Inside the Battle for Chips That Will Power Artificial Intelligence

Joe Weisenthal (0:10)

Hello, and welcome to another episode of the Odd Lots Podcast. I'm Joe Weisenthal.

Tracy Alloway (0:15)

And I'm Tracy Alloway.

Joe Weisenthal (0:17)

Tracy, I'm not sure if you've heard anyone talking about it or anything, but have you heard about this sort of AI thing people have been discussing?

Tracy Alloway (0:24)

Oh, you know what? I discovered this really cool new thing called ChatGPT.

Joe Weisenthal (0:29)

Oh yeah, I saw that website too.

Tracy Alloway (0:30)

Yeah, have you tried it?

Joe Weisenthal (0:33)

I tried it, yeah, I had to write a poem for me.

Tracy Alloway (0:35)

See, pretty cool technology.

Joe Weisenthal (0:38)

We should probably learn more about it.

Tracy Alloway (0:39)

Yeah, I think we should. No, okay, all right, obviously, we're being facetious and joking, but everyone has been talking about AI and these new sort of natural language interfaces that allow you to ask questions or generate all different types of texts and things like that. It feels like everyone is very excited about that space.

Joe Weisenthal (1:02)

Every, like, almost every conversation, like I went out with some friends that I hadn't seen in a long time, like I was at a bar last night. And like the conversation turned to AI within like two minutes. And everyone's talking about the experiments they did. But yes, there is a lot, it's basically like this wall of noise.

And everyone's been talking about, actually about us, because I don't think we have done, as far as I can recall, like an AI episode. We don't want to just add to the noise and get another sort of chin stroke around, but obviously there's a lot there for us to discuss.

Tracy Alloway (1:34)

Totally. And I'm sure this will be the first of many episodes, but one of the ways that it fits into sort of classic Odd Lots lore is via semiconductors, right? If you think about what ChatGPT, for instance, is doing, it's taking words and transforming them into numbers and then spitting those words back out at you.

And the thing that enables it to do that, semiconductors, chips.

Joe Weisenthal (2:01)

Right. So here's like the four things I think I know about this. And so is that A, training the AI models so that they can do that is a computationally intensive process. B, each query is much more computationally intensive than say a Google search.

Three, the company that's absolutely crushing the space and printing money because of this is Nvidia. And four, there is a general scarcity of computing powers so that even if you and I like we're brilliant mathematicians and AI theorists, et cetera, if we wanted to start a Chad GPT competitor, just getting access to the computing power in order to do that would not be trivial even if we had tons of money. Outside of that-

Tracy Alloway (2:46)

I'm gonna buy an out of business crypto mine and take all the chips out of that.

Joe Weisenthal (2:50)

Tracy, they've already been bought.

Someone got that. But that's it. That's basically the extent of my understanding of the nexus between this AI and chips. And I suspect there's more to know than just those facts.

Tracy Alloway (3:02)

Well, I also think having a conversation about semiconductors and AI is a really good way to understand the underlying technology of both those things. So that's what I'm hoping for out of this conversation.

Joe Weisenthal (3:14)

All right, well, you mentioned we've been doing, we've done lots of chips episodes in the past. So we're gonna go back to the future or something like that. We're gonna go back to our first episode, our first guest where we started exploring chips episodes. I think it was the first one that we did sometime, maybe in early 2021

We're gonna be speaking with Stacy Rasgon, Managing Director and Senior Analyst of US Semiconductors and Semiconductor Capital Equipment at Bernstein Research. Someone who's great at breaking all this stuff down has been doing a lot of research on this question now. So Stacy, thank you so much for coming back on Odd Lots.

Stacy Rasgon (3:49)

I am so happy to be back. Thank you so much for having me.

Joe Weisenthal (3:52)

All right, so I'm gonna start with just sort of like, not even a business question, but a sort of semiconductor design question, which is this company, Nvidia, like for years, I just sort of knew them as like, they were the company that made graphics cards for video games. And then for a while they got there like, oh, and they're also good for crypto mining. And they were very popular for a while in Ethereum mining when it used Proof of Work.