Every few weeks a new number arrives to tell us that India has won the artificial intelligence race before it has properly begun. That we are the largest market for this X, or the fastest growing for that Y, and the second-largest base of users for nearly every model that matters.
The figures are real. The story made with them is unreal. Markets exist to price things, and the market for Indian AI has the price wrong in both directions at once.
Advertisement – Scroll to continue
The visible side is overpriced. The durable side is underpriced. We are sold a supply side story while the demand side waits in the shade, holding both the real opportunity and the real danger.
Let me start with what is overpriced. According to Sensor Tower’s State of AI 2026 report, ChatGPT counts roughly 330 million monthly users in India and Gemini around 229 million. Those figures get offered as proof of demand. They measure footfall. Appetite shows itself only when there is a price to pay, and the price in India today sits close to zero, because the frontier labs are subsidising access in a land grab for the world’s largest pool of users.
The narrow gap between the two leaders gives the game away. It exists because Android’s installed base does the distribution work that genuine preference does elsewhere. The top of our market has been won by distribution and subsidy wearing the costume of demand.
Early in my career, putting a new dishwash liquid product into Indian kitchens, I watched free samples by the tons flowing off retailers in a week. The number that mattered arrived a month later, when a homemaker reached for it again and paid with her own money. The first number flattered the internal review. The second one paid the salaries.
WhatsApp owns near total reach in this country and has spent a decade in search of the commerce that was meant to follow. Vast usage, thin value. The capital now flooding into AI infrastructure rests on the same headcount story, on a subsidised price that still awaits its first real test. The day a real price appears, we will learn how much of that 330 million was appetite and how much was a free sample.
Turn the picture over, because the underpriced side is the richer one. Value in this technology is born at the point of use, in the hands of someone solving a real problem. Models commoditise by the quarter . The lasting surplus lives in deployment against the messy, multilingual, high-friction problems that fill Indian life be it credit for the borrower with a thin file, advice for the small farmer or public services in a dozen languages that a model trained abroad meets as strangers.
This is the terrain where India is strong and where capital, mesmerised by the supply side, keeps its eyes elsewhere. The abundant thing draws the attention. The scarce thing sits unpriced. That gap is the textbook shape of a mispricing.
The clearest evidence hides inside the same data everyone celebrates. The positions worth watching grow on session depth, paid conversion and word-of-mouth, far from the headlines that cheer downloads. One cohort that opens its wallet tells you more about this market than three hundred million who only opened the app. That wallet reports demand. The download reports distribution.
There is a third element, the one that ought to keep folks up at night, because its cost still sits off the books. The value of AI in commerce and public life rests entirely on trust, and trust is a bill the current price quietly defers.
Years ago, in 2011-13 , I built the case for online debit in India. As Marketing Director for Visa, I saw an entire business rest on one fragile act: a stranger’s willingness to type her card number into a screen and trust that the money would behave. Reach stayed worthless until that trust arrived.
The same law now governs AI. Deepfake-driven fraud, the synthetic manufacture of opinion, impersonation aimed at the retail investor who has only just entered formal markets, these are the bills that arrive later.
Demand looks healthier than it can sustain, flattered by a charge we keep deferring. The day it lands, through a scandal or a market shock or a wave of fraud against ordinary savers, the whole thing reprices, and it reprices violently. A country welcoming millions of first-time investors and first-time internet users in the same decade stands fully exposed to that reckoning.
All of which gives the phrase ‘Swadeshi AI’ a sharper edge than the slogan usually carries. The real question runs deeper than whether we can stamp a flag on a model. It asks whether India keeps the surplus it is strong enough to create or rents its intelligence from abroad in perpetuity while paying the trust cost at home.
Building at the model layer is the smaller prize. Owning the demand layer, the interface, the deployment, the trusted relationship with the citizen, is the larger one and it currently goes cheap. A free sample is a poor trade for a civilisational asset. A mispricing earns its keep only when it tells you who collects as it corrects.
Three actors stand ready:
- The builder who designs for what Indians will gladly pay for, and earns the wallet that a download only borrows.
- The policymaker who funds distribution and trust, the rails and the redress and the protection of the small saver, with the conviction now reserved for compute, because compute on its own is a tower built on sand.
- The investor who learns to read depth and pays for it, while the crowd still pays for downloads.
I have spent a working life building categories from the demand side, in dishwash and debit cards, motorcycles and cargo electrics, with a few sporting leagues along the way. The lesson travels intact every time. Markets go to whoever first grasps what people will gladly pay for, and why.
The largest free crowd is a vanity, pleasant to report in the morning and slow to pay by evening. India’s AI moment will be settled on this ground. The headcount dazzles. The economics beneath it still await a careful reading.
For now, we are pricing the wrong thing. The correction, when it arrives, will reward the few who priced the right one.
Disclaimer: These are the personal opinions of the author
