The Data Center Problem Is Way Worse Than You Think

[00:00:00] Speaker 1: Paying rent in Minecraft. That was the pitch of at least five different crypto companies. Back in like 2021, you used to be able to buy and rent an artificially scarce digital parcel of land in the metaverse. [00:00:15] Speaker 2: It's come to this, the virtual reality space where a digital private island can cost as much as an average American home. Prices in some areas up more than 400%. [00:00:25] Speaker 3: I'm just thinking like, okay, this is part I'm starting to not get the farther I get into this. [00:00:30] Speaker 1: We've forgotten about it now, but this used to be a pretty well-known thing. Actual real estate people were spending millions of dollars to acquire and develop land in a video game that no one played, and it was getting them good press in the mainstream news outlets. This is after Bitcoin had just gone 5x in a year, NFTs were on track to do the same, so the metaverse started looking like it could be the next big blockchain thing. Everyone wanted in, especially once Zuckerberg made it Facebook's Hail Mary. [00:00:57] Speaker 2: This can be a super profitable investment if you pick the right games to buy land in. [00:01:03] Speaker 4: Of course, we all know how that went. I got completely wiped out, but I believe that it was God that was trying to teach me a very important lesson. [00:01:15] Speaker 1: But that's not what this video is about. We'll get back to all of this later. Right now, I want to talk about these. Data centers. Specifically, hyperscale data centers. Hyperscale. Hyperscale. [00:01:28] Speaker 5: Hyperscale data centers. [00:01:39] Speaker 1: All right. Good to go. All right. So, the book you wrote is called Hyperscale. What are hyperscale data centers? [00:01:48] Speaker 6: Data centers have been around for ages, right? For decades, since the beginning of computers and our need to store a lot of information. But they have changed over time from being a room in a building belonging to a government or a military or a major corporation, to eventually being its own building, but of kind of modest size, not massive, right? To the point now where we have these hyperscale data centers that are just, you know, on a level that is hard to imagine. If you think about, say, you know, I think a warehouse, if you think about like an Amazon warehouse or something like that, might be easier to conceptualize in someone's mind. It's filled with goods that are being shipped out all over the United States, all over the world, you know, depending on, you know, where we're talking about and where Amazon is operating. But this is like a massive warehouse filled with computers, right? [00:02:37] Speaker 1: Okay, so right now I'm on a website called datacentermap.com. You should really check this out. It's a website where you can see where all the data centers are in your area. For example, we found this out while working on this video. OpenAI is building a new hyperscale data center near me, and both my creative director and writer live near a bunch already. This shouldn't be that surprising. Most of us have lived near data centers for decades now, probably without realizing it. Anything you can think of doing on the internet needs computers to process the billions of operations that happen every day and store all the data required to make it work. And that's always meant hundreds of data centers spread all over the place, especially since cloud computing took off in the 2000s. But until now, we've never really paid attention to it because data centers used to be pretty small. That is, until AI pushed data centers to hyperscale. Now, there isn't a single definition of what a hyperscale data center is, other than a really big data center. You know it, I know it. AI requires a ton more computing power than pretty much anything else. So that means that our data centers had to get bigger. But to give you an idea of just how big we're talking, please enjoy this drone footage and a statistic. There are about 10,000 data centers in the world. Hyperscales are relatively new, and they represent only about 10% of the total. But already, they account for 45% of data center capacity. This is France's biggest data center, the Paris Digital Park. It's about the same size as the country's biggest soccer stadium, which seats 81,000 people. Four times Madison Square Garden. And you can tell just how massive it is because the data center and the soccer stadium are a 10-minute drive away from each other, and they're the two biggest things on the map. Now, the Paris Digital Park is visually compelling, but it's hardly the biggest data center out there. In fact, it's not even close to being the biggest one in Europe, nor does Paris even have a super high density of data centers. In Northern Virginia, for example, there are 400 data centers either in construction or already up and running. [00:04:49] Speaker 6: And you can imagine then the demands that that creates on the community and the local area where all those data centers are being located. And often it won't just be Microsoft then that will locate there, but then, say, Amazon might come along too. Google will build nearby. You know, the smaller data center companies that people are less familiar with, they're all kind of setting up in the similar vicinities to one another. [00:05:10] Speaker 1: You've probably heard a little bit about what happens when data centers show up in your neighborhood. It's everything from pretty annoying, bad neighbor stuff. [00:05:18] Speaker 6: Like, say, in Northern Virginia, you see a lot of complaints from the community about noise pollution. So, you know, a lot of these data centers will have a bunch of air conditioning units on the roof in order to keep the building cool. And that means that there is kind of a loud noise. [00:05:37] Speaker 1: And then there's everything else. [00:05:40] Speaker 5: Data centers used about 460 terawatts of electricity. Their projections are that that should probably double to about a thousand by 2030. [00:05:49] Speaker 2: In the last year, her average electricity bill has spiked 23 percent. [00:05:54] Speaker 7: With citizen activists pushing back at the speed of construction and the often opaque deals and tax breaks given to companies by their local officials. [00:06:04] Speaker 1: Let's start with electricity. Take the Paris Digital Park again. At 130 megawatts, the PDP is nothing exceptional. And yet, it uses about the same amount of electricity as 100,000 homes. And that's not me citing someone who hates data centers or AI. That is coming from the CEO of Caterpillar, one of the guys actively trying to get more of these online. Data centers use an insane amount of power. In Ireland, the country in Europe with the most data centers, 21 percent of their energy is powering these centers. In Frankfurt, that number is 40 percent. Now, AI companies are trying to make this seem okay, and they're doing a bad job at that. But the truth is that because of them, our electricity costs here in the U.S. have gone up 12 percent in the last two years. And same thing in Europe, where the Irish now pay more for electricity than anyone else. It shouldn't be that surprising either. The CEO of Exelon, a big electricity company in the U.S., put it really plainly. More demand, less supply, everyone else's costs go up. AI's demand for power is just too much for the system to handle, meaning they both bid prices up and incentivize companies to build more infrastructure, the cost of which they then pass on to you. Things have gotten so bad that the same Exelon exec is saying that Americans will, quote, absolutely face blackouts next year because of the pressure of AI electricity demand. And speaking of pressure, hope you like not having any. Last year, residents in Georgia had a data center move in nearby, and now their water looks like this. [00:07:42] Speaker 7: This is a cold water pressure in the kitchen. Wow. And that's just from the water coming out of your faucet? [00:07:51] Speaker ?: Yeah. [00:07:51] Speaker 3: This is the current drinking water in Morgan County, Georgia, right after a data center was constructed. The metadata center was constructed. [00:08:05] Speaker 1: Look, a lot has been said about AI water use already. Everybody is throwing statistics at each other and comparing how many bottles of water it takes per prompt versus a steak or whatever. But that's the wrong way to think about this. A single data center can use between 1 and 5 million gallons of water a day, or the equivalent of 10 to 50,000 people's usage. That water needs to be clean so as not to gum up the water cooling system, so most companies just tap the public drinking water supply. On the scale of the entire country, 5 million gallons a day isn't that much. 5 million gallons for a single city, though, is absolutely enormous. Especially when you multiply that by 5 or 10, as these companies build more and more to take over entire neighborhoods. Neighborhoods that, more often than not, already had problems getting enough water out to people. And these millions of gallons are nothing when you start counting all the water you actually need just to get a data center up and running. Because it's not just liquid cooling. It's all the extra water needed to produce the electricity the center runs on. And all the water required in the chip production process up the supply chain. And all the water used for the construction materials. And all the water that trained all these AI models these last couple years. [00:09:22] Speaker 6: And so what we're seeing is, you know, a massive expansion in the amount of emissions that these companies are creating as well. And all emissions that are associated with, you know, their data centers. Such that last year, and you know, this isn't in the book because it's newer data. But last year, Google's emissions went up by 18%. Microsoft's went up by 25%. That's just in one single year. And these are companies that back in 2019, 2020, were talking about doing net zero and reducing their emissions and all this kind of stuff. That has really been thrown out the window. [00:10:00] Speaker 1: Howdy, friends. Quick interruption to give you an important update on the Mail Club. As always, I want to make sure I respect your time. So if you're not interested in hearing about our new project, feel free to skip ahead to this timestamp. You can always come back later if you're interested. For those of you who stuck around, thank you so much. For the past few months, my wife and I have been running and refining the Number 3 Post, a snail mail club designed to help build community and solidarity among regular people across the country and abroad. The big updates for this month are gift subscriptions, which a lot of people have asked for, and our holiday policy. The gift option is a prepaid 12-month subscription you can send to a friend, or an enemy, I suppose, along with a little PDF explainer of what the club is about. So if you want to sign someone up as a Christmas gift, you can print out the PDF so they have something physical to open while they wait for their first dispatch to arrive in January. Make sure you follow the instructions on the gift subpage so that the confirmation email goes to you, not the recipient if you want it to be a surprise. It's just the two of us, my wife and me, managing this whole operation, so please do make sure you read the FAQ and all the information on the gift subscription page so we're not having to respond to a bunch of confused emails over the holidays. Speaking of holidays, the second thing is our holiday policy. Every year we're going to pause new signups from November 21st through January 1st, so Kelsey and I are able to spend time with family. Everyone who has subscribed before the November 21st cutoff date will receive the December and January dispatches as normal. We just won't be processing any new subscriptions until the first of the year. So, long story short, if you want to make sure you get the next couple dispatches, make sure you're signed up by November 20th. Anyway, with all that news out of the way, for anyone who hasn't heard of this stuff before, snail mail clubs are pretty simple. You sign up for one you like, then every month you get a physical envelope in your physical mailbox with a bunch of fun stuff inside. For the number three post, every dispatch follows a theme, like September's theme was revolutionary optimism, October's was the horrors of capitalism, and so on. And you'll get a hand-signed letter and a collection of analog goodies. The contents of the dispatch will vary month to month, so you'll see everything from art prints to stickers to journal prompts, recipe cards, curated watch lists, and lots of other fun inclusions. We've got an amazing roster of artists ready to be commissioned for themed work, and we've got some really cool pieces already in the works for the next few months. We're working hard to make this as collaborative an experience as possible, so every dispatch includes a little community bulletin with the artist's social media info, and starting in November's dispatch, we'll also be sharing some positive news from subscribers from around the world. Also, if you're bummed you missed the first few additions, you can purchase past month's dispatches from the archive on our website. Okay, I think that's all the big stuff. If you want to see some more behind-the-scenes info on the technical aspects of the club, I've got a video on my personal channel that goes into more detail. To give us time to prep and pack all the envelopes, the cutoff to receive the next dispatch, technically the next few dispatches since we're pausing new subs for the holidays, is again, November 21st. So, at 11.59pm, November 20th, you've got to get your sub in by then. If you miss the boat, subscriptions will reopen on January 1st. We're really excited to keep growing the number three post, and I hope you're excited to check it out. Alright, back to the show. Look, at this point, I think you get the idea. I could keep going all day talking about how all these new data centers are making the air unbreathable, and stopping cities from building new housing, promising hundreds of jobs only to deliver, like, ten, once they're up and running. But at this point in the video, we need to talk about the why. One thing that shocked me reading Paris' book was this one statistic. 40 to 90% of all commercial data is used just once, if it's even used at all. Look, there are a lot of statistics in this video. There are a lot of statistics in the book. Nothing has thrown me for a loop, like, thinking about how completely pointless this all is. And we kind of knew that already. We can see how desperate AI companies are to make all this have a point. There's an AI button in every app, on every website. Every industry is trying to find its very own powered-by-AI feature. And so far, all it's done is made people psychotic, and given the world's first trillionaire his own personal CSAM machine. I know we here at Second Thought don't use AI. You're absolutely right. A better way to say that, and I appreciate the pushback, is we rely 100% on human invention. And, look, jokes aside, that doesn't mean there are zero uses for AI. But there simply is not a multi-trillion dollar industry here. And yet, this is the only thing supporting American economic growth. It's underneath everything, morphing into this horrifying pillar of the global economy under financial capitalism. We're going to go back to the Paris digital park for a second. The PDP isn't run by Amazon, or Meta, or Google. It's run by a company called Digital Realty. Digital Realty gets its name from being, on paper, a real estate company. It leases tiny little chunks of real estate, i.e. server space, to companies like Amazon, and Meta, and Google. For tax purposes, that means these companies are something called real estate investment trusts, which, to put it simply, lets them do tax evasion legally. But more importantly, it turns these data centers into yet another way for investors to speculate on real estate in an industry that, before we even touched realty, was already a tangled mess of speculative investments. And here's how we get back to where we started this video. We started looking into this mess of data centers and AI, and I couldn't stop thinking about the metaverse. Feels like we've already seen this story play out. Entire country's electrical grids being eaten up by GPUs. Zuckerberg telling us this new thing that uses a thousand times more power is going to be the future. Real estate companies finding shady, speculative investment vehicles. Wall Street predicting trillions in windfall profits. And then, nothing. It all came crashing down. Not because there was no use for the metaverse. It was a video game. Those are great. I'm sure it was a lot of fun for the couple dozen people that played. But it could never live up to the trillions of dollars expected from it. And it was never worth all the resources required. So instead of being this huge, important thing we still talk about, it became a way for scammers and the already rich and powerful to become even more rich and powerful, while ordinary people lost their entire life savings. But we normal people didn't learn our lesson from the metaverse and the crypto bubble more generally. But someone learned. Crypto taught our billionaires that no one could stop them from running their insane little pet projects. No one could stop them from bleeding the public well and the energy grid dry, from polluting the waterways and clogging the internet with mountains of useless data. And they realized that if they just picked something a little more credible than we're all going to live in the computer someday, maybe this time they could even get the government on board. They were right. And now we're living with the consequences. But it doesn't have to be like this. Towards the end of our call, Paris shared something with us that I wanted to make sure we included. [00:17:26] Speaker 6: As part of this book, I went down to New Zealand and met these people at Tehiku Media. And it's basically a Maori media organization, which is the indigenous people of New Zealand. And they have been kind of broadcasting in radio, television, online, in Maori language for decades now. And they wanted to make sure that they were using technology to try to promote the Maori language. But they didn't want their language to be kind of subsumed by these major U.S. tech companies. They wanted to make sure that these were tools that actually cared about the language, that actually helped people learn the language. And so they developed like, you know, kind of a translation and language learning model. You know, that made sure that the pronunciations of Maori were accurate in a way that they often didn't see in these kind of versions created by the major tech companies. And, you know, they were kind of working with a lot less computation. They were very limited in the amount of data that they had. And so they had to kind of be selective about what they used and how they used it in order to train the model. [00:18:35] Speaker 8: It's not just about speech recognition. It's about people, place, and identity. And we want to preserve the economic opportunity for us to lead work that is important and empowering for our own community. [00:18:49] Speaker 6: And for me, this is a really good example of how things can be different, right? [00:18:56] Speaker 1: Right. Cases like this show us that there can be uses for machine learning and AI. Legitimate uses. Things that take, you know, a couple server racks, not entire blocks of GPUs and Olympic swimming pools of liquid cooling every minute. Things that directly help people and, at least in this case, don't make much, if any, money. Things that don't gamble entire economies and our climate on wiping out half the labor force. We deserve that. And if anything, all these protests blocking data centers prove it's what we want. Enough with the tech billionaire bullshit. Things that don't make much more of a good idea.