can find this week’s entire collection below. If you have thoughts you’d like to share, please leave a comment on our stories, or write to [email protected].
Get a premium subscription to The Ken
Unrivaled analysis and powerful stories about businesses from award-winning journalists. Read by 5,00,000+ subscribers globally who want to be prepared for what comes next.
Trusted by 5,00,000+ executives & leaders from the world's most successful organisations & students at top post-graduate campuses
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Rahul leads the promotion and distribution efforts for the journalism produced by The Ken. He works at the intersection of the publication's editorial and business operations.
|
|
Share this edition
You can use the easy share buttons below
|
|
|
|
|
© 2025 The Ken
Do you know anyone else who would like to read this newsletter?
Share this edition with them.