Kevin O'Leary

They were useful for just basic compute and storage of data.

But then when these models started emerging, such as we have going with Anthropic or Gemini,

and any of them you know grok all of it and all of the tools built on top of them the amount of compute power geometrically required geometrically grew and it was no longer 100 megawatts the minimum was minimum was 250 megawatts then four months later the minimum was 500 megawatts

Wait, here it comes.

And then two months after that, it was one gigawatt.

Now, try and get a gigawatt out of the American grid or the Canadian grid or the Mexican grid.

Not a chance in hell.

We're tapped out.

So the Chinese were building these one gigawatt facilities.

So it doesn't matter about the number of facilities.

It matters how many are gigawatt plus for training AI models, because you need a lot of compute for the modern day chip.

And this technology is advancing very, very quickly.

So it's the size of the large ones are the ones that they're beating us on, not the small hundred megawatts.

Nobody cares about those anymore.

You can't train anything on that.

So we're competing now for campuses, 10,000 acres and more at a time.

And that's where the Chinese are kicking our asses.

Sure, that's a good question.

You need, you can't tap in and you can't, you tap into somebody's grid.

Let's take a place like Texas or in Mississippi or Utah, for that matter, where I am.