Jon Quast

We can't generate at the scale that the hyperscalers are talking about.

And there's data to back this up.

30 to 50% of the data centers in 2026 will be delayed because of power shortfalls, according to a report from Sightline Climate.

Morgan Stanley projects a 44 gigawatt shortfall through 2028.

There are already some data centers in California that have been built, but not turned on because the grid needs some fortification.

And this leads to an undeniable conclusion that the current path is unsustainable.

We need something to change.

We either need a power generation breakthrough, a compute breakthrough, or an AI model breakthrough.

And I think that we're a long ways away from having a true compute breakthrough.

or a power generation breakthrough, the AI model breakthrough is the easiest path forward.

And that's what this thing with Google is potentially talking about.

We've maybe come up with a more efficient model.

That said, we do need to tap the brakes just a little bit when it comes to the memory requirements.

So, some memory stocks are selling off because of this.

The Google announcement only optimizes a small part of the memory needs, specifically the key value cache.

It doesn't reduce the memory needs across the board.

It reduces a fraction of a fraction, not a fraction of a whole.

The memory imbalance will likely continue, in my opinion.

Yeah, I want to build on this, Matt.

You talked about taking emotions out of it.