The Critical AI Data Center Power Bottleneck Nobody is Talking about #mxux

This is MXUX. You know, there's a big issue with power and data centers that everyone's missing. It's going to affect the chip trade and everything else. I'm just going to go through this real quick slideshow here. The US power grid is creaking under the weight of AI data center power demand. I mean, it's crazy. Um, the grid is already having trouble. I have a video I'm going to show a clip from later demonstrating this how uh surge uh power surges from data centers are causing um brownouts or uh grid trips that are shutting down parts of the grid. Uh this is happening right now. Uh all right, here's some stats back of the napkin. There there are 1,200 operating AI data centers in the US right now. about half of them, 680 are hypers scale sites, these giant, Google, uh, Amazon sites. They they have the same power use as a small city. Okay? And they're always on 24/7, 365 days a year, running full tilt. Okay? So, this is what's being let into the grid. All right? Now, if you If you deduce this into something that you can conceptualize, that's 680 new small cities added to the grid. Okay? If you go with 50,000 people per city, that's 34 million people added to the power demand. That's about in in a in a period of a year or two uh an increase uh of uh 10% in US population making demands on the power grid. That's crazy. So these 680 hyperscaler sites [snorts] are equal in power demand uh to uh 34 million people and you know that's 10% increase in US population. It's in my opinion got the the grid to the max. The grid has a a failure rate. It's designed to have one failure every 10 years and it's already starting to break that rule. Um, grid operators are right now asking data centers to disconnect and use their on-site power uh at peak demand. And again, there's a video on this. There's more than one video on this. You can you can take a look at this. See, here's the thing. Our power grid is a supply demand function. Okay? It's a spiderweb of transmission lines and substations and so forth. There's no storage of any kind. So what happens is the electricity is generated and then it is shifted from region to region and place to place to meet demand. But if the overall use exceeds the overall supply, there are peaker plants that come on and stuff, but I'm talking about even having those fail. This is this is a problem. Um, so they're asking these data centers under peak supply to to go to their own power supply and and that's called behind the fence uh data center. And there's there's two two types of behind us fence power. There's uh just emergency power that that uh powers the data center until the backup power comes on which is shortterm. But the long-term internal data power is basically based on natural gas. Okay. The short term can be hydrogen, it can be diesel generators but the long term is based on natural gas. Okay. Uh and and and they're they're talking about using bloom boxes. Some experts say the bloom boxes cannot supply the 24/7 uh nature of demand. However, let's just say they can. So, these are running on natural gas. Bloom boxes, the turbines, everyone G Venova, everybody's talking about they're talking about even bringing in jet engines, okay? Which run on jet fuel. That would be a different thing. uh but also generators. Uh so uh Colossus uh one has a natural gas generation plant that was moved from Europe and reassembled onsite for example. So they're telling them to get off the grid. They're telling them to go to their own power. They switch to their own power. There's a little emergency power that takes place there. Probably some cat diesel generators. And then their backup supply kicks in, which I would say right now is uh G Vernova gas turbines. Okay? And some of these gas turbines are actually running this uh business all the way. Now, here's the thing. In six months, according to this analysis I I read, a supply issue with Dometric N domestic natural gas will begin in in the US and it will grow continually to 2030. So the source that they're using for backup power and internal power at these sites is slated to go up. There's going to be a shortage. It's going to be like chip supply. So, we're looking at hitting a hitting a wall here with power, okay? And supply. It's it's more than prices. It's it's supply. You know, electricity and natural gas are going to have limited supplies. Okay? It's going to constrict the overall growth of the se sector. Um, I watched an interview with an old school e grid guy, construction manager for the grid. He says, "Lack of grid power will kill 50% of pending AI SATA centers. I have SA data centers next year." 50% of planned data centers are not going to be built next year. Now, there's still going to be conversions of bit mining sites and so forth, and perhaps sites that already have power allocations will get made or maybe not. We'll we'll have to see on that. But he recommends 50% of pending data AI project construction projects. Okay, let's talk new new construction will be killed next year. So there's 1,600 uh projects are planned. New construction projects of data centers are planned next year. Half of that amount would be 800 data center. So 800 data centers. So 800 data centers uh will not be constructed next year. It will be cancelled. they will be cancelled due to a lack of power and that's either uh natural gas or grid power. And uh in a minute we're going to go through what what possibly um could be done uh about this. Okay, this is clean view.co and just to give you some background on these data centers in the United States, here's some total numbers and so forth, here is a map showing the distribution of the data centers across the country. As you can see, uh many of the uh they're moving they're starting the trend is to move towards these unpopulated areas. But you can see this is this is the data center alley largest operating data centers in the United States, Colossus 2, which I believe is self-powered. I do not believe they have a a utility connection. Byron Childreth, which is I believe on grid. Riot on grid. Edgecore Mesa on grid. Colossus one not on grid. Uh the rest of these I'm not familiar with. Uh Microsoft probably certainly is on grid. Uh so the point is uh only really Elon Musk has managed to keep his large data centers off the grid. This is the plan construction. You can see there 900 a gigawatt is the output of a nuclear power station nuclear power plant. So, I am curious about Fermy because Fermy has a lot going for it except it's so mired in debt. Um, anyway, just to give you an idea of what's going on here. And then, uh, rules for connecting to the grid only go so far if there isn't enough grid. And this is this is what the old school guy I saw said. You know, the grid is is temperamental. It's tightly balanced. That's why Elon Musk did an interview was complaining. You you put an application into connect to the grid, it takes a year. This he said, "Yeah, we have to look at every possible overload situation, how that power would be granted and so forth." So, this is not a very robust grid. And people are talking, well, we need uh, you know, 10,000 mi or 1,000 miles of new high transmission lines, high tension transmission lines. We have the capability to build about 10 miles a year. So, there's some real B bottlenecks. Uh, this is a cool map. It shows the real time demands on the grid. It doesn't necessarily break out uh data centers, but we got two green spots here. We got SPA, which looks like it's in Georgia, and we've got LA is running well. So, uh that'll give you an idea of what's going on. This is this is a uh this is a tough problem. Now I have prepared an I'm going to I'm considering either making this available on the web or uh adding it to the description here but power is the is the is the uh bottleneck. Uh it's not chips. Um AI demand on the grid since about 2022 has risen fivefold and continues to rise. Okay. Um, so alternate sources of power being sused out to meet demand. Now, Bitcoin miners are converting to data centers. This is why Bitcoin is falling everybody because all these data center uh Bitcoin miners are selling their stock of they're getting rid of their Bitcoin to convert their operations to data centers. And that is what's keeping the price of Bitcoin down in my opinion. But we just go through all this. The the point here is all of these alternative sources are backordered several years. Bloom Energy has a four-year back backlog. Um Elon Musk, he just acquired a mobile gas and turbine generator company to try to get around these backlogs. Um so uh AI data center developers have already developed prime areas Virginia and so forth and moving to new corridors that trace major highways. So they're they want to be near usage and near 100% power because they have to be up all the time. Okay. And they're moving to rust belt and other places. um the more remote locations are going to are are going to make inference demands, you know, latency demands. You know, stock might be Wi-Fi, white fiber is one that comes to mind. So, they're going to have transmission problems. But when you look at uh the most relevant practical AI power solution uh and this is this is I just want to mention this sympathetic surging modulation. I was watching an interview with a guy who was a program manager for these and he for data centers and he said they have something happening that no one suspected. [snorts] I don't know if you've ever seen these videos where they have the metronomes and they all eventually sync up. You start one and then the other one and the other one and and after about you know 20 seconds they all are going the same way. This is happening in the data centers. All the power is surging at at peak use. All the power is surging through the chips in a syncopated way. All the cooling pumps are going in a syncopated way and it's causing this sine wave. Okay. Uh and it's and it's creating a sine wave of of power demand. Peak trough. Peak trough. Peak trough. When if this gets high enough, it will uh stress the grid. It'll disconnect from the grid, the data center. It'll automatically go on its power, internal power to maintain power. When it does that vic disconnect from the grid, it throws the balance of the grid off and then the grid has short circuits all the way through it. So, this sympathetic surging is something that they did not realize was going to happen. and it modulates. And these peaks and valleys are what's creating these power surges into the grid and uh causing their backup power to turn on and and then turn off and then turn on and then turn off and then switching on and off the grid. This is this is the start of this uh this is completely new discovered issue that was not recognized during AI data center construction. I feel it would become a bigger issue with the agentic adoption rate skyrocketing. you know, as the the guy was saying, this this comes with, you know, steady peak use. This that's how this it's like a heartbeat. The system all synchronizes. Anyway, I imagine there's a physical also a physical uh part of this. Okay, there are this is the these are the solutions that self-contained behind the behind the fence uh power boom boom boxes there's a four-year backlog hydrogen fuel cells like plug and so forth it's not mature enough for main power right now but it's okay for backup power gas turbines jet engines relocated power plants have been used they're all backlisted 3 years, four years, modular nuclear 10 years. The only practical solution uh is to increase grid efficiency. Now, here's the thing. [snorts] There's there's no capacity to store energy on the grid. It's completely supply and demand. Okay? And because in the past, there were never any good ways to do this. So the it's created the energy is created and it's it's distributed and [snorts] if there's a excess demand peaker plants uh which are natural gas plants and other plants come online to provide more power. [snorts] What they have to do is introduce batteries into the grid. Okay, to act as capacitors. If you know how a stereo works, when you have a big bass boom, uh it takes a lot of power and it'll drain the power out of the amp and the and the stereo stop working. You have a capacitor in there which is a fast discharge battery and when that demand hits, it drains that capacitor and it meets the demand and it doesn't crash the amp. This is what our grid needs. the grid needs is this is the first solution batteries uh strategically placed batteries. Uh and I would add uh that if you look at must uh Colossus 1 and two mega packs mega I think he spent a billion dollars on mega packs. This is how he is running that. He is keeping a steady energy supply. He's using natural gas and mega and mega pack battery packs to supply the you know 99999 up uptime 24/7 uptime off-rid. Okay. So they have to do this. There are presently three USA manufactured batteries for this application. Tesla energy. This This is a sleeper. Elon Musk said, "Well, I think he just made a comment in a sight. I think our battery business could become very significant very soon. That's where these batteries are going to come into play." Okay. Inside defense and outside defense. Uh EOSE, which uh Nvidia uh Jensen Wang mentioned. Okay, here's the thing about EOSE. Tesla, of course, uses uh lithium batteries, which are, you know, once they catch fire, they burn like a volcano. EO and they don't want them in near neighborhoods. They don't want them in public places, okay? Uh, you know, industrial centers, so forth, but not at the substation down the street. EOSE has water zincbased batteries. Okay, these are totally safe. They're environmentally sound. They have no uh toxics in them, toxic substances in them. EOSE is I believe going to be a major player in [snorts] these neighborhood or uh other uh public grids these batteries because uh I don't know they are water-based I don't know if they'll how handle subreezing temperatures however uh in other applications there's no nimi there's not not there's nothing to oppose with these things they don't catch fire if they leak They're they're harmless. EOS. And then you have T1 uh energy which uh Leopold invested in. They're making they're basically a startup however, but they're making um uh industrial battery supplies. Uh there's other publicly traded battery companies, but you know, there's a lot of trouble in this sector. I'm not going to mention any of those companies. I would, you know, I think these three are the main. So now here's the rub. Manufacturing is onshoring by Trump policies and it is and they are building factories. It's going to double the demands on the grid over and above the AI. So uh you know this is just going to stress the domestic residential power supply as well as the commercial power supply. So we we could possibly see mass induction of uh adoption of home and local business solar battery imple implementation to provide peak power to the grid and lessen peak demand on the grid. And this would ensure homeowners and others of of having a power supply handy. Of course, you know, the Trump administration has has changed the outlook on solar, but this is a reality. Okay. So, um this is introducing a capacitor, a system capacitor in your own home. And they these also can feed energy back into the grid as well. And uh I'll just talk about uh T1 energy again here. Here now this is interesting. Corporate filings show that X AI spent over $1 billion on Tesla storage megapac battery units between 24 and mid 26 to support massive AI training. As I said, Elon has basically got self-contained uh gigantic uh data centers, and this this could be the model going forward for terrestrial data centers. But again, massive deployment, $1 billion $1 billion worth of batteries involved. So I mean we have to this this is the quickest most viable solution all this nuclear and all this stuff this is going to take years and u and this is talking about the old school grid manager here it's talking about how 50% of the projects are going to be cancelled. Uh, look this up. Lowden County data centers. This uh is the uh flickering of the grid caused by data center demand. Uh this is a um other they're already planning for this. I'm not going to click on this link right now. you guys can [snorts] get a I I'll put this in the description hopefully and then you can click on the link there. But [snorts] people are looking at this and again according to these heavy duty analysts natural gas is not going to be any way out. Uh all these prices are going to go up within the next year. They have serious doubts that natural gas powered balloon boxes can meet 24-hour 24/7 power demands of AI. People don't don't think they can. These experts are saying, "Yeah, they can handle it, but not the 20 not the demands they're making." So, what's the result of all this? It's going to be a slow grid upgrade. Uh natural gas prices are going to go up. Power prices are going to go up. Um, consumers will face rising electric and gas prices. Experts recommend homeowners install solar and battery systems now to cut future utility bills. I would see that, too. Uh, this is a great interview. The natural gas crisis no one sees coming. Okay. Um, this is about, [laughter] excuse me, [snorts] [cough] [clears throat] this is about Musk buying the port turbine company for a billion dollars. So anyway, here's the conclusion. Now, what's all this mean? Here's what's bullish. Green light. SpaceX orbital data center. Tesla obviously get the units out in space and use solar. Tesla mega packs and home solar. Absolutely. Okay. Absolutely. Um EOSE Tesla and T1 battery grid uh grid battery optimization. Absolutely. Now uh there's cubit and a couple other uh there's d is it dsquare uh more efficient hybrid uh quantum processing and processors there they're putting u quantum uh is playing a you can now use quantum for part of your processing the thing is quantum is so much quicker and it uses less energy in that extent. uh but you know the um uh Nvidia chips are going to a 800vt architecture which is even going to be more heat more power and it's already the most one of the most super effic I mean the main users have super efficiency in the uh power I mean cooling maybe is the thing um industrial grid scale solar power supply now this is something that is new and this is something that MA is doing and Elon Musk is big on it. Uh these guys are buying industrial you know square mileile large industrial solar panel farms. Okay. And the and this is one of the reasons I'm I'm bullish on Tesla and also uh everybody's discounting Meta. Meta Meta is building capacity and now they're buying this solar up. Um they're doing what uh Amazon did to create AWS. It's not readily apparent. I think Meta is going to come out a winner in the end. Uh but you can look at these links. I I'll put this in the in the u uh thing in theuh description, but um solar is a big part of this. Batteries are a big part of this. Okay. Um and then I have down here AGI, which I am AGI pill. I think it's going to emerge or if it hasn't emerged behind doors already, you know, can AGI come up with a better solution? Is there a fusion or a fision or fusion reactor or some other solution that we can't think of that AGI could think of? But anyway, so here are some companies uh that are uh interesting. I I mentioned WiFi, White Fiber. Um, they have a lot of experience working with Sarabus and Sarabus has a 40 billion contract, $40 billion contract with OpenAI and I think they may end up playing a big role in this inference data transmission. That's another company I mentioned, but certainly Tesla, certainly SpaceX, uh, you know, the red light. Okay, now I'm not giving it financial advice in any of this. You do your own DD. Uh, Bitcoin stalls as more miners sell their Bitcoin to raise cash and convert to AI data centers. Okay. I I don't know if I don't know if Bitcoin is going to recover soon. Picks and shovels are going to lose 50% of new domestic construction. Now, there is international construction. whether there will be international constraints on the Nvidia chips or whatever. But you can see these data centers are getting bombed in Saudi Arabia. So there's still an international market, but I don't think it's nearly the size of the US market as well. I think the international market is China where we are not going to go. And by the way, China has a modern DC power grid which is something we don't have. So AI compute expansion will be limited uh without an existing so some of these places have arranged for grid connections you know 2 three years ago if they if they have a an existing grid connection then they're okay I uh did has one ke is another one um But if you don't have a grid and a grid connection is no guarantee now either. Uh if you don't have a grid connection then till new so industrial solar and grid battery oper optimizations are engaged you're out in the cold private private power uh uh generation [clears throat] gas diesel turbines are backlog three plus years bloom boxes are backlogged four years. So that means as we have this constriction increase in demand token supply growth to terrestrial token supply growth is going to stall I think and token cost is going to increase. Okay. Now they can very you know use different models and so there could be breakthroughs in software but in general supply and demand as you know they're already at m near max production with these existing systems hypers scale data centers that have long-term power natural gas supply contracts and this is Amazon Google uh Meta um so forth. Uh hyperscalers that have uh long-term power are going to have a power mode. Okay. And they are going to dominate terrestrial compute. So the winners Microsoft who who just had earnings. Now they're discounting MA. I I have a different feeling about MA, but they got a moat now. This is their moat. This is why everyone is, you know, they're playing musical chairs and somebody's going to get not going to have a chair and they're not going to have any power and these guys that have the chairs that have the power are going to dominate everything. Okay. Uh this is this is why I'm so bullish on Starlink and and Tesla and SpaceX and the merger and so forth. Uh orbital AI makes is going to be the game changer. Okay. and Musk is determined to get the project started in 2028. Uh, which is going to be perfect timing with all of this. But it really is, it doesn't, it sounds so out of control and crazy, but it is a viable alternative there. This is this is a big problem. Power and then I'm just going to close with this. power power grid problems are already evident and will increase exponentially with mass AI integration. You can look at the Oracle Oracle is another one with the building a mode but you can look at or Oracle's demand. I mean these these guys that are saying well AI isn't going to make AI is already making money. It is going to be, you know, it's just like getting the internet. Everybody, every workplace is going to have, everybody's going to use it. And the token demand, the the nature is old software would scale. You'd make it once and sell it a thousand times. Okay? And that's why these companies like Microsoft made so much money. they got fixed co, you know, the cost is fixed that no matter how many units they sell. With AI, you have to create the tokens every time you use it. So, in other words, um there's it's like the power grid. It's supply and demand. It's a constant flow, you know. I don't know. Uh I I I [clears throat] think I heard there was a firms there's firms starting now that are going to arbitrage tokens and stuff. This this may come and so forth. But anyway, the point is there's a very big elephant in the room that nobody's looking at and it's this natural gas and uh electricity issue in the United domestically in the United States and you know with export restrictions and everything and with the key demand and key development in the United States this is something that uh is critical and you know there's going to be winners as I mentioned the hyperscalers that are in place now I think Tesla XAI uh the OSC you can do your own research on this I'll try to put the text from this either on my substack or uh in the description all right this is MXUX good luck in the market Um there there's not a lot of news on this right now, but it's coming. Okay, this is the uh video. Uh this is a this is just happened. Uh a data center tripped a grid. And I'm just going to play a little bit of this and then I'm going to mute it and talk over it. But okay, this is a guy. He's talking about monitoring the grid. Uh but the point is this data center tripped uh a break not a breaker but it tripped an alarm with the grid and it sent shock waves through the entire grid uh in u the United States actually. It spiked then dropped before leveling off. And this is NBC4 News Washington. From Chicago to Boston, the source, he said, data center alley. This one event serving either one building or whatever the data center was complex is in Ashurn impacted the entire region here. This is the point. This is what's happening with the data centers. they are again I mentioned earlier or later with the modulation that's taking place uh the the thing is this is going to start happening more and more all the time uh will there be a technology to break through I don't know right now the grid is the the improvements to the grid are glacial now this is the failure that took place it spread. You can see this is a map that's the dark is the outages that's that's spread across the entire country. Okay. And that's from one data center. Okay. So, uh as I said the changes in the grid are glacial. This is going to be a bigger and bigger problem going forward. It's going to take time uh to solve the grid problems. And I don't know, as I said, right now they're voluntarily asking data centers during peak use to go onto their own private p power, you know, can they do that? Will they do that? Will they be forced to do that? You know, anyway, there's a lot of headwinds with data data centers. There's nimi, a lot of uh bands and uh but this is a real problem. This is an engineering problem. This is a physics problem. So anyway, that's a great uh great story. I'll put a link in the description.