The Dirty Secret Behind AI Data Centers No One Wants to Talk About

[00:00:00] Speaker 1: There's this concept right now called the Ramageddon and it's nasty and here's why. Did you know 10 years ago if I ask you, how many AI data centers do you think we had in America in 2016? You know what the answer to the question is? Zero. You know how many we have today? 4,000 is the low number, up to 5,400 AI data centers and it's so bad right now that people don't want them to be built in their cities. Missouri just had a story a couple days ago, AI data center they wanted to build. One guy ran against four other city councils. He beat everybody. He had one campaign. No AI data center if you vote me in. The state of Maine is saying, hey, you guys pump the brakes. You can't come and build AI data centers. Why is this taking place? While we're talking about all of this stuff, data comes out that 70% of all these chips that they build, you know these chips like these DDR5 rams that you put in your computer or maybe in the PlayStation that you buy? 70% of these chips that they're building that we would use for consumer electronics. Guess what is being used right now? For AI data centers. TSMC, I believe it's called Taiwan Semiconductor Manufacturing Company. That company built a center in Arizona to build their chips. They said, it's going to cost us $11 billion. And they said, not $60 billion project. You know how much it is right now they just broke around? $165 billion to build this project. It's so scary while they're talking about this that I want to show you this chart. Look at this here. This is what the price of an average DDR5 for 16 gigabyte RAM was. This is July of 2024. Same price, around $125. All of a sudden, they start developing these AI data centers. And then look what happens. Boom! Skyrockets. That's December of 2025. When I tell you the prices here today or what's going on, you're going to be blown away. Anyways, having said that, today, we're going to talk about this concept of Ramageddon and a company that may disrupt the whole thing. That cost may go back down to what it used to be pre-COVID. Okay, so if you're wondering if I'm limping, it's because I just tore my ACL and I'm wearing this brace here on my knee. But having said that, I'm going to do this standing up. If you get value out in today's video, give it a thumbs up and subscribe too. So let's get right into it. All right. So Ramageddon, is it this serious, Pat? It kind of is. Why? Let me simplify it for you. So imagine you bought a PlayStation and I'm going to start a little bit backwards for you. In 2020, it's the last PlayStation model that they made for $499. You would assume that that PlayStation, if you bought it for 2026 years ago, if you wanted to sell on eBay today, it probably would be worth $300, right? But do you know if you have a used PlayStation right now, you put it on eBay, you're not selling it for $499. You're selling it today for $599. You're selling it for $600 today. You know why? Because these chips are so expensive today. So let's kind of get into it. So these words that we're going to be using, we're going to learn it together. RAM. What is RAM? Random access memory, the short-term memory of a computer. Data is wiped when the device turns off. DRAM. Dynamic RAM, type of RAM used in almost all modern devices. Now, when we go to the DDR, the double data rate, the standard language for memory, each generation DDR4, DDR5. DDR5 is the latest one that we built. And that's what we're using today for the 2026 computers. They use a DDR5. Now, HBM is the high bandwidth memory. Instead of laying chips flat, you put layers. So 8 to 16 layers, it just goes up and you stack them up vertically. 3D goes all the way up, significantly faster and more power efficient. The DDR5 takes 8 to 16 weeks to make. The HBMs takes 3 to 6 months to make, but obviously they're a lot more powerful. So you kind of get an idea. The cost for these things can be $120, $30 to $60 if you're talking about what it costs them to make it. To sell it, of course, it's a lot higher. The HBMs are more expensive. It could be $1,200, $1,400, up to $3,000, depending on where you're making it. And keep in mind, the wafer you need, we'll put the image here to know what the wafer looks like. You know who's a quiet king of materials when you want to make this kind of stuff? Everybody needs the wafer. Most of it are built in Japan. Now, if you want the chips, you got, I want to say Taiwan and South Korea that are making the chips. I think most of it is probably in Taiwan. That's why you hear the Taiwan semiconductor manufacturing company. They control so much of it. This is why Taiwan is such a powerful country when it comes down to the chips. Well, let's go through it. Memory-hungry AIs, like Chad, GBT, Claude, and Gemini, are using massive amounts of memory. Okay. In 2026, this is the kicker here. AI data centers will absorb 70% of all global DRAM production. So imagine the stuff we used to use it for. These AI data centers are like, "Hey, we need it all." And they're signing massive contracts. Wait till you hear what Sam Altman committed to two separate companies on how many he was planning on buying in 2030. The number is insane when you hear what he said to them. And then later on, he's like, "I don't know if I said it." Both companies didn't fully know about it. It's like, well, it was kind of a non-binding contract. So relax, guys. I didn't give you money yet because of some things that took place. So stay with me here. The big three monopoly, Samsung, SK Hennix, and Micron control 93% to 95% of the RAM market. 93% to 95% of the market they control. So they have the market right now. Now all have shifted, focusing on DRAMs to HBMs for AIs. And HBM offers a much higher profits. One HBM module to NVIDIA could be the equivalent of 15 to 20 different DDR sticks total. So imagine one HBM versus selling 15 to 20 DDRs. Producing one byte of HBM requires 300% more production capacity than one byte of DDR. So imagine the electricity, the power, the number of employees, people, the bandwidth, all the costs that you need, leaving consumers with industry scraps. We get the regular stuff. The players get the big stuff and manufacturers aggressively cut production to clear inventory. 2022 to 2023 and AI demand surged in 24 in waves we've never seen before. So let me kind of give you some numbers here. What happened during COVID? Do you remember during COVID like regular used car prices went up? Like my car, I bought for $100,000. I put 40, 50,000 miles on it. Like, you know, somebody will buy your car right now for $110,000, $120,000. I'm like, what are you talking about? Yeah, because we don't have chips. And so used cars were selling more than regular cars because like, wait, I can't. Yes, this was happening during COVID. So to give you an idea, pre-COVID 2019, the average price for 16 gigabyte, $60 to $80 price per gigabyte was $430,000. COVID peak 2021, it goes to $750,000 per gigabyte. Then it drops back to back down to the normal price of $310,000 post-COVID low. Now guess what it is? $1250,000, four times higher to what the low was post-COVID. Wait, what? So during COVID, we needed these chips because we couldn't find them to build cars. What are you talking about? Yes, AI just happened. I mean, open AI is not like it's 22 years old. If you look at Claude and all these other companies, these are not companies that have been doing these AI data centers for 20 years. AI data centers are something new that happened the last 10 years, less than 10 years. So the prices have gone up and they're sitting there saying, wait, what is this going to turn into if they continue? I already told you the PlayStation example, but let me give you an idea what this goes to. While everybody is talking about all of this cost, guess what ends up happening? Okay. The reason why these prices went up, this guy named Sam Altman, in October of 2025, he signs a letter of intent allegedly with Samsung and SK Hynix, in a letter that he wrote. It's called the ghost order in 2025. He said, by around 2029, I'm going to order 900,000 RAM wafers per month from you guys. And he's telling these two companies, neither company knows that he's telling the other company. They're thinking they're the only ones. But remember, it is a letter of intent. So it's not a full on commitment. So roughly 40% of the total global DRAM production reserved for only open AI. He's saying this behind closed doors, just a few months ago, by the way, we're talking six months ago, primary driver of Ramageddon price hikes. A lot of people are saying it's because of Sam Altman. The prices went up 171% like the standard 64 gigabyte DDR5 kit price went from $190 to $700 in three months because he said, I want 40% of market share because I need this business. Give it to me. We're going to be okay with it, right? March of 2026, the market learns about these letters of intent are non-binding, no obligation to buy. One of the companies, Micron Technology, the stock drops 22% and people are asking, why did this happen? Why would he make that commitment? Did these guys start already building the facilities and the products? They haven't started building the products, but they did start getting the team together to commit to building the products that Sam Altman said he needed. So guess what Google comes out and does? They write a research paper, Google's TurboQuant Breakthrough, announced March 25th software level compression algorithm for large language models reduce AI memory demand by 600% with zero loss in accuracy, instance efficiency gain, no need to corner the market on HBMs, proves that Ramageddon was a temporary market shock and not a permanent physical limitation, retail demand fatigue, consumers are tired of the high prices, so they're just stopped buying. Meaning the prices went so high, people are like, look, I'm not even buying them. Companies started selling you for a discount. They're like, look, we got to get rid of the stagnant inventory. We'll give you a 15 to 20% discount. So all of this stuff is going on at the same time. And the market is trying to find out what's going to happen with this. These guys are wanting to build more data centers. What's going to happen to open AI? What's going to happen to consumer prices, stuff that you and I are buying on top of everything else that's going on with tariffs and the war and gas and oil. Now I got to deal with this stuff. Now, here's what's crazy. While we're talking about all this other stuff, there's 1500 AI data centers that are right down the docket about to build. There's a new article that came out just this week saying half of them are on pause because the speed of chips being made is slow. There's a story of Oracle. I want to say Oracle, open AI and SoftBank. It's one of those banks that have committed to a hundred billion dollars of different facilities upwards of a half a trillion dollars just being built in Texas with Microsoft is building a $7 billion facility. I believe in Wisconsin and Microsoft quietly saying, we got to go nuclear. So what happens here? If they continue going this way and the cost comes to you, and the prices go up, guess what White House is now dealing with? Affordability on steroids. So let me get this straight. I'm dealing with gas prices and I'm dealing with tariffs and I'm dealing with AI data centers, increasing the cost of electricity, coming to a city near you, time out. Midterms are around. We got to do something about this. So either some of these guys are going to be put on a pause, pre-midterms, or they got to figure out a new way to make this work so you are not affected by it. So more and more and more companies are going to have to sit there and say, look, maybe we got to be open to nuclear energy. The one word nobody wants to use, but more and more are starting to realize this may be something we may want to look into the concerns of the past that we had. We have a lot better controls today. That's what they're claiming that they have. There's a lot more knowledge about how to handle it with all the different mistakes that we made. And by the way, you know, who's been going nuclear while America hasn't for the last 10, 20 years? China. China's quietly just select America. Don't worry. Don't build nuclear. It's not safe, but we'll build it. You guys take your time. No, no, no. Don't build it. We'll build it. And then imagine if they do, then we realize, wait a minute, they're advancing faster at the chips and what are we going to be doing? So they're competing. The market is very, very aggressive right now when it comes on to this topic. But the reality is, if you're watching this right now saying, now I realize why these chips are getting more expensive, why I'm paying more money for this. So this is the ramageddon war that's going on behind closed doors that a lot of people in business, specifically on the AI space, are talking about right now. So if you got value out of this video, give it a thumbs up, subscribe to the channel. And if you have any questions for me, feel free to connect me. This is my QR code. It's the only place where I respond to 100% of questions that get sent my way in audio. You can send me a question. I'll respond to you. And if you liked this video, we did another video about AI and nuclear. If you've never seen it, click here to watch the video. Take care, everybody. Bye-bye, bye-bye.