Is AI the next dot-com bubble?

Everyone is drawing the dot-com comparison right now. The froth, the valuations, it's 1999 all over again.

But here's what I keep coming back to: the internet worked, eventually, because the layers held.

TCP/IP didn't decide to become a browser. AWS didn't ship its own social network (at least not really). Each layer of the stack was standardised and stable enough that you could build on top of it.

I'm not sure AI has that yet.

Right now, models are expanding into applications. App providers are training their own models. Cloud companies are building models, hosting them, and shipping their own products on top. Every player at every layer has the incentive and the capability to invade the one next to it.

The boundaries don't just shift. They dissipate. In the internet era, value accrued at the layer that was hardest to replicate. You found your defensible ground and you built on it.

In AI, that defensible position has a shelf life measured in AI years (weeks!) What looked like a moat six months ago, whether a fine-tuned model, a proprietary dataset, or a killer interface, gets absorbed into someone else's layer.

None of this means AI is a bubble waiting to pop. It may or could. And probably will have some catharsis at some point. The underlying technology is transformative, and while the dot-com crash didn't make the internet less real. It cleared the noise.

The winners in AI will look different from the winners of the web era. Less about owning a layer. More about something harder to name: speed of adaptation, depth of relationships, institutional trust, and the ability to keep moving when the ground beneath you is sliding away all the time.