Data Centers: A Necessary Evil…Kinda

Hey, everybody. Peter Zeihan here, coming to you from Colorado. And today we are going to talk about data centers. You may have noticed that there’s a bit of a political brouhaha about them, the idea that they absorb a lot of electricity and a lot of water and are generally making life loud in rural America. There’s something to all of that. I want to talk about the movement against data centers and what data centers actually do, because there’s a lot of misconceptions there. 

So first, the data centers themselves. Most of the brouhaha has to do with artificial intelligence. The core of any AI system is something called a graphics processing unit, which is a chip that is normally involved in personal computers to handle graphics processing. Ergo the name. When large language models, which are the current iteration of AI that everyone’s all gung ho about. When they got big a few years ago, the people who are doing AI were looking around for preexisting types of hardware that were good at doing multiple types of processing of data in parallel, and GPUs were the only thing out there. So most AI data centers are either going to use GPUs directly or something that’s descended from that lineage, that works very similarly, but at the core is a relatively small chip, in most cases about the size of a postage stamp, integrated and packaged in different ways. But the problem is, as anyone who has a gaming console will tell you, or anyone who runs a multiplayer computer system for games that has a lot of graphics, is that these things get hot. So normally they exist one at a time under a fan. They run for a few hours and then they turn off. But in a data center, they’re running 24/7 hot, and they typically have to be hooked up to a coolant system that for most systems is either air or water based. And those systems are running 24/7. And the return, the hot air or hot water that’s taken away is typically more than 180 degrees. So the cooling system is very power intensive, whether it’s air or water. And if it’s water based, it’s a question of whether it’s using an older system or newer system. The older systems have evaporation towers to cool off the heat. That’s where your water demand comes from. Your newer systems are more closed circuit, but you still have to fill it up with several millions of gallons of water on the front end. And then of course, there’s leakage. So there’s a constant demand, not as much as the older systems, but still very, very real. 

Anyway. Writ large, over 80% of the water use and over half of the electricity use for a data center does come from cooling, and there’s not a lot you can do in the short term to change that. I mean, you can move away from the evaporation towers, of course, but that requires different sorts of hardware and generally is incorporated into the system on the front end. They are working on new types of GPUs that incorporate the cooling into the actual chip architecture. But the first of those is not expected to come out this decade. The problem is that there’s a lag from the point that you design a chip, and then you build out the logistical system that is necessary to incorporate the chip into a data center. Years. Simply fine tuning the production cycle of a fabrication facility takes 6 to 12 months, based on the piece of hardware you’re trying to do, and then it is somewhere between 90 days and 180 days before your first production run actually generates your chip. Because once you have your silicon disk, you have to etch it and bake it and cool it dozens of times. That’s not a day process. That’s a multi-month process just for their first run. At that point, you then take your finished semiconductor wafer and you break it up into the various components. And those get incorporated into various intermediate products. That’s another 60 to 180 days. So from the point that you finally decide you have your perfect design to the point that you have enough of those finished products to put into a data center, you’re talking 5 to 7 years. That’s not going to happen before 2030 at the absolute soonest. 

So we’re stuck more or less with the environment that we have today, which means a lot of power, a fair amount of water, and the demand is very real. Some people say that the Chinese are trying to manipulate public opinion to go against data centers, and no they’re really not. The Chinese are not nearly as good at that as the Russians are. And the Russians don’t care about this topic because they can’t make any of the chips, and they’re not using any of the chips because they don’t have an AI sector, and their data centers are pretty piss poor, anyway. The Chinese do have a bot farm that has something like 200 million accounts, but only 200 of those accounts have been assigned to participate in the anti AI space. And almost all of the followers of those accounts are just one another. The Chinese just don’t understand how to manipulate a free society in the way that the Russians do, because the Russians have had a century of practice, and so it’s just really a non-factor. Most of the opposition is real. That doesn’t mean that it doesn’t cross over into other unrelated sectors. So, for example, there’s a lot of people in the green movement who see any sort of fossil fuel as a sin and data centers, because they have to run 24 hours a day, can’t just can’t use solar and wind for an appreciable percentage of their demand because it’s not reliable enough. And of course, the sun doesn’t show at night. Now you can put in batteries into a data center, and all of them have batteries, but it’s mostly for power smoothing, so there’s no interruption in current as opposed to store solar or wind power, because just the fields you would need would be immense. So most of them have to run on nuclear, which is not an energy sector in the United States. As expanding people talk about small modular reactors becoming a thing, but small modular reactors are somewhere between 5MW and 300MW, with most of them probably around 50. By the way, we haven’t built a prototype yet, so that is all very perspective. And let’s say you have 50 a 50 megawatt SMR and you have a one gigawatt data center. Okay. So you’re going to put you’re going to put 20 nuclear reactors into a single data center. The math just really isn’t there. If you don’t use nuclear, your only other options are coal or natural gas. And so you get these green factions that are opposed to data centers because of the type of power that they undoubtedly are going to have to use. There is not a single data center on the planet that runs on solar or wind at all, and there won’t be ever, unless our nature of what solar and wind can do changes are a bit, a lot, radically. That’s also a problem for a different decade. 

Okay, so that’s the front end. Now let’s talk about what the data centers actually do. Just because opposition is real doesn’t mean it’s not without consequence. So let’s break this into three big categories. First, the things that you might not have thought of but are probably not a reach for you. Big data processing and obviously artificial intelligence in that category. But so is graphics processing. So is any type of research. So is climate modeling. It used to be that you would have a mainframe or a series of servers in your company, and you would use them to do the big data heavy work, and it would take hours for them to crunch the numbers. That has all been offloaded to data centers now. When people talk about the cloud, the cloud is 100% data centers. So whether you’re saving your own files or you’re doing a climate model, or you’re in the energy sector and you’re trying to figure out how to frack a well more effectively, or are you doing biological research, every bit of that isn’t just stored in data centers. It’s processed in data centers. And it has been for years. The second category are things that you probably realize already. Your email. Anytime you use your computer to send or receive any information, whether that’s for an insurance claim or for personal use, all of that is going to be using a data center. So for whatever reason, we don’t have enough data centers. All of a sudden you have to do all your computing and storage local, and that’s an order of magnitude more cost for you. One of the reasons why laptop computers have kind of faded in terms of general use is a lot of the tasks that they used to do have now been offloaded to the data center sector and aren’t being done at all. So think of all the accounts that you have, for example, for your bank, for your fire insurance, for your home insurance, for your mortgage, for your checking account, for your savings account, for your debit account, for your credit cards. All of that is done almost 100% in data centers today, which means we get to a third category. The things you probably haven’t thought about. Anything that can be turned into zeros and ones now largely has been. So. For example, it used to be that you had a phone and you had a phone line. Well, once we switched over to cellular communications, your phone handset turns your voice into a series of zeros and ones and sends it off to a tower. That tower sends it to a data center. That data center routes it to a different tower. And that tower then beams those zeros and ones to whoever the receiver of the call is. Basically anything you do on your phone, no matter what it is, including voice. It’s now processed through a data center. So that’s your entire social media history and participation. That’s every app you have, regardless of what it does. Now it is something that has been digitized, and you have to apply that thought process to everything in the economy. If you go back to, say, the 50s and the 60s, well before data centers, everything that involved interaction among companies was done on paper. So just to pick one example, your grocery store, every day, they would go through and do inventory and figure out what they needed to order, and they would put it down on paper, and that paper would be physically handed off via the mail to the person who would then fill those orders and send it back and stock the shelves. By the time you got to the 70s and 80s, that paper had been partially digitized via the fax machine, but everything still had to be done manually, and something like 10% of the American workforce at its peak was involved in this sort of data processing, which was really just putting pen to paper and then sending it off in some way. So once we got to the digital age, we started transferring all of this as the zeros and ones. And once faxes started to give way to email, we started sending them as attachments or now eventually just an app process. And so anytime your at a grocery store and you buy something, it no longer has to be manually counted. The system itself says, okay, you just bought two things of yogurt and a thing of eggs that goes back to the back room where they tallied up moment by moment. Where they set thresholds. We’re like, okay, if we keep going at this pace, we’re going to write out of yogurt in three days, in eggs in four days. So let’s preemptively place an order that’s all automated. It’s all done through data centers. 

So if you want to go into a world without data centers, you have to consider what you’re giving up. Everything that you can now do at home. You will probably have to go in and communicate with a real person. Anything that is digitized. You’re gonna have to do with paper. And so for things like, say, grocery stores and anyone who has to deal with inventory, you’re talking about needing literally millions of workers to do very, very, very painfully low value added jobs where minimum wage wouldn’t begin to cover the cost of those people. So you’re talking about slower service, like by weeks, you’re talking about less product variety, you’re talking about longer lead times and basically condemning 10 million Americans to have jobs that have no value add. So do data centers have negative effects? I’d never try to argue against that point, but keep in mind that without them and without their ongoing expansion, you really are talking about dialing back the clock to where we had a far lower standard of living and far lower personal incomes.  

I will note a couple of things that are exceptions. Health insurance is still done the old fashioned way, which is one of the reasons why we all hate it so much. So keep that in mind. If you want your experience with your insurance company to become the norm for your entire economic life, then a world without data centers is exactly what you’re after.