We'll win by understanding our unique problems more deeply than anyone else, by leveraging our messy, beautiful, irreplaceable data, and by building solutions that actually work for real people dealing with real challenges.

My father tells this story about the 1970s in Mumbai. Back then, owning a television was the ultimate status symbol. Every family that could afford it wanted a Crown, a Weston, or a Dynora sitting proudly in their living room.

Here's the funny thing: today, nobody even remembers Crown, Weston, or Dynora existed. Those brands vanished into history. But everyone remembers the shows they watched, the news channels they followed, the content that hooked them.

The TV manufacturers? They made decent money. The media houses that figured out what to beam through those screens? They built empires.

I'm incredibly bullish on India's AI future. But not because I think we're going to create the next best GPT model.

Look, there are some heroic efforts happening. The IndiaAI Mission is building public compute infrastructure. Sarvam is working on sovereign LLMs. Krutrim is pushing up the stack. I genuinely hope they succeed, and I'm cheering them on.

But let me be blunt: training frontier AI models is brutally expensive. We're talking hundreds of billions of dollars and massive energy capacity being poured into these efforts in the US and China right now.

And honestly? We're late to this particular party. I don't see India developing an industry leading model that beats ChatGPT, DeepSeek, or Grok -- at least not for the next few years. So what do we do?

A friend of mine spent three months last year working with a logistics company in Pune. This place had been around for forty years. Their systems? An absolute disaster.

Excel sheets everywhere, no rhyme or reason to the organization. Delivery notes scribbled by hand and stuffed into cardboard boxes. The founder kept apologizing for how chaotic everything was, clearly embarrassed.

Then my friend did something interesting. He started training an AI model on all that "messy" data. Suddenly, the chaos transformed into gold.

Think about what four decades of operations actually means: delivery patterns across different seasons, customer quirks documented over years, driver routes that evolved organically, even the shortcuts guys discovered during Mumbai's brutal monsoons. No flashy logistics startup had this kind of information. No Silicon Valley algorithm could replicate it because they simply don't have access to these India-specific datasets that never made it online.

This is the dirty secret about AI that I explore throughout my book: the models are converging.

Try it yourself. Ask ChatGPT and Claude the exact same question. You'll get nearly identical answers. The base intelligence is becoming commoditised.

But feed these models your proprietary data -- the stuff that exists only in your filing cabinets, your Excel sheets, your employees' heads -- and suddenly you've built something nobody else can replicate. You've created a moat.

And here's what blows my mind: India Inc is sitting on mountains of this gold, and most of them don't even realise it.

Banking transaction histories spanning decades. Hospital patient journeys with uniquely Indian health challenges. Warehouse inventory patterns that account for regional festivals and customs. Insurance claim notes written in a mix of English and local languages, capturing nuances that matter.

This isn't just data. This is context that no amount of internet scraping can ever capture.

The biggest mistake I see -- and I write about this extensively in the book -- is companies that start with "We should do something with AI" instead of "Our customers are bleeding from this specific problem."

Think about Google's origin story. Larry Page and Sergey Brin didn't wake up one morning and say, "Let's build a search algorithm." They were frustrated grad students at Stanford who found searching the internet impossibly difficult for their research. So they built a better search engine as a hobby project in their dorm room, initially just for themselves and their colleagues.

Or take Zerodha, India's largest stock broker. Nithin Kamath didn't start with "Let's use technology to disrupt finance." He'd spent a decade as a trader, getting increasingly frustrated with existing brokers. They charged percentage-based fees that could run into thousands of rupees per trade. Their platforms were clunky. Everything felt opaque and designed to confuse. So he and his brother Nikhil built Zerodha to solve their own pain points as active traders.

The pattern is always the same: start with the pain, not the technology.

India has no shortage of problems begging for solutions. We're facing severe water stress and extreme heat. Our primary education system reaches millions of kids, but quality varies wildly. Justice moves at a glacial pace.

Let me give you a specific example that hit me hard recently.

I visited Tata Memorial Hospital in Mumbai, and it opened my eyes to the scale of our healthcare challenge. India diagnoses 1.4 million new cancer cases every single year. That's the highest in the world.

Now, here's where it gets interesting. We have 800 million ABHA IDs -- that's Ayushman Bharat Health Accounts, basically a unique health identifier for every Indian -- connected through ABDM, our national digital health infrastructure that links hospitals, doctors, and patients.

Imagine what's possible: AI that dramatically reduces time-to-diagnosis by spotting subtle patterns in medical scans. Treatment protocols that adjust to Indian genetic profiles and dietary habits. Insurance claims that get processed in minutes instead of months. Medical documentation available in Kannada, Tamil, Bengali -- not just English.

This isn't about building the smartest AI model in the world. It's about building the most useful one for real people with real tumours.

And here's the kicker: when you solve India's cancer problem well, you're actually training AI models on some of the most comprehensive oncology data anywhere on the planet. With 1.4 million cases annually and our unique disease mix, we could end up building vertical specific AI models -- for cancer care, in this case -- that work globally. No other country has datasets this large and diverse.

You'd be solving a local problem and accidentally creating a global product.

For most companies, the winning strategy isn't to compete with OpenAI or Google on model development. It's to compose.

Take a powerful general model (ChatGPT, Claude, whatever). Blend it with industry-specific data. Add Indian language capabilities through Bhashini, the government's language AI platform. Then fine-tune everything with your domain expertise.

Ship AI agents, copilots, and workflows that actually move a meaningful business metric this quarter. Not someday. This quarter.

The glamour is in the slick generative AI demo that wows everyone in the boardroom.

I recommend what I call the "Love 20" approach in my book.

Find twenty customers who will genuinely love your product. Not like it. Not tolerate it. Love it.

Then build painful-detail features specifically for them. Obsess over their workflows. Understand their vocabulary. Fix the things that drive them crazy every single day.

Twenty customers who love you will evangelise your product better than any marketing campaign. They'll tell their peers. They'll tolerate your bugs. They'll give you the feedback that actually matters.

The internet has demolished geographic moats. A team in California can launch a pilot project in Kolkata by Thursday morning. Being "the local player" doesn't automatically protect you anymore -- not in India, not anywhere.

But flip that around: a Bangalore startup can serve customers in Belgium just as easily.

Geography is secondary now. What actually matters?

Your understanding of what's driving customers insane. Your ability to deliver an experience that exceeds their expectations. The unique advantages you build that competitors can't easily replicate.

Get those three things right, and the whole world becomes your playground.

India won't win the AI race by building better foundation models. That ship has sailed, at least for now.

We'll win by understanding our unique problems more deeply than anyone else, by leveraging our messy, beautiful, irreplaceable data, and by building solutions that actually work for real people dealing with real challenges.

The plumbing is ready. The data is waiting in those dusty filing cabinets and ancient Excel sheets. The pain points are staring us in the face every single day. Now we just need to build.