Sundar Pichai is an unlikely tech CEO. He does not pace the stage or reach for applause. If anything, the man at the helm of Google speaks in a slightly flat cadence, almost like he’s reading a weather forecast. But let that not fool you–he is among the smartest men in the world and which is why when he took the stage at Google I/O, the company’s annual developer conference earlier this week, those who understood what he was actually talking about, went quiet. Pichai announced the end of the internet as we have understood it for twenty-five years.
Until now, the relationship between the internet and us was clean and predictable. Have a question? Type it into a white box. The machine ‘retrieves’ some blue links with results. Pichai said ‘search’ will no longer ‘retrieve’. Instead, it will reason, build, and act for us. It will be a Personal AI Agent. ‘Agentic AI’ is what that’s called.
This agent can cross-reference medical reports, negotiate with retailers, execute payments, and file paperwork while its owners sleep. So, the white search box is not being upgraded. It is being retired. What replaces it is something far more consequential — and far more expensive — than most people yet understand.
When we type a search, it does not read the words, it shreds them. Every sentence is chopped into tiny fragments called tokens. ‘Unbelievable’ becomes ‘un’, ‘believ’, ‘able’. Each fragment is converted into a number, plotted on a mathematical grid, and processed by a physical chip running at extraordinary heat.
Why is that? Think of the way wheat becomes atta. Nobody makes a roti from whole grain. The wheat must first be crushed into a fine, uniform powder before it can be shaped into something usable. AI does something similar with language. Tokens are words ground down into machine-processable fragments.
Pichai revealed this week that Google now processes 3.2 quadrillion such fragments every single month. It is a thermodynamic statement. Every token calculation generates heat. Cooling that heat is no longer a background cost. It is the central engineering constraint of our time. This is why Google’s infrastructure spending has gone from $31 billion in 2022 to upwards of an estimated $180 billion today.
For most of the world, this is a story about corporate competition: Google versus OpenAI, Gemini versus ChatGPT. Who can spend more to do more? But let’s get this right. No one spends that kind of money to serve users more efficiently. If Google is willing to spend upward of $180 billion to make itself impossible to replace. In management lingo, it’s called building a ‘moat’.
Talking of management, celebrated author and teacher Mohanbir Sawhney of Kellog School of Management, says every CXO sits with one question: when the hype recedes, what are the structural assets that actually remain? It is a deceptively simple question, but it is the right one. Because after every product announcement, every developer conference, every breathless headline, the same truth keeps reasserting itself. Whoever owns the physical layer wins. Not temporarily. Permanently.
That is the real strategic issue for India. The question is not whether the country will use these tools. The question is whether it will remain a user forever. A user pays. An owner earns. Over twenty years, that difference compounds into something larger than a business model. It becomes the difference between a digital economy and a digital colony.
So Sawhney’s question comes back to India in sharper form: when the hype around agentic AI finally recedes — and it will — what will India have built that still remains?
This matters because India has long welcomed global technology majors while struggling to build indigenous technology at scale. In the current environment, it is easy to invite hyperscalers in, offer tax holidays, and let them anchor their computing pipelines on Indian soil.
But that also opens the door to compute-colonialism, a structural asymmetry in which India absorbs the environmental cost, the water tables bear the thermal load, and the strategic intellectual property remains locked in boardrooms elsewhere. In practical terms, that could mean Indian queries, Indian forms, Indian commerce, and Indian public services running on infrastructure the country does not own.
There is, however, a lesson India already knows how to apply. When mobile payments arrived, it did not wait for a mature credit card infrastructure to evolve. It leapfrogged the legacy system and built UPI, a payments stack that now processes more transactions than most developed economies combined.
If history is a pointer then, Agentic AI will diffuse in much the same way. Indians will create new systems that remove friction. The adoption is not in question. The question is: whose infrastructure will it run on?
