AI is not one story

And breathe.

There’s a lot of noise about whether AI is in a bubble.
But over the past month, the conversations I’ve been part of - founder pitches, VC financings in AI application layer, diligence - paint a picture that’s far more uneven than the headlines suggest.

And that’s the point most commentary keeps missing.

People keep calling AI a bubble because they’re staring at one coherent “AI market” - and that market doesn’t exist.

There are three markets moving at three different speeds:

1) Infrastructure: the chips, data centres, supply chains.
2) Applications: the part that actually creates value for end users.
3) AGI: the long-horizon belief system.

Each layer has different investors, different risk profiles, different economics.
They’re not inflating or deflating together.
They’re not meant to.

So trying to diagnose them as one organism is like taking three independent heartbeats and calling it a single pulse.

And here’s something else people forget:
in every vertical - bubble or not - only a few companies ever generate returns above their cost of capital.
That’s not froth.
That’s capitalism.

Retail, fintech, biotech, SaaS - they all consolidate around a few winners.
Dozens of companies falling short doesn’t signal a bubble.
It signals competition.

There’s no reason to hold AI to a higher bar unless there’s a compelling reason to do so.

Yet, people do - because they’re conflating company risk with technology risk.

Consumers don’t care which AI company wins.

This isn’t social media, where switching platforms means abandoning a social graph or an identity.
If one model disappears tomorrow, people will simply open the next one.
No loyalty.
No friction.
No memory.

Companies may come and go, but the technology isn’t going anywhere.
AI isn’t a “company story.”
It’s a cost-curve story.

And ironically, if there is a bubble anywhere, it’s probably in infrastructure - the most capital-intensive layer with the weakest pricing power. Billions in debt raised now for GPU capacity that won’t be fully monetised for years.

But even if infra is overbuilt, the result isn’t collapse - it’s cheaper inputs.

If GPU capacity gets ahead of demand, the cost of goods for the entire app layer drops:

• cheaper inference

• cheaper training

• cheaper experimentation

Value doesn’t evaporate - it migrates.
A capex bubble becomes a margin-expansion story for B2B and consumer apps.

Unlike the Metaverse (remember that…?) - a top-down attempt to manufacture a future - AI is spreading bottom-up.

Kids are using it for homework.
Students for revision.
Creators for ideas.

People dismissed that once.
The same way they dismissed TikTok as a place for dance videos - until it became the default discovery engine for an entire generation.

This is how real platforms grow:
quietly, peer-to-peer, long before institutions catch up.

Which is why the question “Are we in a bubble?” is the wrong one.
The more useful question is: what part of the stack are we actually talking about?
Because once you separate the layers, the narrative stops looking like mania -
and starts looking like a transition that’s messy, uneven, and entirely normal.

✍️ Note to self: The market isn’t one thing. Neither is the truth.