Tyler Crowe

But also, there are bottlenecks behind the bottlenecks, right?

You have companies like ASML, Lamb Research, as well as KLA Corporation.

These companies that we think of like...

the bottleneck, it's like, oh, Taiwan Semi or Intel, they're like the only game in town in terms of chip manufacturing.

Well, ASML is the only game in town when it is the equipment to make the chip factories.

And I'm very curious when I hear these companies saying, we're going to do this, that they're all going to have to put in orders with these chip manufacturing equipment companies.

And I do wonder to Arm's ambitious goals,

How long are they going to have to wait in line for this equipment?

How long is it going to take to build out?

We've been talking about cost inflation and things like that.

And I bring this up specifically because I've been thinking about this a lot lately, that as much as this is an explosive growth and we have, you know, AI infrastructure basically finding any chip that they can find, whether it's, you know,

reused crypto mining or whatever.

It seems like whatever spare parts or compute power we can get their hands on, they're going to use it.

But it is still a cyclical industry.

As ambitious as all this growth is, how much capacity expansion can we have in chip manufacturing before something really starts to shift?

Even if we have this five-year growth period and we bring all this new capacity online, we could be looking at it six, seven years from now and all of a sudden we're way over capacity.

And I feel like that's a major risk for especially somebody like Arm Holdings who doesn't have this yet and wants to get into it.

So are you guys seeing something similar or is it like, ah, I think you're just kind of shaking at the wrong problem here?

Yeah, and just wrapping it up here, I think I'm more or less in line with you guys, but I reserve the right on some curveball of algorithmic efficiency where power and compute use goes way down relative to what we're seeing out of Anthropic, OpenAI, and the big power users today.

Maybe they start seeing some sort of deep seek-esque drop in compute power per token or however we want to measure it.