00:00Yeah, I mean, it's very large scale. Six gigawatts of capacity is equivalent to the output of six
00:06nuclear reactors. It's enough power for several million homes. The average size of a data center
00:11is about 300 to 500 megawatts. So it is a sort of multi-year massive commitment. And, you know,
00:17the market has faded a little bit in terms of AMD share reaction. But it's quite a big bet by
00:24Meta, that the AMD's technology, in particular, its AI accelerators, the chips that will be used
00:30to run models and what we call inference are effective. But the warrants bit, I agree,
00:36very interesting. Like, is it or is it not circular financing? I'm always up to debate it.
00:42And of course, we know that Meta announced a tie up with NVIDIA just last week. So this is about
00:48Meta also diversifying its partners in this build out. How much do we know? Or is, can we guess
00:57that Meta is spending on this partnership on this transaction? Well, on this AMD one, the way that
01:03Lisa Su and AMD put it is that it's double digit billions per gigawatt. So it's 10s of billions
01:09of dollars over the five years. Remember, AMD is a distant number two player with respect,
01:15you know, it's all of its revenues are about 10 billion dollars a quarter, which includes the
01:20data center part. NVIDIA's just data center part is more than 50 billion dollars. So, you know,
01:26they've got some ways to catch up. But yeah, it's very significant scale. And on the Meta side,
01:32right, you know, Mark Zuckerberg said it next to the president, he's committing to spending about
01:36600 billion dollars over the next five years or more. And, you know, he's always said that,
01:42like, if they misspend, quote, 200 billion dollars here or there, the risk of not spending that money
01:49is greater for them. So the stock's like down half a percentage point right now. But yeah, you know,
01:54most people, this is consistent with what we'd expect from Meta.
01:57Ed, what I'm seeing as I'm coming up to speed and all things AI and the lingo that's used in
02:02that
02:02industry, when Meta agrees to buy AMD chips, computers, okay, I get that. Where does the whole concept of
02:10measuring this in terms of gigawatts and power? How does that work into the story?
02:14Oh, that's a really, no, that's a really smart question. Basically, like, data is transferred
02:20via electrons. There's work in more nascent fields of transferring data via photons, light, in other
02:26words. But inside the server, that's how data passes through electrons. There is a heavy electricity
02:31draw to power the whole system, not just to run the GPUs and run the computations, but also like
02:37everything involved in cooling and heat dissipation. Like we're talking about electricity
02:42and that's why it's measured such. And like, to go back to Scarlett's earlier question,
02:46like this is six gigawatts just between Meta and AMD, but Meta has plans to install like tens of
02:54gigawatts by the end of this decade. And then beyond that, like hundreds of gigawatts, you know,
02:58completely different scale to what we've been talking about. That's why we track so closely in
03:03parallel what is happening with grid and energy infrastructure, because the burden is, you know,
03:09it's not, you can't even compute it with the human brain.
Comments