The AI math gets absurd at current GPU prices

The economics of AI are getting harder to defend at current GPU prices.

The hyperscalers currently generate roughly $70 billion a year in AI cloud revenue.

But according to the Financial Times’ calculation, they would need about $636 billion annually to achieve roughly the 30% return on invested capital they historically targeted.

That’s roughly nine times the current revenue run rate.

And that only covers the hyperscalers.

The AI labs and application companies need to make money too. Once those margins are included, consumer and enterprise AI spending could need to approach $2 trillion a year.

That’s the problem.

Not whether AI works.

Not whether people use it.

Whether people will actually spend anything close to $2 trillion a year on AI software.

Consumers have spent decades getting used to free digital products. Convincing them to suddenly pay enough to support a hardware industry spending hundreds of billions of dollars is another matter.

The industry can keep selling GPUs.

It can keep building data centers.

But eventually the customer has to pay the bill.

At $70 billion of current AI cloud revenue, the gap to $636 billion is enormous.

And the gap gets even larger once everyone in the AI stack wants a return.

The technology can be real and the economics can still be terrible.

Not financial advice.