Limitless: An AI Podcast - Elon Will Handle the Energy Bottleneck Himself: Buys APR Energy for $1B

Episode Date: July 22, 2026

There's power bottleneck in AI infrastructure, so Elon bought an energy company himself. From APR Energy, behind-the-meter generation, and nuclear, everyone is trying to secure electricity for data centers. We also cover rising DRAM, HBM, and NAND prices, and wonder if energy may be the next major AI infrastructure trade.------🌌 LIMITLESS HQ ⬇️NEWSLETTER:    https://limitlessft.substack.com/FOLLOW ON X:   https://x.com/LimitlessFTSPOTIFY:             https://open.spotify.com/show/5oV29YUL8AzzwXkxEXlRMQAPPLE:                 https://podcasts.apple.com/us/podcast/limitless-podcast/id1813210890RSS FEED:           https://limitlessft.substack.com/------TIMESTAMPS0:00 AI Power Bottleneck2:44 Energy Demand Explodes4:12 Elon’s Permitting Loophole7:23 Memory Trade Rotation10:08 Korea’s Memory Shock12:27 Electricity Becomes the Trade14:51 Power Stack Layers16:15 Bloom’s Portable Fix17:35 GE Vernova Leads19:29 Nuclear Gets Interesting21:44 The Power Trade Thesis22:44 Bullish on Electricity24:09 Model Agnostic Demand25:15 Short-Term and Space Power26:00 Closing------RESOURCESJosh: https://x.com/JoshKaleEjaaz: https://x.com/cryptopunk7213------Not financial or tax advice. See our investment disclosures here:https://www.bankless.com/disclosures⁠Josh works with Anthropic as a contractor. All views expressed are his own and do not represent Anthropic, its leadership, or its affiliates. Nothing in this episode is investment advice.

Elon Musk just spent over a billion dollars of his own money to buy a company that no one's ever heard of. It's called APR. And the strangest thing about this company is it doesn't make AI chips. It doesn't make rockets. It doesn't make any sort of AI model. It makes gas turbines that sit on a trailer, get wheeled into your data center campus, and powers it on all your GPUs in a matter of days. Now, this is coming from the guy that spent two decades philosophizing about solar energy.

But when you unpack why he made this purchase, it reveals a new trade in the AI space, which explains why memory stocks are crashing currently. Energy is explicitly abundant in the US, but bringing that energy online to power the billions of dollars worth of GPUs that are going online this year is an incredibly hard feat. And there's a small category of companies that are helping do this in 2026. Yeah, there's a company called APR Energy, which is the name of the company which he purchased. And it has a pretty funny backstory. It was founded in Jacksonville in 2004, then listed on the London Stock Exchange in 2011, and in 2013, it bought GE's energy rental business to become the largest mobile gas turbine

leaser in the world. And then ever since then, it's been passed around and taken private to private equity, and due to some investigative work, and searching through filings, we were able to see that Elon Musk bought it for about a billion dollars. So what did he get for a billion dollars? The answer is about a gigawatt of power. For reference, a gigawatt of power powers roughly 750,000 homes, and it's also the output for one-sized nuclear reactor. This gigawatt is also enough power for about 600,000 H-100-500-classed GPUs, which is,

I mean, as far as I'm concerned, the largest coherent cluster. So if he's actually able to get another gigawatt, that's amazing. But the question that left me is, why is this owned by Elon and not SpaceX AI or Tesla? It's like, why of all the categories is he choosing this one, Why this particular company and why on his personal balance sheet? Yeah, so I have a few theories as to why he might want to do this. One obvious one is he might want this to apply to a bunch of different companies. The other main one being Tesla, and Tesla, for what it's worth, is meant to be an exclusive company separate to SpaceX AI,

although there are rumors that they might eventually merge at some point in the future. So I think this is more like a tax-slash-structural reason why he purchased the company on his own time. In fact, the only reason why we learned about this purchase is because one of the companies that was invested in APR had to disclose personal returns of $50 million. That's the only reason why we were able to realize that Elon made this purchase.

Now, if we zoom out for a second, we've heard that energy and power is the next bottleneck for AI. But I think a lot of people don't actually understand why and what the problem is. I just want to touch on that very briefly. Now, if we rewind about three years, 2023, the demand for data center energy was roughly at around 23 gigawatts, which at the time was gargant to it. We did not have near that amount available to bring online.

Fast forward to 2026, that has more than doubled to 46.5 gigawatts, I believe, in total. Now, the issue is, in this year alone, we are only planning to bring online 12 gigawatts. That's significantly less than that 46 target, right? But it gets even worse. We are currently on track, and we're near over half the year now, on track to bring online five gigawatts of that 12 gigawatts commitment. So the point I'm trying to make is there is a significant bottleneck, and it's very, very slow to bring energy online.

Now, if you might ask, well, don't we have an abundance of energy in the Western America in general? The answer is, yeah, we do. But it's incredibly hard to bring this energy online. We need access to high voltage transformers. We need access to grid infrastructure. We need all the permitting and regulations around this. All of this slows everything down to the tune of five to seven years. I'm not exaggerating.

We're talking about like almost over half a decade to do any of this. Now, what Elon's purchase shows us is he's kind of going into the gray area. Now, there's this thing called the Clean Air Act. And it's basically this act which states you can't bring gas turbines onto your data center. in order to power up your GPUs. That's kind of like, not illegal, but it's allegedly kind of like not malpractice. Now, the fact that he's purchased this company

allows him to bring permitted gas turbines onto his data center ground under that clean air act. So it doesn't trigger any kind of like warnings or anything like that, which allows him to power up his GPUs. So Elon's being very smart. He's trying to be the fastest scaler for bringing GPS online so he can build the best model.

That's what Metz is trying to do, and that's what he's trying to do with GROC for, point five and subsequent models. Yeah, it's a race to power these GPUs. And like you mentioned, the infrastructure is the problem here when it comes to creating power. It is less the actual power. We have all of the resources. We have the oil.

We have the natural gas. We just have a very difficult time plugging it into the infrastructure that we need to turn these things on. And if you look at the country like veins, all the power lines that run through, the grid is really strained and it's in a really difficult place. And I remember having this conversation a decade ago when I was first getting excited about electric cars and Tesla's and the impact that just charging all of those cars would have on the grid and how difficult it would be to sustain. And at that time, we were barely keeping up. Now we are still

barely keeping up. So to put on cities worth of energy online into these existing grid structures, it causes a lot of trouble and a lot of problems, which leads us to the one way to actually do that, which is more of a modular solution. Instead of tapping into the grid, you have to bring your own power, bring your own electrons. And there's three key categories to doing that. The first is solar. Solar has the largest footprint in terms of these options. Solar generally works by just absorbing a lot of sun, store it in batteries, feed the batteries into the data center. This is good, but it doesn't allow you to get as much as quickly as densely. And the permitting, as we know, is a really difficult thing. The second option is nuclear. Nuclear is a kind of far cry from being

readily available for data centers today, which brings us to the third, which is just bring your own turbine. Bring these turbines, tap into a net gas line, and they're on their way because these turbines can just generate electricity on site. And in a way, these data centers are starting to build their own electrical grids. And then perhaps they can reverse their way into feeding the existing grid. But for now, they just need to get energy to the data centers. And the best way of doing that is just bring your own power grid. And that's kind of the basis of this investment here with Elon is like he's now able to buy a company that owns these turbines and he's going to bring them over to the data center and power these chips on fast

than everyone else. This is what it's come down to. You have to actually acquire private companies and roll out your own infrastructure if you want to build this. Isn't that insane that we are in a position where in order to get access to like a base commodity, electricity, to power the thing that's going to make your country, your GDP increase significantly, you have to go through all this archaic permitting laws and stuff. New York State just banned a bunch of data centers. So that's going to delay data centers in New York by like five years. It's just crazy that we have to go through all this red tape. But nevertheless, this is a workaround. And if gas turbines on a trailer sounds familiar, you are probably thinking of other companies like Bloom Energy and publicly traded stocks and

companies that might be fixing or solving this bottleneck. We'll get to that eventually later on in this episode. But before we do that, we have to talk about the movement of funds and capital within the AI trade. Because for a lot of time on this show, Josh, we have spoken about a little component called memory. And memory feeds into a ton of things, most significantly, the GPUs that power the training and inference of all these different AI models. And one common trend that's happened with these memory pieces is the prices of these things have skyrocketed. To the tune, I think it's on average, three to 500 percent over the last nine months. So it is absolutely insane the types of demand that we're seeing for memory. High bandwidth memory is the most obvious one, but also Nand Flash, the stuff

that's created by Sandisk and stuff like that. And these stocks have taken a little bit of a battering recently. Yeah, memory stocks have been absolutely crushed. It's been a really difficult time to invest in them if you've only owned them for two weeks. If you've won them for any longer, congratulations, you're still doing incredible. But there has been this whiplash, right? That's kind of happened in memory stocks where they went up, seemingly 20% every single day for months on end. And now they've started to back off collectively about 20% off of the highs. And this is interesting because when you compare the actual price of memory stocks to the price of memory, the price of memory is continuing its up only trend. In fact, just this month, it looks like

the price, the average price of DRAM is up almost 20% on the month. Which is crazy. Yes, this is the shirt that I was looking at. When you look at the month of July, the cost of DRAM is up 20% and the stocks are down 20% even though the demand has not changed one bit. So that's why it seems like the market is a little spooked and rightfully so. I mean, these stocks just ran up a tremendous amount. When you look at the actual core costs of goods sold relative to the margin that they're able to charge for these things, it's still continuing to go up only and that demand curve hasn't slowed. And this is the interesting thing here is like memory stocks are getting crushed. Memory prices are still going up. But it seems like people

are just kind of tired of that narrative. And the money is just starting to flow into other more exciting places because, I mean, perhaps everyone's just made enough on the memory trade for now. I think the markets are incredibly emotional. And if we use memory as an example, to your point, demand hasn't wavered. It's just gone exponentially up. The prices of these things going up, prove it. But also the LTAs, the long-term agreements, which is basically the contracts that these memory supplies are signing with clients. There was this crazy stat for SK-Hinex, which is, I think the number one or number two memory supplier in the world, there's only three of them. Between 13 to 15 customers secured around 40% of their projected profit for next year.

So they've already sold out their supply for the entirety of 2027, and 40% of that profit has already been committed to, so they have to pay regardless of what happens with memory supply next year. 13 to 15 people. So the point I'm making is the demand is overwhelming at this point and nothing has actually changed. If you look at Micron, let's pull up their chart over here. Over the last month, they're down around 24%. But when I look at their forwards earnings, their 7x forwards earnings on 350% revenue and 85%. 85% gross margins. There is very little businesses, especially in the hardware realm, that makes that types of margin. And, And the point is like, you can call a memory bubble as soon as there's oversupply for this type of commodity.

But the truth is, the fabricators, the plants, the factories that are creating these things aren't in overabundance just yet. And that bottleneck isn't going to get unlocked until around 2030. But we see people getting frustrated. So I think this is just temporary. And if you want to look for a reason, a thing to blame, look no further than our friends over in South Korea. Now, a reminder for everyone, the two biggest memory supplies. are based in Korea. It is SK Hynix and Samsung. And there was a lot of red on the chart, as I'm showing you on my screen here, over the last two weeks, because a lot of investors in Korea were

over leveraged to the tune of about a billion dollars. And so the market has shed the equivalent value over the last week to the tune of $1.5 trillion. Now, of course, that is an order of magnitude larger. I was being a little kind of tongue in cheek there. But the point is, the market is overreactive and I think this is just oversold. The fundamentals still sit and memory is still an important trade, but people are looking for other things and maybe that's power and electricity for now. Yeah, I think that's probably the rotation that's happening is people are like, I'm done playing with this toy for now. And the fundamentals are still very strong, but I mean, everyone's just gotten a huge win and are probably just looking for something else. Our guess,

and the reason we're recording this episode is because it seems as if that is going to be energy, it is going to be powering these things on. So like everyone's like, okay, well, we understand memory now, we understand the complex. S.K. Heinex just went public. Perhaps that marked like the end of people's enthusiasm towards it. Where does the money rotate out to? Well, probably this electricity trade. And we've mentioned this a few times in the past, but the electricity trade, at least for me personally, is the most exciting of all of the trades because it's something that is so durable and so necessary in any advancement of any society ever. So even in the case that all the data centers turn off tomorrow, there's still this unbelievable demand for electricity

in everything that we do. And that's why the electricity trade is this really exciting thing that I'm kind of happy to see dollars moving over to. Because, I mean, yeah, if we look at these charts that we have here,

the U.S. data center power demand is basically doubling over the next 20, or over the course of 24 months, from 31 gigawatts to 66 gigawatts in 24 months. That's crazy. And then what it does to the old grid

is the total U.S. electricity demand. It used to be 1%. It is now climbing to 3%. That number is going to continue to be up only. So there's a increasing demand of these data centers. It's not going anywhere. In fact, it's going straight vertical. And an incapability of us to deliver on that demand with these existing infrastructure. And you just mentioned, it's like crazy that we have to have all these permits and everything, but at least we have this system in place that allows people

like Elon to come and buy these turbines and like solve the problem. So what's really cool now is this opens up this entire new world where, all right, the grid is kind of cooked, but we need electricity. Yes. Who are the people who are most creatively solving this problem to get these data centers online as fast as possible? And that's where the money is going to be flowing. It's like, if you can create an electron of energy for cheaper than someone else, and you could drop it at the data center, that's basically an infinite money glitch. Whoever can figure that out will get paid as much money as these large cap companies are able to fund because that is the single bottleneck that no one's seemingly been able to solve. I mean, we look at the gas turbines. How long does it take

just to get a blade from a turbine? They're backed up for years with purchase. orders. So this is a really difficult challenge. And that's where the focus is now. Who's solving that challenge? So who is solving that charge? Maybe we should answer that question and get into kind of like the companies that are both private and publicly traded that might be solving the problem that we've just spent the last kind of 10 minutes explaining. So now we've talked about the AI infra stack. There are many different layers that previous episodes should definitely tune in and check those out. But within the power stack specifically, there are its own layers. Now, you mentioned earlier, Josh, that there are modular ways to approach it, different types of companies, nuclear, you've got kind of like solar, those kinds of

things. There's also different quick fix ways to solve the power constraint. And I'm going to run through a few, starting with layer one, which I call the quick fix. That's what Elon just did. He purchased a literal company for a billion dollars that creates these gas turbines, places them on a trailer, like a trailer truck trailer and wheels them in to your data center. A literal trailer. I'm not exaggerating here, right? And they park it behind the data center. There's actually a term for this.

I was watching the All In podcast and they explained it very well. It's called Behind the Meter. So this is kind of like a gray area where you can kind of park it behind the official meter and power up that meter. And technically it's legal. It's all good to do. And it's a fast way to power on your GPUs, but there's a few issues with it. Number one, you can't power up.

that many GPUs. He's got one gigawatt and, you know, his data centers on average are going to be scaled up to three gigawatts. So there's still kind of like a fixed thing there. You can get it online in a couple days, which is great, but it only lasts for maybe like a six to 12 months. So it's not like a long-term solution. Now, the layer below that, we start seeing names that are very popular, at least in this ecosphere, which is called Bloom Energy. Now, Bloom Energy has this thing called a fuel oxide cell. It's kind of like this massive gigantic box that, again, is portable. You bring it onto your site similar to the gas turbines on a trailer, but it converts natural gas into energy way more efficiently, and it can last around four to seven years. Now, the reason why

people are so hyped up about Bloom Energy is because typically, if you have to wait five to seven years to even get access to the transformer, if you can get access to this early on, and it's kind of like a medium-term fix, four to seven years, you can scale your data centers way quicker and maybe even train a frontier model way sooner than your competitors. So that's what the likes of meta and a bunch of data centers in Mexico are doing. But there is an issue with this. And I want to find the specific tweet here. Yeah, over here, Bloom Energy shares are trading lower after New Mexico regulators rejected for the second time permits for a gas pipeline. So the point that we were making earlier on in this show is you can bring these amazing fuel.

oxide sells, but it runs into the red tape issue, which is we can't get access to permitting to be able to run these things. So you end up facing the delay on its own. That's why Elon purchased his company APR for a billion dollars. And GE Vernova has kind of positioned himself at the center of this trade. They produce the turbines and the grid equipment. And the stock has just continued to go up and up and up 300% in three years. And it seems like they're taking orders all the way through 2031. So the revenue is very pretty. predictable. In 2025, their orders doubled year over year to $7.1 billion. And it's just this like unbelievable company where, again, similar to Bloom Energy, if you can build the electrons, they will

come. And as soon as you hit a wall, there's someone else who has the permitting, who hasn't hit the wall. And it looks like this company in particular is one that hasn't. And it's been this really nice, slow and steady growth. You haven't seen the crazy exponentials like an ETF like DRAM has. It's just slow, steady, really strong growth when it comes to building turbines, generating electricity, and getting the permitting required to actually put these things online. I like to think of GE-Vanova as the old staple of the power industry. They've been around for a while, so they know how to deal with the old-school transformers,

the high-voltage types of things that you need to kind of equipment that you need to build, Pioneer. They have all those supply chain relationships, but they also understand the quick-fix solutions that we mentioned earlier from the bloom energies, from the APRs, and they also spin up and create gas turbines. if you look at the company and the clientele that they've signed, we're talking about $7 billion deals with Microsoft. We've got Open AI being one of their primary customers. They're all in the

process of powering up a bunch of these company's data centers. So the point is, G. Vanova is kind of like the TSM of power, dare I say, and they've kind of been around, they've kind of proven their method, and they're seeing consistent demand. I think they're growing on average around 30% year over year. But if I had to take a bet for this particular company, again, not investment advice, I would say that growth goes exponential over the next six to 12 months when people realize that you need power and energy to bring these things online. Now, the final rung of our stack that I'm showing you over here, it says layer four. It talks about utilities and IPPs, but one thing that isn't mentioned right now is the nuclear side of things. And there

is a company that we've interviewed the founder of, and they're building kind of like the modular approach to nuclear. Now, typically, if you look at any nuclear company, you're not seeing any of these really come online until 2031 to 2035, but Valor is an example of a company that is going to speed this up pretty significantly. Yeah, also just to clarify an IPP, it's an independent power producer, which is a company that owns plants and sells electricity into the market rather than serving a regulated territory. And this is a really important distinction because like we mentioned, selling back into a market is far better than having to deal with red tape. So you figure out the red tape on your own.

own, you produce the power and then you sell it back. And that seems to be the best way that these companies are kind of working this out. And in fact, larger companies have been working with these IPPs to lock in 10 to 20 years of electricity at fixed terms. So Microsoft, for example, they have three-mile island, which, I mean, for those who are familiar, is an very old, previously shut down nuclear reactor. They brought it back online and their deal length is 20 years. and they get 100% of the output. So the time horizons on these are huge, and it shows the real scale of what this kind of power trend is going to look like over time,

where there's obviously no shortage of energy demand. People are locking it up for decades at a time, as if there's very long-term bonds, because electricity really is a form of currency now. It's like if you can get the electrons, you can power on the intelligence, you could serve the tokens, you can make money, and in a way, earning 20 years of electricity, you getting that lockdown is more powerful than issuing like a 20-year bond who can give you

3% because you can create a lot more upside with that energy. So this is an important trend that I think we're going to be following as well. It's just seeing what companies are able to lock up guaranteed energy for a really long time. And the nuclear opportunity, like you mentioned, is so cool. It's just a little early. No one's quite got them online yet. No one has figured out the permitting. It's not there. But man, when that comes online, that's going to be huge. You know, as we talk about kind of like the structure of these energy and power companies and the deals that they're starting to sign, but it hasn't been really abundantly signed just yet. It's not really guaranteed growth rate.

It's a little bit murky. People aren't really sure if this is going to be the next trade. You know what it reminds me of? It reminds me of the memory trade before it became the memory trade. Before they started signing massive long-term agreements and price hikes of three to 500 percent. it sounds exactly the same. You know, history doesn't repeat, but it rhymes. And I'm sensing this for the power trade as well.

People don't really know about G. Venova. Maybe people in the comments are going to say, like, hey, like, you know, I'm an old schooler here, and I have heard of it. But, like, most people don't. And they don't understand the importance.

It's a very different type of problem. Like, with memory, it's easy. It's kind of like memory, okay, you need to remember context about everyone. I want the AI to remember me. Cool. But power, it's kind of just like, yeah, can you just get the electricity?

Don't we have, like, an abundance of electricity? and it's a much more nuanced problem. And I think people are now only starting to unpack what that investment sector might actually look like. So brings us to the question, bull and bear case, Josh. I think you should go first. What is your, are you bullish or bearish

about this power trade over the next, let's say six to 12 months and then like three to five years? Yeah, I'm bullish on the power trade until the end of time. I think like at the limit, electrons are more valuable than dollars. and I think this continues to be true for basically infinity. There has been a singular trend throughout all of history

where more energy equals more productivity, equals more innovation, equals more prosperity. And the more electrons you could throw at a problem, the better the success outcomes become. And this seems like a one-word trade forever. In terms of these swing trading things that are happening, I don't know.

I mean, memory was the first trade of the first half of 2026. And it's ironic that it kind of peaked the same week that Ska Hanix listed on the NASDAQ. And now the question is, like, what is the next scarce input everyone's running to? Elon kind of answered it with his billion dollar acquisition. He said, hey, it is electricity. It is electrons.

Go do this. So in terms of short term, seems good. In terms of long term, seems amazing. Medium term, who knows. But I think the demand for electricity is up only forever. Okay. That's pretty bullish. I'm afraid to say that I'm also equally as bullish.

I'll give you a few different reasons as to why. Number one, I love that this layer of the AI InfraStack is completely agnostic to whatever your model is, whoever created the model, and most importantly, which country created the model. A big topic of debate over the last week is China versus the USA. China just released an amazing open source model, the largest ever, 2.8 trillion parameters, Kimi K3. And it's now the case that over 56% of tokens, at least on OpenRouter, which is an open American platform, is being spent on Chinese open source models. So the big question is, should we allow people to use these open source models

and does it matter that they're not using an American-based model? Well, you know who isn't worried? Jensen Huang, who creates GPUs that power all these different types of models. And you know who doesn't care below Jensen Huang? It's all the people that are helping power and energize all these GPUs. It's going to be the same case. It doesn't matter if your model is open, whether it's cheap, whether it's expensive, whether it's frontier, whether it's not.

The point is you're going to need energy and you're going to need GPS to do this. It's a commodity that is going to be required regardless of the situation or geopolitical situation as is. Now, when we talk about energy in general, we can't talk about Elon purchasing APR and also Elon launching data centers into space to literally harness the energy of the sun.

He's been the biggest proponent of energy harnessing for a while now, and he's taking two different approaches on the ground. He's purchasing companies like APR for short-term fixes, but he's also, in his long-term plan, aiming to harness one of the largest energy sources in our galaxy. And so if that doesn't show you how bullish people are on the future demand of energy, I don't know what will. So I think this is just going to be a growing trend that people are going to become more amicable to, like they did with the memory trade over the next couple of years. And you just need to sit tight. And that is the

episode on the state of energy, the state of power, the state of electricity, where the possible next trade is going. Again, as always not financial advice. I don't even own any of this stuff here, although I probably should. But it's just interesting. It's exciting to talk about. And it feels directionally correct, at least. So now you have an understanding of loosely how things work, why it is so important and why a lot of people are equally as excited about this opportunity. If you enjoyed this episode, please do not forget to share with a friend, with a family member who might also enjoy maybe electricity, maybe one of the other episodes that we've recorded this week. If you liked it, please don't forget to

rate us five stars on your favorite podcast platform where you get your podcast. And yeah, is that everything? Anything else each other before we let people go here? The final call is limitless has gone independent. We are keeping the lights on at home. And we would love to look for a partner. So if you are someone that has a product or service, or if you know of someone that has a product or service that would be amazing to share with our limitless audience, please reach out. Our email is in the description below or reach out to us on X. But that should be it. And we will see you, folks, on the next one. See you next time.

Thank you.