AI HAS STARTED A LAYOFF RACE (Pt.4)

Here is the plot twist nobody wants to talk about:

AI can be useful. AI can be everywhere. And it can still fail to pay back the money spent scaling it.

Yes, really.

The most expensive illusion of the AI era is this:

“People pay for a subscription. So the model must be profitable.”

No.

Maybe one answer pays for itself. Maybe the product does.

But do the data centres pay for themselves? The GPUs? The electricity? Training the next model? The cost of borrowing all that money?

That is a very different question. And, publicly, we still do not have a clean answer.

There are three levels of “profitable”:

One request pays for itself. Your API payment or subscription brings in more than that answer costs to generate.

The product pays for itself. Revenue covers the answers, the teams, safety, support, and the next round of training.

The infrastructure pays for itself. Now add the GPUs, data centres, power, networks — and the money raised to build all of it.

Most AI conversations stop at level one.

“People pay $20 a month” sounds like a business model.

It is not an answer to whether that business model can repay tens of billions poured into chips and data centres.

And this changes how we should look at the AI race.

A company can cut staff because it expects AI to save money.

But if the AI stack itself needs permanent, enormous spending to stay useful, the savings may not be as simple as the pitch deck promised.

Next: who is definitely winning from AI — and why most business owners are not on that list. Sorry.

AI HAS STARTED A LAYOFF RACE (Pt.4)

Here is the plot twist nobody wants to talk about:

AI can be useful. AI can be everywhere. And it can still fail to pay back the money spent scaling it.

Yes, really.

The most expensive illusion of the AI era is this:

“People pay for a subscription. So the model must be profitable.”

No.

Maybe one answer pays for itself. Maybe the product does.

But do the data centres pay for themselves? The GPUs? The electricity? Training the next model? The cost of borrowing all that money?

That is a very different question. And, publicly, we still do not have a clean answer.

There are three levels of “profitable”:

One request pays for itself. Your API payment or subscription brings in more than that answer costs to generate.

The product pays for itself. Revenue covers the answers, the teams, safety, support, and the next round of training.

The infrastructure pays for itself. Now add the GPUs, data centres, power, networks — and the money raised to build all of it.

Most AI conversations stop at level one.

“People pay $20 a month” sounds like a business model.

It is not an answer to whether that business model can repay tens of billions poured into chips and data centres.

And this changes how we should look at the AI race.

A company can cut staff because it expects AI to save money.

But if the AI stack itself needs permanent, enormous spending to stay useful, the savings may not be as simple as the pitch deck promised.

Next: who is definitely winning from AI — and why most business owners are not on that list. Sorry.