Amazon Dumps NVIDIA for AMD: Will This Move Pay Off?

Craig: so amazon is dumping nvidia for amd trying to say that and loosen nvidia's power hold and move forward with amd chips to power their ai. do you think this move is gonna pay off?

Jon: yeah. so, sorry. hold on one second. so i think we've gotten it wrong. i think, amazon is going at creating their own chip, and openai is going after amd.

Craig: everything according to my stuff is that openai know? oh, yeah. openai is planning to release their new chip in twenty twenty six, and they're switching to that, but they're dropping intel. because they were using intel chips. amd is dropping nvidia. sorry. amazon's dropping nvidia in favor of amd, and they're talking potentially about using their own chips down the road, but we don't know. we were talking about that. i think they're all talking about using their own chips down the road. yeah. you know? and this goes back to our discussion where the only companies making money and they are the hardware companies because all the software companies have to spend, spend, spend. and they're realizing now that their biggest dollar amount is buying these chips to power their data center, and they need to start making them themselves.

Jon: yeah. yeah. so the ours technical technical article on amazon was, what i was looking at was, the terrarium. how do you pronounce? the, terrarium two chipset, which they're with which is anthropic is using right now.

Craig: yeah. and that's the article i have pulled up here now. it was just one of the two. so amazon's poised to roll out their newest artificial intelligence chips as a big tech group seeks returns on a multibillion dollar semiconductor investment and to reduce its reliant on market leader nvidia. everybody seems to think that nvidia is too big for its preaches, and we need to stop relying on them as a whole. yeah. and so they're starting to change that. so amazon's working on their new chips. trainium is already being tested by anthropic. the openai competitor has secured four billion in backing from amazon, as well as databricks, deutsche telekom, and japan's ricoh, and stockmark?

Jon: yeah. so this is cool. so it's gonna be available in the amazon e c two t m two instances, which contain sixteen trainium two chips per instance. it's got some cool features. it's designed to deliver four times faster training performance compared to the first generation of trainium chips, and it's estimated to provide around six hundred and fifty teraflops of low precision compute performance per chip. and this is cool. it features ninety six gigabytes of high bandwidth memory, which is three times more capacity than the previous generation. the architecture itself is a multi tile system in package, with two compute tiles and additional chiplets. so, this is really cool. what baffles me is that there was this company, canadian company, grok with a q, who originally were designing chips, but then moved to software, which is weird because you would think they would keep making the hardware because that's where the money is. right?

Craig: software is cheap. hardware is expensive.

Jon: yep. they already designed a chip. yeah. it's billions

Craig: of dollars. you know? look at everyone pulling out a different hardware stuff because they spent so much. so, dave brown, the vice president of compute and networking at aws, says we wanna be the absolute best place to run nvidia, but at the same time, we think it's healthy to have an alternative. and amazon, refers to their inferentia, which is another line of a specialist ai chips is already forty percent cheaper to run for generating responses from ai models. now everyone's running out to build their own asics as colby, was mentioning here. it's very similar to, in the bitcoin heyday. many of companies were running out to, do optimized bitcoin, asics that would, you know, solve that algorithm faster to generate more bitcoins.

Jon: yeah. but yeah.

Craig: so it's the same thing. everybody's getting into this game. we'll have to see there's only so much room to go around when it comes to chip manufacturing because it is expensive. so we'll have to see who really can carry this ball forward, and it's really nice to see. i like it when other people are getting in the game. like, i like it when there's competition, so it's nice to see amd coming out, with this hardware and being competitive. and, apparently, grok is still making hardware. this is controlling the ecosystem because vertical integration is easier to solve than, their bulky chips. awesome.

Jon: thanks, cole. me. yeah.

Craig: we appreciate that. and, you know, again, hardware is expensive. if your chips are buggy as colby, you know, indicated there and whatnot, you need to fix that before you go into a high end manufacturing or else look at all the trouble, intel's been in because of a few slip ups on chips, and we think, you know, they're gonna be bought out.

Jon: yeah.

Craig: so everybody from openai to apple is looking to build their own chips, as they seek lower production costs, higher margins, greater availability, and more control, the question becomes, are they gonna move into the hardware game? amazon's always moved into the hardware game and, you know, semi successfully. so we'll see if they're building their own chips. are they gonna do it? if openai is building their own chips, are they gonna do it? are they gonna meaning, are they gonna sell those chips out to people, or they're proprietary only to themselves?

Jon: yeah. that's interesting. i like, who would they sell to? because everyone is developing their own chips now. right? so i guess they could sell to other startups looking to start up. but maybe if microsoft develops their own or if amazon develops their own, i don't think they're gonna buy it from open the a chip from openai. maybe. who knows? but, it all depends on the quality of the chip and the performance of the chip at the end of the day.

Craig: right? well, we haven't heard anybody like facebook buying, meta making their own chips yet, but, maybe that's just because i haven't read those articles. but there's plenty of other a eye companies that would buy those chips to run ai data centers. so don't be

Jon: actually, meta is developing its only ichip. well,

Craig: there you go. yeah. there you go. i just hadn't read about it yet.

Jon: i know. there's so much tech news. we miss a lot of it. right?

Craig: well, we're with ai coming out every day every day, there's, like, new stuff. it's in and we, you know, have multiple stuff going on. but it's this helps us stay ahead, of course. but i'm really interested in all these new companies. you know? nvidia log, twenty six point three billion in revenue for ai dentists data center chips in the second fiscal quarter of twenty twenty four. so that's just in a quarter. so no wonder, you know, these companies are trying to say we need to, one, get a hold of this, manage our supply, because they can't give us what we want on demand. the prices are too high. it's do this, do that. so they're going to customize these things each, and we'll have to see. i am curious to see if they're gonna turn it not just into, a cost reduction and maybe a, company advantage from a hardware perspective that if they'll turn it into a profit center as well.

Jon: yeah. we'll have to see. it's it's interesting. i guess we're in the chip race now. ai chip race or gpu chip race and

Craig: yep.

Jon: you know, we'll have to see, where it takes us.