Monologue: Blame AI For Making Everything More Expensive

0:02 More media. Hello and welcome to this week's Bear Offline Monologue. I'm your host ed Ze Trump, and today I'm talking about the great parasite of AI and how it's making everything more expensive. And to be clear, before we go any further, what you're seeing in these remarkable sales from video and all the RAM companies in Broadcommon and all them isn't the measure of industry demand, but of speculative purchasing based on the vegas the projections. Because the actual demand for AI, the actual revenue generated by this industry is kind of pathetic. And honestly, we don't really have any breason to build these data centers. Nobody has trouble accessing these services. I mean, it's just being done because

0:48 they've got nothing else to do. And I know that sounds kind of flippant, but really is what's happening to make that's worse. The more the bubble inflates, the more expensive everything gets, the more data centers that get funded, the more that are ordered, The more resources this takes, and the more the RAM companies can crank up the prices and fuck over the customer. You see, aigpus is a special kind of memory, high bandwidth memory that takes up more of the limited amount of fab space where eske Heinex, Micron and Samsung build their RAM. These are also the only three companies that really can build HBM.

1:21 It's a triopoly, I guess you'd call them. And because in Nvidia broad common, other chip manufacturers quite literally can't build the GPUs that hyperscalers crave without them, their memory companies can charge effectively whatever they want. Aigpus have become a massive amount of guaranteed revenue. Is the AI bubble inflates, leaving less fab space for the regular RAM, creating a supply shortage, or put another way, a reason for the memory companies to bump up the prices aggressively. And yeah, they're allowed to do that. There's nobody to stop them.

1:49 It's not like we have regulatory authorities that would bother I should also add that in Vidia is one of the largest, if not the largest HBM buyer in the world. Fifty five to sixty five percent of all high bandwidth memory sales go to Nvidia, and this makes it actually a victim of this same inflation because memory is getting more expensive across the board, even for high bandwidth memory.

2:10 Gold and Sachs reports that high bandwidth memory is on course to increase by as much as ninety percent year over year in twenty twenty seven. A linear increase in costs for every single in Vidio GPU and every single new one that one of the four hyperscalers OATHS insists on pre ordering only serves to inflate a bubble for effectively no reason. They're not making a profit. Shit, they won't even talk about revenues. Maybe they're not really making that much outside of open AI and anthropics compute, like I've been saying. But good lord, good lord, good golly, Miss Molly, do these data centers take up a shit ton of RAM? Analysts estimate that by the time that in Video ships its first next generation Vera Roubin GPUs,

2:49 it'll cost eighteen dollars and forty cents per gigabyte of high bandwidth memory sold two hundred and eighty eight gigabytes at a time. In fact, let me give you some scale as to how much RAM and a data center actually needs, because it's an astonishing amount. The current generation GB three hundred racks the vast seventy two GPUs servers that fill out modern data centers, and they have a CPU for every two GPUs have about twenty point seven terabytes of high bandwidth memory and seventeen terabytes of LPDDR five X RAM. And by the way, that's the RAM that you find in smartphones and mobile devices, which is also contributing to the supply chain shortages and driving up the price. Anyway, a giga what data center is about

3:29 four nine hundred and thirty three of these raks, working out to about one point ninety four billion dollars in high bandwidth memory alone for a single data center, a number that's certain to increase across the board, even for current generation gear because the price of high bandwidth memory is going up. I haven't even included the many separate computers and different machines inside the data center that require RAM, two entire racks of power management gear that themselves require RAM or HVAC systems alone. They're effectively servers unto themselves, and every the data center is basically a giant warehouse of different RAM dependent devices and machines, each controlled by other RAM dependent machines or to do one thing. AI

4:08 that's yet to prove it even has a meaningful business model or one that I don't know makes more money than it loses. And that high bandwidth memory, by the way, is also not particularly useful for anything else, at least not at the scale that's being built, meaning that once there's a noted Kapex pullback, somebody is going to be left with a bag, by which I mean somebody is going to either have too much high bandwidth memory or gear that they can't sell with it put on the side that won't be useful or really sellable to someone else.

4:34 And I must be clear how much more of this they are selling than usual. Back in twenty twenty two, they sold about six hundred million gigabytes of high bandwidth memory, so sounds like a lot until you realize they're expecting to sell seven trillion or more of it in twenty twenty six. That's that's a problem. That's a problem because the more of it they sell, the more expensive it gets.

4:57 But that's they're going to have years worth this crap just either sitting around or being attached to in video GPUs that can't sell to anyone. And the longer it takes for Capex to retract, the more expensive everything becomes. As a result, the more GPUs that get sold, the more capacity that gets put towards high bandwidth memory, and the more that Micron, Skhienex and Samson could charge for it, which makes it more expensive to buy aigpus, which increases the amount that hyperscalats are spending on AI capex for effectively the same amount of gear. The longer that Hyperscale has sustain this pace, the larger the return needs to be.

5:30 And at this point none of them have disclosed their actual AI revenues, which heavily suggests that there's yet to be even a single dollar of profit, and the actual quarterly revenues are pretty piss poor. Yet the more they commit, the more committed they have to be. Pulling back at this point will prove to the markets that all of the hyperscalats have committed far too much to too much capacity.

5:49 Yet not pulling back means that hyperscalers will continue to turn their free cash flows negative in pursuit of some indeterminate goal. It's a vicious cycle, made worse by the fact that every very spin of the Capex wheel increases the price of just about every consumer electronic in the world, creating a market wide inflation for what amounts to a speculative asset bubble, and will leave them sitting with a bunch of GPUs they don't really need now and certainly won't in the future. What great works are these data centers in search for? What possible achievement can you point to at this time that remotely justifies the current expenditures, let alone the trillion plus more plan for the future.

6:28 This is the vicious cycle of the AI bubble. The more data centers they build, the more expensive it becomes to build them. Because of the cartel of memory manufacturers that adjust the prices up at basically whatever rate they want to and hyperscalers that have no idea what to do other than spending more money. RAM is ineffectively everything, and thus everything is going to get more expensive from here. Samsung, sk high Inex and Micron have committed more than two trillion dollars to building out more capacity, even though this bubble has been inflated based on building high bandwidth memory that is being built at a scale beyond any reasonable level of intry demand. Will that capacity because you can

7:02 build other things in the same RAM capacity could theoretically be used for something else. The question is how they intend to pay for it, and how many of these actual commitments they've made with their own money, and indeed, how many of the commitments made to them by hyperscalers buy in video can be canceled? And I expect we see a large and more grotesque version of the massive write offs and losses across memory manufacturers that we saw when the supply chain crisis ended in twenty twenty three, after the supply chain crisis that started in twenty twenty one.

7:31 And when the time comes, know who to blame. Sam Altman, Dario Amadei, Sundar Pashai, Andy Jesse, and Mark Zuckerbo the men who created this entirely avoidable and horribly destructive situation. They all fucking suck. They could have done literally anything else. They could have built housing, they could have fixed their own products, but they chose to do this because they have no vision, These met a losers. I despise them for what they've done, and I hate them for what they've done to the computer. Thanks for listening. We'll be back next week with a special memory focused episode. Would the Burg of Gamers Nexus And I'll be back, of course, with another monologue. Cheers, madeirs