Diffusion matters more than models

For most of my career, whenever I met an EVP or a CXO from a global company, my pitch was fairly consistent: India is not just a market. It is the world's most underutilised business-model test bed. Prototype here. Pressure-test here. Learn here.

For a long time, that sounded like strategy-speak. Today it is simply reality.

The conditions for diffusion

A billion connected people. Some of the cheapest mobile data in the world. Digital public infrastructure that quietly solved identity, payments and data rails before most countries had even framed the problem.

The outcome is fascinating. For the first time in our planet's history, frontier AI is reaching ordinary citizens before many governments have fully wrapped their heads around it. ChatGPT, Gemini, Perplexity and others, free, in the hands of people who were never part of any "early adopter" club.

So I wasn't surprised, but I was genuinely moved, when my auto-rickshaw driver showed me how he uses AI. In his own language. Without a tutorial. Without anyone selling it to him.

Not top-down, not enterprise-first

This is diffusion at a scale we have never witnessed before. Not top-down. Not enterprise-first. Straight into everyday life.

Previous technology waves in India followed a familiar path. Large companies adopted first, consumers followed years later, and rural India later still. AI is moving in the opposite direction. Individuals are experimenting faster than most organisations.

Less obsession with models. More obsession with "so what?"

Invention gets the headlines, diffusion creates the wealth

Economic historians have made this point for a long time. The countries that benefited most from past technologies were not always the ones that invented them. They were the ones that spread them fastest and most widely through their economies.

Electricity is the classic example. The dynamo existed for decades before factories saw real productivity gains. The gains came only when factories were redesigned around electric power, with new layouts, new processes and new skills. The invention was necessary. The diffusion was where the value appeared.

AI is likely to follow the same pattern. A handful of companies and countries will build the frontier models. Many more will decide how much value those models create, depending on how quickly businesses, schools, hospitals and governments learn to use them well.

That is good news for India. We may not lead the race to build the biggest models. We can lead the race to put them to work.

Why diffusion is harder than adoption

Downloading an app is adoption. Changing how a clinic schedules patients, how a farmer plans a season or how a small manufacturer quotes a job is diffusion. The second is much slower and much more valuable.

A few things stand in the way.

• Skills. People need to know what AI is good at, where it fails and how to check its work.
• Trust. Citizens and businesses need confidence that their data is handled responsibly and that answers are reliable enough to act on.
• Integration. Most value sits inside existing workflows and systems, not in a separate chat window.
• Incentives. Organisations need a reason to redesign how they work, not just to try a new tool.

None of these are solved by a better model. They are solved by entrepreneurs, managers, teachers and public servants doing the patient work of redesign.

The questions that matter

Which is why, as we head into the AI Impact Summit, I feel our conversations need a subtle shift.

Less obsession with models, parameters, benchmarks and agent social networks. More obsession with the questions that decide whether any of this matters.

• What real value is being created?
• Who captures it?
• What new business models emerge?
• And most importantly, what does this mean for the median young person in India, and the median human globally?

These are uncomfortable questions for an industry that likes to measure itself in tokens and leaderboard positions. They are also the only ones that will matter in ten years.

What leaders can do now

For business leaders, the practical lesson is to measure AI by outcomes rather than activity. The number of licences bought or pilots launched says little. What matters is cycle time cut, errors avoided, customers served and new revenue earned.

For policymakers, the lesson is to treat diffusion as infrastructure. Training at scale, open datasets in Indian languages, procurement that rewards real outcomes, and rules that build trust without freezing experimentation.

A tech-tury, not a techade

This isn't just another technology cycle. It feels closer to a civilisational inflection point. Not a "techade". A tech-tury.

If India can turn mass adoption into mass value creation, the rest of the world will study it. If it can't, we will have the world's largest population of AI users and very little to show for it. The difference between those two outcomes will not be decided in a research lab. It will be decided in millions of ordinary workplaces.