I understand supply demand has driven solid state memory prices up because of AI compute, but why are spinning drives going up in cost too?

42:32I understand supply demand has driven solid state memory prices up because of AI compute, but why are spinning drives going up in cost too? I have a feeling I know what you're going to say, but go ahead, Carl. So AI data centers are that exactly. They do both AI, which is processing of tokens but they're also data centers which means they need data so i'll use grok as an

43:00example because grok kind of tells you what it's looking at when it's thinking so if grok goes out and looks at like a cnn web story uses it looks at wikipedia it looks at grokopedia it looks at all these different things it's not actually going out to the internet to look at those it's actually looking at its own hard drives spinny drives and these large capacity hard drives

43:17because it's scraping the public internet so we know that grok has already scraped the entirety of of Wikipedia because of Crocopedia. So we know that, but it's also scraping all these news. And this is where the news ones are kind of funny because Croc only needs one subscription behind the paywall and it can scrape everything. And it's like, oh, you're going behind the paywall.

43:36So that's one of the arguments, but anyway. So it's scraping all this data from everywhere. And it's kind of working if you remember how a fusion drive worked back in Apple. So Apple has essentially there's the processor and then you have the faster SSD drive. And then you have the spinning drive behind it for stuff you use less regularly now in this scenario it's not ssds essentially ssds are used

43:59for more short term the hard drives is really taking the place of the ssd and a fusion drive and the public internet is taking the place of the spinny drive and a fusion drive so we're just stepping one step away so it will put stuff that it regularly gets asked to look at into hard drives physical hard drives on its own server so it can get to it quickly without any data costs because

44:17it's got to pay for ISP charges going in and out of the building and if the stuff gets very irregular it will go out and look at the web but if it gets asked to look at something like two or three times it will simply scrape that or maybe it scrapes it every time depending on how much and this is the thing this is why they have so many large capacity spinny drives is because every time it gets asked

44:37something and it doesn't have it in its own storage on spinny drives it has to go out to the web within the algorithm it says how many times do I have to look at this website before I scrape it and bring it into the hard drive so i don't have to go out to web again but it does have to look at a delta to see if something has changed so and again in the algorithm if the information is uh

44:56old and it's not going to change like a story from two years ago like a new story then it can just scrape that and not have to go look at the delta but this is kind of all the stuff that's happening behind the scenes when you ask something and this is why apple um enlisted google's help to do outward looking ai requests because google is doing this too google is literally scraping

45:16from the internet for, I guess, Gemini is their product. Apple have their own deal with Google. So they're not technically using Gemini, but they're using Google's storage on their hard drives. That's what Apple and Google is really doing because Apple can go out and look at the internet itself from its own data centers, but it doesn't want to do that. It wants to use a store of the internet

45:36that's on hard drives in someone's building, which in Apple's case, it's Google. So it is kind of convoluted, but the reason is AI data centers need need as much lithography as possible, and that's for the processing. Lithography again for SSD, so that's lithography time taken up and that's why it's expensive to get SSDs and RAM. But then it also needs an enormous amount of spinning drives because it needs just an

45:58enormous amount of data that can be accessed relatively quickly, you know, within a few seconds, for it's essentially the data that you're asking for it to look at because it doesn't want to go out to the public internet every time, and the public internet is sometimes fickle. So they just simply want the data in their own building so they can provide a better product than, you know, the competitor.

46:18Yeah, the answer is that the two are inexorably linked, no matter how you do it. Anytime prices go up on one, the prices will follow on the other. Courtney, you have something to add real quick? Well, also on those spinning drives is RAM cache involved, and that means they got to put those chips on those spinning drives too. And since all those RAM chips and, you know, flash RAM and especially static RAM and dynamic RAM comes out of the same foundries, you know, they can't make those spinning drives unless they have the cache.

46:46So, you know, they're going to run in, you know, even if they're only using, you know, a couple of megabytes of cache, they still got to put the RAM in there and they still got to come out of the same foundry. So that's why the prices are going up. They're not, you know, they can hold up production because of the lack of live stream.