Stacy Rasgon

And by the way, in and out of semis, you know, if AI demand ever does roll over, like, we're all screwed.

Yeah.

In semis and out of semis, it almost feels like even in the broader economy, it's the data center spending and AI spending that really is supporting everything right now.

But what do you think?

Well, I mean, it depends on how much compute you need.

And so this gets now to the usage and the return on these assets.

And again, it's still early days, right?

And the amount of revenue that is getting generated from these assets is still probably relatively small.

Most of the spending we've seen so far has been used.

They built these massive compute clusters to train these models.

I don't know how much your listeners know about AI models, but you can sort of think about...

like a massive amount of data and parameters that have to be set to create a model that can perform a task.

And so most of the compute spending so far has been to train these models.

So what we need to see is these models actually get used more for something, right?

This is actually called inference, right?

And it's still early days, but we're already seeing evidence.

You can take a company like an Anthropic, for example, who discloses their annualized run rate of revenue.

And Anthropic makes something that's called CLOD that's used for agentic coding, for example.

This is where you can actually use the AI model to help you code.

And I mean, their revenue has gone vertical.