What They're Not Telling You About Data Centers

[00:00:00] Speaker 1: An acre of land is 66 feet across, 660 feet long. It is the unit by which all things that happen on it are measured. A farmer might make $1,200 per acre of farming corn. Raise cattle instead and one acre may produce $3,000 in a good year. Rent it out to someone who wants to build a windmill on it, $5,000. But choose to use that acre instead for an AI data center, that number goes way up. About $10 million per year. A number that until very recently could only be claimed by the land in prestigious areas of the most prestigious cities. New York, Paris, Tokyo, London. It wasn't until AI was placed in front of the words data center that a shift occurred. A shift that has taken just three short, fast-moving years. Moving data centers from something nobody gave a passing thought about to something else entirely. Something more profound than electricity or fire. Compute is going to be the currency of the future. Meta will invest hundreds of billions of dollars into compute build-out. Infrastructure we call factories, AI factories. [00:00:58] Speaker 2: It's the largest AI infrastructure project by far in history. Colossal data centers, very, very massive structures. [00:01:06] Speaker 1: Bridges, power plants, skyscrapers, stadiums, docks, railroads. In the long history of the U.S. building major things, just two come anywhere close to the backlash being felt by AI data centers right now. There was the U.S. interstate system in the 1950s, a major fight over what land exactly all this new pavement would be poured over, how close they would be to people's everyday life. People really didn't want a highway running through their neighborhood. Then there was the nuclear decade of the 1970s, grinding to a halt as major protests erupted at every site that wanted to break ground on a new plant. People were scared nuclear plants might kill them. [00:01:39] Speaker 3: Looking at the alternatives for energy, nuclear fission must be stopped. [00:01:44] Speaker 1: Go out and ask a thousand people today, and the amount of people who say they don't want a data center next to them is far more than the number who said they didn't want a nuclear plant next to them back in the 70s. This was duck and cover drills in school, countries testing ever bigger nuclear weapons. How were nuclear plants more popular at that moment in time than data centers are now? What exactly is this fight over data centers actually about? This video is sponsored by ZocDoc. More on them later. The internet had a rule, and the rule was a problem. The backbone of the internet, the long wires between cities, were owned and paid for by the government. Meant for universities and labs, you weren't allowed to use them for business. You could send a professor's email across it, you could not send a customer's. Zero commercial activity allowed, which was fine, until people showed up on the internet who wanted to be customers. And a few businesses, eight to be exact, showed up who wanted to serve them. If a customer of one wanted to email a customer of the other, the only wire between them was the one they weren't allowed to use. So the two companies had to physically touch their wires together somewhere. A line coming out of one building, a line coming out of the other, and physically meeting somewhere. That was the only way two customers could talk. It's kind of crazy to think about how they solved it. Just a couple of guys, one from each of these first companies, ran a wire from the street, one for each company, and into the first meeting place, which was just a plain office suite in a building in Tyson's Corner, Virginia. Two companies at first, then another, and another. Within two months, eight offices were connected together, each company running a line from their building and into this room. Once more companies joined in that room filled with cables, they moved across the street to a parking garage. Actually, a tiny room off the garage of an office building on Boone Boulevard. One server, one plug. That was it. Nobody worked there. Nothing was computed there. It was a place for a handful of companies to touch their wires together. Yes, I said it again, and I'll probably say it a fourth time. This was the first public internet and the world's first data center, a room whose whole point was simply a meeting place for other buildings around it to touch. By the time the government shut off the wire companies weren't allowed to use, around half the world's internet traffic was already passing through that little room in that garage in Tyson's Corner, Virginia. Visit Tyson's Corner today, and it will appear as any town USA. There is nothing that would alert you to the fact that this place, this little room in a garage, was the spark that created what is 30 minutes down the road. The largest collection of data centers the world has ever seen. Data center alley. Data center after data center after data center. Picture a super Walmart, all the aisles, how much time it takes to walk from one end to the other. In this 20-mile area, there are 250 super Walmart's worth of data centers, 52 million square feet of floor space, where servers let millions of companies touch their wires together. The center of gravity had formed. There was a sense, a lot of it, in being near the hub where every company around the world was exchanging information with each other. The average person will not wait more than half a second, Google said. So if you want us to show people your ads, you must be located in this area. When a person lands on a website with ads on it, we'll send out a broadcast and a profile of who we think this person on the site is. You have 100 milliseconds to analyze that profile and tell us how much you're willing to spend to have us put your ad in front of them. Receive the broadcast in San Francisco and 35 milliseconds are eaten up waiting for it to arrive. Then another 35 milliseconds to send it back. 70% of your time was gone before you had even looked at who the person might be. Analyze the profile better than everyone else. That was how you made money. Your bids would be better. Every millisecond of time spent sending things back and forth mattered. A lot. So not only did dozens of companies put servers in the same area as Google, they paid extra money to put their servers in the exact same building as Google. This is how all the free stuff we use on the internet works. And it is the most intense, highest volume compute event on the planet. Type in a website, hit enter, and Google fires off an auction billions of times per day. Google search was a great product, better than the rest. But this auction machine is what made Google into the company it is today. For this, and many other interactions like it, once the internet really started to take off, it was not an option to be somewhere everyone else wasn't. If you've heard of Loudoun County, Virginia, that it is the richest county in the United States, this is the place they're talking about. Now it is important to point out three things here. One, the population here is not rich because the data centers are here. It's a suburb of Washington, D.C. Number two, the data centers are not here because the population is rich. They're here because they needed to be here. Number three, the local government here is rich because of taxes they put on the data centers. Three different things, just so there's no confusion. The who's who of the internet all had or have space here. AOL was a few miles down the road. AWS started here in 2006. The first region you could rent spare compute from was US East 1, something every software engineer knows very well. It is the default setting for every cloud provider. And it routes right to Northern Virginia because that's where everyone already was. Window lists, long, low, nobody protested. Few knew they were there. A data center in 2015 had every photo you had ever taken already. Netflix keeps copies of every single movie in its library in dozens of data centers around the country. WhatsApp messages all flow through these. Data centers are the internet. And until very recently, they drew about as much attention as the warehouse next door holding lawn chairs. Then someone decided to put a lot of a very different type of computer inside. Is that blinking? No. I thought I charged that last night. How long is this going to take? 16, 16 watts? What is a watt? You know what a watt is? About one hour. All right. We are good, though. Battery is back. No blinking. Nice. Okay. So plug in your phone. It's about 10 watts. Click the TV on, 100 watts. Turn on the microwave, 1,000 watts. The labels on all the electronics that we have around where it says stuff about electricity, usually in watts, those labels are not really made for you and me. It's the maximum amount of electricity this thing might ask for in one second. Needed by electricians so they have a general idea of what size wire to run to what part of the house. Each run to a different part of the house, branching off the main wire coming from the street. Each of these runs, the wire chosen for them, can only handle so much electricity flowing through it. Try to run your washing machine and dishwasher on a small size wire and you'll trip the breaker, which is why the electricity always goes out in that one section of the house. Hey. Even though you've told them a dozen times we don't put the big space heater in that room for a reason. It's not even cold. Put a blanket on. It is also the reason your washer, dryer, oven, and dishwasher are plugged into those big outlets, connected to wires large enough to handle the extra electricity they need. This system of different wires, different plugs, different circuit breakers, but ultimately run back to the power plant that makes the electricity, it's the same for all the factories, office buildings, stadiums, everything that uses electricity. It's just scaled way up. Wires are bigger, breakers are much bigger, the transformer that takes it off the main line and channels it into your building. Far bigger. An average American home around the clock averages about 1,200 watts, which is a little more than a microwave. If every home in New Orleans sat down and hit play on Netflix at the same time, 10,000 watts of servers would be running to deliver all of that video. It's about eight homes worth of electricity. But then we started plugging in these new computers, the computers that would force all of us to reckon with what exactly would happen when instead of servers sending us Netflix videos, we installed the special servers that allowed us to ask for almost anything to be created on the spot and sent back. The movie coming from Netflix was made years ago. Netflix compressed it, copied it a few thousand times, and put the copies on servers that sit inside a data center, probably within a few miles of your house. Press play on saving private rhyme. The signal you sent goes to a box that finds the file and pushes it back down the wire to you. That is it, every second of video you've ever watched was finished before you asked for it. Ask an AI for a video or anything else for that matter and nothing exists yet. There is no file, there is no template, there is an empty void that must be filled. If it can't figure out how to fill the void, there is no reason for it to exist. So think about what a video is. It's just a lot of still images strung together, frames, playing in sequence one after another. For AI to make a two second video, it has to make 48 different images. Start to fill the void by creating the first frame, but being video, the next one must be different than the first, but it can't be too different. It needs to make sense. So it checks the second one against the first, then thinks about how that might make sense against the third, makes the third, does the same check, the fourth, on and on. Each one a guess saying, I think I'm doing a good job filling the void with something this random human who gave me this task will like. Each frame 2 million pixels built from everything the model has ever seen. This is the whole difference between the old data center world and the new one. The chip that delivers, it's called the CPU. It's the one in your laptop, eight cores, maybe 16, very fast, very good at doing one thing after another. A kitchen with eight outstanding chefs all working on their own dish. The chip that creates, the void filling chip, that's called a GPU. Something like 17,000 cores. Each one is slow and dumb, not even close to a chef. But all 17,000 of them are working on the same frame at the same time. 17,000 cooks working on the same dish at the same time. And it is here that we arrive at what might be the true sticky center of what the data center debate is actually about. The Netflix servers were 10,000 watts. Okay, everybody is watching those movies. These machines that fill an empty void require 120,000 watts. That's 100 homes worth of electricity. Leading to a very tricky question that has resulted in what has to be something many of us have now experienced. Yeah, you just type in the box and it'll do it for you. [00:11:32] Speaker 3: What do you mean it'll do it for you? [00:11:34] Speaker 1: Anything you want. [00:11:35] Speaker 3: I don't know what that means. Just ask it something. [00:11:38] Speaker 1: All of this electricity, and what an interesting piece of the puzzle electricity is, which we will soon see. But all of this electricity, all of this land, all of this money, and when the average person has the audacity to ask something like, well, what do I get from this? The best response is, hmm, I don't know exactly. Something cool. It can write emails. Either that or it might kill you. But if you don't mind, I'd like to build another city-sized box across the street so we can create more cool stuff. This is perhaps the ultimate example of a Lulu. We'll come back to that in just a second and turn to me, finally booking the appointment I kept pushing back, thanks to the sponsor of this video, ZocDoc. I needed to see a dentist. Not an emergency, just overdue. And every time the thought showed up, I came up with a good reason to wait. Got to take a look through this edit, a lot of reading to catch up on, and one more research pass until finally, it is late and booking feels like a whole new thing I don't have the brain for. Later becomes next week, which becomes next month. The real problem was much less the actual visit and more the making of the appointment. Open a browser, hunt for a dentist, check insurance, read reviews on three different sites, call during business hours, sit on hold, wait for a callback that lands while I can't answer. That version of the task is why people like me stall. So I stopped stalling and opened ZocDoc on my phone. Local doctors who take my plan, real patient reviews next to the names, and actual appointment openings, I could see and tap. Compare a couple, pick a time that fits, book it. The thing I'd been protecting my whole day from took a couple minutes between tasks. If you've got an I'll do it later appointment sitting in the back of your head, try it from your phone at ZocDoc.com forward slash Maxonomics. Local, in-network, real openings, booked without turning your afternoon into hold music. So thank you to ZocDoc for sponsoring this video. Now let's get back to it. Do you know why modern prison cells don't look like this anymore and instead look like this? Inmates can throw stuff out. One variant of that is a fluid assault. A guard walks by, one of the inmates inside has acquired fluid of some type, they throw that fluid on the guard. an assault by fluid. These are the kinds of doors the Dakota Territory Prison have. Going all the way back to 1881, this place has stood, barely changed for nearly 150 years. Its cells are 30% smaller than the industry standard. There are no windows or line of sight to the outside world. Many of the cells, showers, are in one long basement room where 40 men shower at the same time with two guards present and there is exactly one stairwell that allows you to exit the room. It's a 19th century prison. Probably a great one for the time. But as time has gone by, there are many things that are wrong with this prison. Not just from the prisoner perspective, but from the people who have to staff it. It is impossible to get it up to life and fire safety codes. So bad that a federal judge ruled that it did indeed violate every prisoner's 8th and 14th amendment rights. That was all the way back in 1984. And that is about as far as the whole thing has gotten because prisons are Lulu, a locally unwanted land use. Finally, in 2021, the state said, no more. This has to be fixed. And a judge agreed. A prison, like a jail, is a necessary government function and it must be located somewhere. The state starts offering money to people who own land. $70,000 an acre and a place where land goes for $10,000 an acre. One said yes, who thought he was selling for a fairground. As soon as he found it was for a prison, he flipped no. We'll build it on the outskirts of that place. What about that place? No, no, no, no. Protest after protest hopping from one place to the next pressuring the mayor and politicians to say no. 40 years to get from judge to breaking ground which finally happened April 2026 on a section of land next to a concrete plant. Landfills, airports, power lines, highways, many of the things that we all use or benefit from indirectly are exactly the things that absolutely no way do not build that near me. They don't pass the kitchen table test where you sit down with your family and you talk about the new landfill that they want to put in right over there and you all acknowledge that yes, our garbage will go there and everyone's garbage needs to go somewhere. The value of our land will probably go down there. It's probably going to be traffic, maybe even a smell. Despite knowing it's something thousands of other people will use, it's so easy to hit no on that. Data centers do not pass the kitchen table test either, yet anyway. But for a reason most people haven't really considered or don't want to put into words. All the benefits of a state prison, it all stays in the state. Bad people from your immediate area are taken away and put inside explicitly so they cannot harm you. That is a public service the person who lives next to the prison can clearly see. Even if they don't love the laws or how the legal system works or even that the prison is there, the prison, it is built to serve them. Just ask it something. Like you're speaking to a person, just like you're speaking to me. [00:16:20] Speaker 3: Like what? Okay, okay. What does a data center do? [00:16:24] Speaker 1: Every two seconds, roughly. Depends on exactly what you're asking and what you're doing. One million people are able to send something to one of the mega AI data centers being built and it will fill the void and send it back. Back is wherever. There's a small chance it's the house a mile away from the data center but probably not. There are a thousand people in some of those towns. Most of them don't even use AI. This is a much harder kitchen table test. A local fight over where the prison should go, brutal, but it's honest and easy to understand. A data center breaks the old moral geometry of this is our mess so this is our dump. A data center is not matched to its users. It is matched to four inputs power, water, fiber, and a deal. That deal part, the details of it, the kitchen table is where that deal is made and the people around it are focused on exactly two things. [00:17:09] Speaker 3: This is the current drinking water in Morgan County, Georgia. [00:17:14] Speaker 2: Researchers say the amount of water used by data centers can be up to millions of gallons during hot days. [00:17:19] Speaker 3: The house was built in 2005 and the water level in our well has dropped 15 feet. [00:17:24] Speaker 1: The thing falling just behind first on that list, number two, water. Data centers get hot. Historically, the cheap way to cool it is the same way we as humans do. You sweat it out. Put water into the building, put the heat that comes out of the chips into that water, let the water evaporate, the heat leaves with it. Running that loop all day, a big data center running a particular type of cooling system can go through a million gallons. If you put that data center in a dry county with a small utility, that is a number the neighbors are going to question and if they can't get a clean answer, that might be the bigger issue. Like at this site in Oregon, Google was using more than a quarter of the town's water in the city. It wasn't even Google went to court to keep that a secret. That's the part people are actually mad about. It appears to be much less about the gallons, much more about not being told. If you back out and look at water usage by activities that are in a house, taking showers, doing dishes, drinking it, this is how it breaks down. Farms are the biggest consumers of water in the U.S. almonds, pistachios, golf courses rank up there and this little box down in the corner is data centers. Doesn't mean you can put a data center anywhere. It doesn't mean you can be reckless but actual documented harm to residents, meaning someone lost water or had it degraded because of a data center is a very short list. Of the 5,400 data centers, the cases that get named over and over are the same handful. The one that goes around is Newton County, Georgia about muddy colored water. It's not the city water everyone uses. It's four private wells and by all counts just to get all sides of the situation including the county commissioner, the cloudy water appeared many years ago, years before Metis started building its data center. But either way, the industry reads the same polls everyone else does. New buildings are being designed to run closed loop which means they use the same water for 10 years. Water goes in one time and then it just recirculates or a new type of air design system or no water at all because a million gallon headline in a dry county is going to cost more than cooling savings. But the number one thing what 77% of people are focused on it is the number that shows up at the top of every poll poll after poll all political parties. The number one thing being chatted about around the kitchen table is this going to make my electricity bill go up? [00:19:31] Speaker 2: Announcing the formation of Stargate the construction of colossal data centers. [00:19:36] Speaker 4: NVIDIA announcing that it will invest 100 billion dollars into OpenAI. That project will use about 10 gigawatts of power. [00:19:44] Speaker 1: That's a lot of gigawatts. 1,000 watts how much a microwave pulls is 1 kilowatt. 1,000 kilowatts which is 1,000 microwaves is 1 megawatt. So naturally we just keep the pattern going. 1,000 megawatts or what is 1,000,000 microwaves is 1 gigawatt. Or to make it a little more concrete imagine if every home apartment, house, condo whatever all turn their microwave on at the same time in a city like Chicago. It's about 1 gigawatt. Add up all the data center proposals and it's very easy to start wondering alright that's a lot of microwaves being turned on at the same time where is that going to come from? Across the entire U.S. this is how much electricity can be generated this year. Of that there's a decent amount that won't be used surplus. By 2030 the U.S. barely generates more still surplus though. By 2035 there's less electricity available than 2030 and the surplus disappears. So where is this electricity going to come from? People refer to the grid as though it's an interstate highway system a connected web where each part connects to the whole? It is not. The United States of America all 50 states right as they should be but this is not how the electric system works tidy inside of each state. This is. This is the electric states of America. When you turn your microwave on there's nothing that says the electricity that flows into it comes from the power plant closest to you. It might or it might come from any of the other power plants inside the bounds of the electric state in which a person lives. A plant that is fired up at that moment and making electricity when you hit start flowing out of any of the hundreds of plants in each zone through transmission lines into local lines into your house just as movie night is about to start. As electricity took off in the 1800s power plants were small and local. They connected their own lines to their own customers. Power plants though started to be made bigger and they were made more powerful. It started making a lot of sense to have a few hundred big ones instead of thousands of smaller ones. So they naturally started connecting one set of wires to another. Small utilities and power plants combining to make a bigger one. It was more reliable. The electricity could be made cheaper just like it wouldn't make sense to have one company that made gasoline in each town. Instead a few giant refineries make gasoline for the whole region. Gasoline and diesel are exactly like this actually. If you look at the map it's the same kind of weird map for oil storage and refining. This weird map is why it doesn't make any sense to look at how much total electricity the U.S. can make. What matters is how much each region can make and whether each region has more than enough or not enough. And this is where things get interesting because just like the 50 individual states of the U.S. each one its own little experiment the electric states of America are also experiments. They do not all do things the same way. How decisions to build more power plants are made what kinds of plants to have all the way to what to put on our bills that we discuss around the kitchen table. Most of the U.S. population is in a region just like the one that hosts the world's most data centers. No one company owns every part of the system. One company owns the power plant. Another the big transmission lines. Another one owns the lines in your neighborhood and sends you your bill. And yet another organization is responsible for getting all of those companies to cooperate and to run hundreds of auctions that decide the price you pay for your electricity. Those auctions work like this. Auction number one. Every hour every plant joins an auction saying the minimum amount of money they'll sell their power for. Bids roll in. The organization running the auction looks at all the electricity being asked for and what their historical models say to expect over the next hour. They sort the bids by amount. The last bid that covers all the power they expect to need that is the clearing price and that price becomes the price for every power plant that submitted a lower bid. In other words if I say $5 and you say $10 and we need your electricity to cover the amount we need to deliver to customers your bid is accepted. You get paid $10 I also get paid $10. Auction two. Same type of auction but this is a single major auction that happens once every year asking a simple question what do we have to pay you to make sure you are available on the worst hottest summer afternoon when electricity demand is at its highest two years from now the people holding the auction look at how much electricity they think they'll need a forecast looking two years into the future how much do we need how much do we think that will grow how many plants are going to retire how many are going to be built add up the numbers and the price that comes in from the plant that gets the total amount over the forecast that's what every plant gets paid and that amount is split across everyone's bill it is an insurance payment a just-in-case to make sure the lights stay on on the worst day if we mosey on down to the south things get much easier only two areas of the country live in this type of system the southeast and a big swath of the west these areas are much simpler one company like the southern company in Georgia owns the plants the wires and sends you the bill what is called vertically integrated once per year they publish a big forecast to determine how much electricity they'll need to keep their customers happy the following year southern goes to the george government and says here's what it's going to cost to keep the lights on next year please let us charge that a commission looks at it says yes no or not that much and whatever number they negotiate is what customers pay there's no hourly dog fight no two-year reservation fee one company one hearing one break a little farther west in texas though this region is unique in the entire country it is the only region that works like this as simple as can be and interestingly note that the borders of this region are entirely within the state borders a very intentional decision because that is the only reason they're allowed to do it the way that they do if you've ever heard the term interstate commerce that is what texas does not want to trip connect to a system that serves customers outside of texas and the federal government is then allowed to put rules on texas about how it runs its system and texas does not want those rules the system would become much more complicated than how it currently works which is just a round-the-clock auction we need this much electricity how much will you sell it for it and a plant only gets paid when it actually makes the power that is it most hours that's cheap on a brutal afternoon when every air conditioner in dallas is running full blast or a freak winter storm like a couple years ago the price of that hour is allowed to go as high as five thousand dollars all that to say it's not the same everywhere three different systems will lead to three at least somewhat different outcomes that's not to say any of them solves the demand for electricity entirely but this is where things stand now this is how much surplus each region will have this year everyone is in the green by a decent amount fast forward to 2030 and the picture starts to shift significantly the markets where stories of rising electric bills have swirled are already in deficit out to 2035 and those same areas with bill increases today get dramatically worse [00:26:06] Speaker 4: your rising power bill and how all these data centers are playing a role should dominion customers continue sharing those costs [00:26:14] Speaker 3: who should pay for it it has to be the data center industry [00:26:17] Speaker 2: what's at risk is every person every household runs the risk of a doubling or even a tripling of the amount they pay [00:26:23] Speaker 1: not all of it is data centers some of it is regular upkeep some is maintenance and some is an error where those two year out projections didn't actually happen the organization that holds the auction said 30 gigawatts were coming but only five did the charge from that calculation already appeared on customers bills so that's the case against it if someone shows up in your town asking for enough electricity to power a city and the power company needs billions of dollars of new wires and substations and power plants to serve them there's an extremely reasonable question to ask who pays for all of that because if the answer is you if this company's decision to build here makes the number on your kitchen table go up then suddenly the world's AI ambitions have become your electric bill but there is another side of this fight because there are two ways to look at a company showing up and asking for a billion watts of electricity one is where are we possibly going to find another billion watts the other is someone just showed up offering to buy a billion watts and those are very different ways of looking at the exact same thing for most of modern American history electricity demand grew slowly enough so slowly that the system could mostly grow slowly with it AI has definitively changed that for the first time in generations enormous customers are showing up saying if you can make vastly more electricity we will buy it we will buy a lot of it not for an hour not when it's cheap all day every single day and suddenly all of these weird electric states we just looked at systems built differently regulated differently priced differently have been dropped into the same experiment can you build can you build quickly can you build without making everyone already living there pay for it if you can what happens to that place that does you become a place capable of producing enormous quantities of reliable inexpensive electricity in an economy that increasingly wants enormous quantities of reliable inexpensive electricity AI needs that factories need that advanced manufacturing needs that and whatever comes after AI may need it too so maybe these data centers aren't just consuming the electric system maybe they are placing an enormous bet on which electric systems can grow and now every one of the electric states has to decide whether it wants to take that bet so the four biggest companies in America are spending they're betting something like 700 billion this year on these buildings last year was about 400 billion the entire interstate highway system every mile if you adjust for inflation costs about 500 billion dollars that was spent over 40 years these companies are spending more than that in 12 months remember U.S. East 1 the default setting for the whole internet the reason so much of what happens online passes through Virginia right now when someone in London or Sao Paulo or Jakarta asks an AI to fill the void there's a very good chance that void gets filled here in the United States and that is the other side of the argument the people spending all this money believe compute is going to become one of the most important inputs in the global economy that compute is going to become an export and one that is at least as important as oil and if they're right the question isn't just who uses the AI it's where the AI gets made because when a country becomes the early center of an important new technology it doesn't capture only the thing itself the money comes the companies come the suppliers come the engineers come and then people figure out what else they can build because all of those things are already there we've actually watched this happen once already in the story why did everybody put their servers in Northern Virginia because everybody else had already put their servers in Northern Virginia the center of gravity formed and then the center of gravity became a reason to keep building there the argument from people building these boxes is that AI could work the same way except now the competition isn't between Virginia and California it's between countries and this chart is what America's lead looks like today a formidable lead but not insurmountable Loudoun County US East 1 it's not wealthy because the data centers are there we established that already but its government has collected enormous amounts of money from them money that has helped pay for very nice schools and public services and allowed the county to lower property taxes by almost 40% so here is the pitch do that again except bigger except for the entire United States if the world is going to spend trillions of dollars building the infrastructure behind artificial intelligence build a disproportionate amount of it here generate the electricity here build the infrastructure here collect the taxes here develop the expertise here and hope that whatever gets built on top of all that compute gets built here too that is the bet and it's not a ridiculous one but there is a problem with it because the person sitting at the kitchen table it's not the United States of America they're the person sitting at the kitchen table so let's bring this full circle a nuclear plant is a building that makes electricity for the people who live around it a data center is a building that uses electricity on behalf of people who mostly live somewhere else probably far away same category of unwanted thing opposite direction of flow if you're sitting around the kitchen table the test it ends right there before anyone says a word about what AI is or is not but the connection to nuclear plants it's not complete because protests are not what killed American nuclear power a breakdown [00:31:16] Speaker 5: in an atomic power plant is probably the worst nuclear reactor accident to date an unexpected release of more radiation today from the Three Mile Island nuclear power plant has led to a series of consequences [00:31:27] Speaker 1: after Three Mile Island every plant under construction had to be redesigned mid-built costs doubled then they tripled as just one example in Washington state a group of public utilities had started five plants at once and when the money ran out they defaulted all what today would be almost ten billion dollars in debt it's the largest utility default in American history and around the country another hundred plants were cancelled and when the projects went bad regular people regular electricity customers were often left holding the bill so for 30 years nobody broke ground on a new one nobody had pockets deep enough to backstop the cost increases something people didn't want near them anyway that had a high chance of becoming a bill they would pay for non-starter but the interstates the interstates went the other way for a reason that had nothing to do with them being popular quite simply the federal government paid 90 cents of every dollar to construct them the fights were real some highways got stopped whole neighborhoods got saved but the fights were only really ever about where the road went never about who paid for the bill it's a messy fight but a fight that was survivable today now just three years into this whole thing everybody is intently watching the bill [00:32:33] Speaker 3: a 12-month moratorium was signed in mid-July by Governor Kathy Hochul [00:32:37] Speaker 1: halting the development of data AI centers [00:32:39] Speaker 3: Governor Greg Abbott is pausing all data center permits across Texas [00:32:43] Speaker 1: these announcements you would be stretching to call these bans by any definition ban isn't really the right word here these are pieces of paper saying if you want to build a data center you have to prove you can pay for it especially the electricity Ohio reserve enormous amounts of electricity and don't show up you still owe 85% of the cost Virginia data centers are being separated into their own customer class with their own rates Louisiana three new gas plants for one metadata center which Meta has to pay for and Pennsylvania a company agreed to buy power from a nuclear plant for 20 years so its owner would restart it that plant is Three Mile Island [00:33:19] Speaker 3: Constellation Energy said that it will reopen the Three Mile Island nuclear power plant selling all of the power to Microsoft [00:33:26] Speaker 1: the building that put the final nail in the coffin of American nuclear power is being restarted for a data center a thing that is less popular than nuclear plants this time the customer can afford the bill in the last great build out when costs exploded it would land on regular people ordinary electric customers today nobody really knows exactly how AI plays out anybody who says they do is selling something but the data center fight could be might be maybe even likely is a fight over a much older and a much simpler question when the cost lands who is sitting underneath it the town of a thousand people or whoever built the box there are actually other things I'd like to go on about like car factories versus data centers people accept factories they come with jobs data centers not so much and about half a dozen other things but I am way over time I have to end this so who knows maybe there's another angle to this whole thing we'll explore another time but until that time thank you for watching oh and hit that subscribe button it is down there somewhere and I will see you in the next video see ya