Nilesh Jasani: Snap out of the DeepSeek delusion and invest big in basic research

Hardly any financial analyst globally has changed a single conclusion since DeepSeek revealed process innovations. Instead, justifications have been rolled out in support of previous conclusions.

DeepSeek’s reinforcement of pre-existing biases has also been notable among commentators in India, many of whom still find it prudent to avoid foundational research—no longer because it is too expensive or something we cannot do, or because large players are already far ahead, which was their earlier argument, but because it is too cheap and easy; everyone is doing it, so there’s no need for India to join a crowded race.

Also Read: India must wake up on basic R&D for technology before it gets too late

Effectively, their advice is to sit on the sidelines and watch the rest of the world burn through money and energy on AI models. Then, as the dust settles, we should use our unique data to create applications on top of others’ work to suit our needs. Sounds safe, right?

Wrong. This ‘wait-and-see’ approach a risky bet that could leave India far behind in the global innovation race. Here’s why:

The race won’t slow down—It’s speeding up: Advocates of caution assume that the AI race will eventually stabilize, giving latecomers a chance to benefit from others’ mis-trials, mistakes, and premature wins. But that’s not how it works. AI development is accelerating. Machines are building, testing and refining AI models with minimal human input now. This self-improving loop means that each advancement lays the ground for the next, and the cycle is getting quicker.

Of course, there will be saturation and stagnation in some fields—such as the largest language models—but the overall innovation pace is both increasing and moving in various directions, with the world’s best capital, talent and machines behind them.

Also Read: What America’s technology denial and China’s AI success imply for India

To cite one example, just a year ago, AI models relied heavily on human-curated data and manual fine-tuning. Now, models like DeepSeek are generating their own training data, optimizing their architectures on the fly and outperforming the most advanced systems from giants like OpenAI and Google.

If we opt out now, waiting for some mythical stability, we’ll find that the rest of the world has raced so far ahead that catching up will be nearly impossible. While focusing on applications tailored to India’s unique data and needs is important, it’s not enough. It’s like building houses while relying on external suppliers for key raw materials, architectural designs and foundations.

Lessons of the internet era—Build, don’t just use: In the previous internet era, India didn’t just sit back and build applications for local use. We became a global powerhouse in software development, offering services and products worldwide. This not only boosted our economy, but also gave us the financial strength to purchase what we could not build.

Also Read: Rahul Jacob: DeepSeek’s big AI shake-up holds policy lessons for India

The same principle applies to AI. Yes, applications that cater to India’s unique needs are important, but stopping there is shortsighted. To benefit from the AI revolution, we need to create products and technologies that the world wants. Given the nation’s AI talent, it is possible that India will emerge as a key provider of efficient and cheap technologies and solutions.

Foundational research holds the key to future innovations: This isn’t just about building better chatbots or models that remain the most cutting-edge only for a short time. It’s the cornerstone of future innovations in fields like biotechnology, autonomous vehicles and even space exploration.

Soon,we may see DeepSeek moments in other fields; say, in protein folding, where some new entrant might come up with better and cheaper ways of identifying protein structures than Google’s Alphafold. This is possible, as DeepSeek has shown, across domains. The only thing that is given is that those who do not try from the ground up have no chance of coming up with such solutions.

Also Read: DeepSeek’s breakthrough is a pivotal moment for the democratization of AI

This is why US-based robotics companies like Figure AI are moving away from relying on models like OpenAI’s and developing their own foundational models in this field. They are reportedly on the brink of breakthroughs that could substantially alter robotics.

Imagine if Indian tech companies partnered with pharmaceutical giants to develop AI models designed for drug discovery. We could lead the world in protein folding research or DNA analysis that would let our companies pioneer new treatments.

If we opt out now, waiting for some mythical stability, we’ll find that the rest of the world has raced so far ahead that catching up will be nearly impossible.

But that won’t happen if we’re only focused on building apps for local use. If we don’t invest in foundational research, someone else will—and we will be left playing catch-up in more and more sectors where we have a chance to lead.

Don’t let fear hold us back: Yes, AI development is fast, chaotic, and full of risks of failures and losses. But opting out of foundational research isn’t a safe bet. Rather, it’s a surefire way to fall behind. We can’t afford to let fear of rapid change or short-term investment losses keep us from participating in the unfolding innovation era. Some moves have been made, but we must join the race in earnest.

The author is a Singapore-based innovation investor for LC GenInnov Fund.