The AI Bubble Deflates, The Technology Doesn't

Every downturn in tech gets narrated as the end of the thing that boomed. AI's version of that story is already being written, a deflating bubble, tightening capital, a reckoning coming. I think the bubble and the technology are two separate stories, and conflating them is the mistake.

the bubble was always about capital, not usage

A bubble deflating is a correction in how capital gets priced, not a verdict on whether the underlying technology sticks. The dot-com crash wiped out valuations built on nothing, but the web itself kept growing through it, more users, more usage, more of daily life running on it, the whole time the capital markets were busy punishing the excess. Nobody today argues the internet didn't work out because pets.com died. I expect the same split here. Whatever correction hits AI valuations doesn't touch whether people keep using the tools, and usage is the actual signal I care about, not the funding headlines.

usage compounds, it doesn't plateau

The bubble narrative assumes AI's value is front-loaded, a burst of hype that has nowhere to go but down. My own history with new technology argues the opposite. I emailed Greg directly to get early access to GPT-3, and I was using Instagram back when it was still finding its shape. My first reaction to a lot of that was to dismiss it, the flaws were obvious and I walked away more than once. But I kept giving those products a second chance, and the pattern was always the same, more exposure, tolerating the bugs, watching other people figure out uses I hadn't thought of, and eventually refining my own usage until it actually clicked. That's not hype fading, that's a skill compounding through an ugly middle phase most people quit before reaching. A downturn in capital doesn't erase that middle phase, if anything, it's the people willing to sit through it who end up furthest ahead, while the tourists who only showed up for the headline round sizes leave.

this still feels like the start, not the middle

Where I actually land on this is that it feels like the opening chapter of the AI era, not a late one. There are frontiers nobody's properly started on yet, globally and specifically in India. That's not a hedge against the bubble talk, it's a bet that the interesting building, the kind only visible from inside a market, hasn't happened yet, regardless of what the next two quarters of AI valuations do.

localisation here means behaviour, not language

When I say India is a frontier, I don't mean language localisation. Indians behave differently as a low-trust society, and building for that behaviour has always come from someone inside the market watching it closely, not from a global product adding a local skin. Flipkart built cash on delivery because Indians were skeptical of paying online, and Amazon had to adopt it after. Ola built OTP verification into starting a ride because drivers sometimes started rides without ever picking up the passenger, and Uber had to adopt that too, it's standard everywhere now. Rapido pushed it further with OTPs checked continuously through a ride, and Uber picked that up as well. None of that was a language problem, it was a trust problem, solved by someone close enough to the behavior to see it, then copied by the giant once it worked. I expect the same sequence in AI, infrastructure, model, or application layer, a local product surfaces the behavior-driven fix first, and the global giant adopts it after.

conclusion

What I watch for instead is usage, not headlines, whether the people around a product keep getting more fluent with it or whether they quietly drift away once the novelty fades. The bubble will do whatever bubbles do. The technology's story is a longer, separate one, and I've watched that story before, it just took me a while to stop dismissing it each time.