To play devil's advocate a little, the intelligence of frontier models has grown beyond belief in the last year, and recent releases suggest that we're still no where close to reaching an intelligence plateau.

I say that as someone actually using one for difficult tasks.

A bit over a year ago, the model I was using couldn't debug its way out of a paper bag.

Now it's getting incredibly close to a point of "if you can dream it / AI can build it" with very little handholding and error making along the way.

Not only that, but incredible efficiencies are being obtained in driving the cost of that intelligence down.

The latest xAI Grok release reports intelligence on par with Anthropic's Fable 5 Max but at 85% of the cost.

That's almost an order of magnitude less cost, and there's no obvious signs we're close to bottoming out yet.

I was a lot more skeptical about AI hype years ago, but seeing is now believing.

That said, none of this means there's not excess build-out happening. I'm certain that's happening, and I have no doubt there's going to be some massive pain and consolidation in many of the companies at some point.

But on the plus side, having this immense capital and build-out available helps ensure that America can stay on the bleeding edge of technology and ahead of competitors in other countries.

It's a double edged sword.