00:00 Speaker A
Let’s start here, Corey. Uh we’re wrapping up 2025. You’re looking at the Mag 7, Corey. And the divergences here are, they’re interesting, Corey, they’re noteworthy. and over the past year, Google, it’s a rocket ship, we know this. Nvidia, strong. Amazon, not so much. And I’m curious what you make of that, Corey. What what you think investors are looking for? Is is it monetization, Corey, or investors looking for, okay, who controls the whole stack? Explain it for us. What do you think’s going on?
00:46 Corey
I think the construct of the Mag 7 is a market phenomenon, not a business phenomenon. Which is to say you have very different businesses uh under the the the auspices of the Mag 7 that were gathered together because of kind of the size of the businesses in general, the size of the market caps of the companies, I should say. And and the direction of their stocks. And as those start to fall apart as they naturally those those correlations naturally will because they are in different businesses, the whole notion of a Mag 7 is going to look sillier and sillier over time.
01:21 Speaker A
Let me ask you about a name. I’m excited to get your take here. Nvidia. This deal with Grock, Corey. I’ve not I I I want to get the Cory take here to remind viewers. So licenses the AI technology for reportedly here 20 billion in cash. Key execs join Nvidia. What do you make of that deal, Corey? Was it a smart deal by Jensen Wong? And why did Jensen Wong decide, you know what, I’m going to structure this as licensing, not an acquisition?
01:54 Corey
Right, those are two important different questions. So let’s talk about what they’re doing first and why it’s different and we can talk about this for days, except that we can’t. But but let’s let’s see what we can do here. So uh the way that Grock chips are so different than Nvidia chips is the way they profit uh the way they process information. Specifically, uh there’s a it’s a phrase that’s really hot in Silicon Valley right now, deterministic programming. So deterministic or processing, I should say. Deterministic sort of takes a big chunk of data and analyzes it all at once. So, if you were to ask the example that Grock likes to use, if you were to ask a large language model using, you know, if you were to ask chat GPT using Nvidia chips, what uh sentence uses all 26 words in the alphabet, and the answer might be the quick brown fox jumps over the lazy dog. Well, as you type that question in letter by letter, Nvidia’s uh chips and and the chat GPT model is trying to figure it out letter by letter. W H A T S E N every what sentence? It’s it’s trying to determine what the next thing you’re going to is and it can be kind of slow. Whereas a deterministic model, using a a Grock chip looks at the whole question, figures out the answer and dishes it out to you quickly. It’s a different kind of processing that uses less high bandwidth memory, uses less processing power. The chip must be more powerful to do so, but program to do that. It’s a very different approach. and as the as the the artificial intelligence world goes from building the models, training, to using the models, inference, the Grock chip is a different approach that might work better. Nvidia had reasons to be concerned that they were going to lose share on the marketplace over time because of this deterministic approach being more simple and more elegant and and cheaper to use over time. They’re smart to get in there to try to uh make sure nobody else can take their lunch.
Add Comment