Elon Needs More Power Than Earth Can Give Him

Ajouté : 2026-05-25

[00:00:00]StarCloud one separation confirmed.

[00:00:02]>> The problem is that data centers take up a ton of space and they need a huge amount of energy.

[00:00:08]>> This is the beginning of a future where most new data centers are being built in space.

[00:00:13]>> The world is about to be surrounded by over 1 million satellites. Elon Musk will get paid over a trillion dollar for it. And the power required to power these satellites will be the same as 85 billion homes in the US. Yes, all of what you just heard is true and it's happening and almost nobody knows it's happening and even less people truly understand the insane implications of this. But after watching this video, you'll be part of the very lucky few. On January 30th this year, SpaceX filed paperwork with the FCC to launch 1 million satellites into Earth's orbit.

[00:00:49]Humanity combined thus far has launched just around 14,000 satellites in total ever. Now, that same week, a startup literally nobody has heard of, confirmed that a single Nvidia H100 chip, the exact chip running every major AI lab on the planet, was sitting in orbit, 325 kilometers away above Earth, running a full AI model, transmitting the output back to the ground. Then a few months later, the CEO of Enthropic got on stage at his own developer conference and admitted that his company had planned for roughly 10 times growth, and what they actually saw was a freaking 80 times growth. The reason they couldn't keep up, in his own words, was basically compute. They literally could not get enough of it to run their AI on Earth right now. Now, hold all these things together in your head. You got 1 million satellites, AI running in orbit, and the fastest growing software company in history saying the ground isn't enough.

[00:01:43]And as a bonus, there's one sleeping giant that nobody in this conversation is even watching yet. By the end of this video, you're going to understand why all four of these things are pointing at the same place. I've been covering this space for roughly 14 years now. And honestly, this is easily one of the craziest things I've ever seen in my life. And I've seen some serious stuff.

[00:02:04]But before we get there, I need to show you something first. Because nothing that I'm about to explain right now makes any sense unless you understand the problem that these companies are actually trying to solve for. And the problem is way more fundamental than you think. And the way it's going to impact all our lives is not well understood at all by 99% of the people alive today.

[00:02:29]Let's start with a number. This will help you understand the scale of this problem. $725 billion. That is what Meta, Amazon, Microsoft, and Google have committed to spend on data centers to power AI systems in 2026 alone. That's a 77% jump over last year's record of 410 billion.

[00:02:48]And you've got the four richest companies in human history funding that with three4ers of a trillion dollars on just one thing in just one year. When you add everyone else, 2026 is the first trillion dollar year for compute spending in all of human history. And you might think with that kind of money, any problem is solvable, right? And the shocking thing is that that level of spending is nowhere near close enough to the amount of money that's needed to solve for the problem these companies are trying to solve for. And if a trillion dollars can't fix it, what's actually happening is something structural. Something is broken. The actual bottleneck is something way more boring. And it's just electricity.

[00:03:29]Freaking boring ass electricity. These AI models that all of us use today are so power hungry that entire city grids cannot keep up with their demands. The IEA projected data centers will eat 1100 terowatt hours in 2026. That is the entire electricity consumption of Japan.

[00:03:47]It's equal to an entire country just to run servers. To be honest, that's freaking shocking. It really is crazy.

[00:03:52]Now think about Northern Virginia. There are 603 data centers in that region.

[00:03:57]Virginia is basically the freaking straight of hormuse for AI at this point. Then Dominion Energy, the company that's powering all of the data centers in that region, came out publicly and said they are massively constrained to power new data centers. The biggest data center hub is already maxed out until 2028. So just sit with that for a second. The most powerful companies in the world with basically unlimited money cannot build fast enough to keep up with the demand because there is literally not enough power left to give them. Now what's interesting is the country of Singapore tried to be the alternative.

[00:04:30]They eventually faced the same problem and capped new data centers at five megawatts only which is equivalent to one small data center. Amazon and Microsoft need 8 to 24 megawatt just to turn the lights on and run basic operations. The rest is just sitting there delayed or even cancelled. And it's not because these companies ran out of money. All the checks are written and the money is there. It's ready to be deployed. Billions are literally just sitting there waiting to be spent today.

[00:05:00]What's actually stopping them is energy with land, water, and regulatory hurdles to turn these centers on. Things you cannot just manufacture or fund raise your way around, especially today with AI approval ratings in the US being at or below ice. It's a tough time right now to build more AI capacity on Earth.

[00:05:17]You even have companies like Microsoft signing the largest corporate nuclear energy deal in history just to guarantee future power supply. They found a decommissioned nuclear plant called 3M Island and said, "Yeah, we'll just fire it back up because it's the same one that set the US into a freaking frenzy back in the 70s when it melted down."

[00:05:36]Which, by the way, with the benefit of hindsight, was probably one of the biggest overreactions and massive missteps by the US in its history. It landed the country in this current position that it's in with extremely old, decrepit, and wholly incapable energy grids to keep up with the insane growth that the world is experiencing right now with the AI boom. We are so behind. Real quick, if you're enjoying this video, about 75% that watch my channel often aren't subscribed. And the easiest and free way to support this channel is by clicking the subscribe button right below this video. Thank you so much. So, let's just be really clear about what's actually happening. There are some laws being worked on right now to slow down this AI thing because there's fears around job loss and some of that is warranted, but as of right now, this is a problem about the limits of our current energy generation systems, which makes what Anthropic just did next so interesting. Anthropic are the makers of Claude. Their revenue went from 9 to $30 billion in roughly 4 months. That's the fastest ramp in software history. Nobody's even close.

[00:06:37]Then Daario, the CEO, gets on stage at their own developer conference and admits out loud that they were expecting a 10 times growth this year. And what they actually discovered was that they were experiencing an 80 times growth forecast. He called it just crazy and too hard to handle. Their users are growing so rapidly that they know this was inevitable. Then through a deal with SpaceX, Anthropic publicly said they want to develop multiple gigawatts of orbital AI compute capacity, which means SpaceX will use its latest rocket called Starship to send literal data centers in space to use the infinite power of the sun to process the compute necessary for their AI systems and beam the intelligence back down to Earth for us to use. Now, it's very interesting to have Daario, the CEO, when someone like that that's running a giant company like Anthropic, they're looking at everything that's happening in Virginia and Singapore and the freaking nuclear pants with Microsoft and concludes that the ground is not going to be enough to power the data centers. That is a signal worth paying attention to. There is a permanent answer to this problem. It just might not be on the ground. And the wow thing is that it's already happening.

[00:07:47]A company basically nobody had ever heard of launched a satellite in November of 2025 in space called StarCloud. Inside it was a single H100 Nvidia chip. That chip is the one running on every major AI lab right now.

[00:08:00]They ran Google's GMA model, a full large language model in orbit. It got processed up there and they got responses back down here on Earth. And the first thing it said was, "Greeting Earthlings," which is either the most on brand thing a rogue AI could say in space or the best marketing moment of 2025. Maybe both. Either way, remember that moment because it matters more than it sounds right now. StarCloud was founded in January 2024. It was backed by the obvious investors in Nvidia and Y Combinator. They also trained a tiny model called nano GPT on Shakespeare up there, making it officially the first model ever trained in space. actually insane to say out loud. But that's actually not even the most important thing that they proved. Here's what this actually proves. Space chips and data center chips are different categories.

[00:08:49]Space chips are very slow because they have to be radiation hardened. They're built to survive. Data centers on Earth are the opposite. They're super powerful, but they're very fragile. This is because space is much harsher, and it requires way more shielding and different processes to make sure the chips can actually survive the radiation in space. This has traditionally made space-based compute way, way, way harder and way more expensive. Nobody really tried to mix those things together thus far up to this point because it's so hard. But StarCloud did very recently.

[00:09:20]Everything I'm about to describe is built on this insanely gigantic moment.

[00:09:25]If you can run actual AI in space, why is everyone down here still fighting over land and power grids and water rights?

[00:09:37]But first, I want to talk about what could actually kill this whole idea.

[00:09:41]There are real problems. Starting with radiation. Running one H100 chip for demo is super cool, but running a 100,000 GPUs doing non-stop AI training for weeks in a radiation environment is a completely different problem. Google tested their chips in a proton beam simulating 5 years of cosmic radiation, and the chips survived. But the memory showed real sensitivity. It didn't actually kill the chips, but it was very impactful. Then you've got maintenance.

[00:10:08]A dead GPU on the ground gets swapped out in 20 minutes by some guy that probably is hung over or hasn't had his coffee yet. In orbit, you just ride it off. And it's not just a dead chip either. It's the chip plus the full launch slot to replace it. That math is brutal. It's genuinely brutal. And that is the reason why this conversation has not really happened until very recently.

[00:10:28]Then refresh cycles. An H100 launch today would be multiple generations behind. by the time you'd wanted to actually replace it. You can't exactly send someone up there right now with a screwdriver, you know, to swap out the chip. And so when you think about these problems, you're like, "Okay, that's genuinely worrying. Maybe this doesn't actually work. Maybe it's just a cool demo and nothing more." But then you think about the solution and it's one answer. Every single one of these problems has the same exact answer.

[00:10:56]Launch costs. That's it. If you have cheap enough launches, the maintenance problem solves itself. When it costs almost nothing to send stuff into space and you're able to make robots that automatically swap or replace these chips, this stops being science fiction and it becomes a no-brainer. Especially with the advent of humano robots about to take off, which we've covered very deeply on this channel. The redundancy problem solves itself. It goes from impossible to hard. and hard is completely different than impossible.

[00:11:28]Literally, think about it. Electricity wasn't impossible when it first was invented. It was just very difficult.

[00:11:34]Same thing with the railroad or the internet or fire. Space-based computer is exactly the same thing. And there's exactly one company on Earth that controls launch costs. And what that company has been quietly building in the background while everyone else was watching the AI news just like you are is something I genuinely was not prepared for.

[00:11:57]Most people think SpaceX is a rocket company. Some people are not realizing it's also an AI company. And what's funny is both of those are still selling it short by a lot. Here's what they've actually been quietly building. While everyone was watching rockets land and losing their freaking minds on X and on YouTube, a lot of things have been happening. Layer one is launches. Falcon 9, which is SpaceX's workhorse, already costs around $2,700 to $3,000 per kilogram to orbit. It's the cheapest in the world by far, but Starship is targeting under $200 per kilogram and eventually under $20. For comparison, the space shuttle costs $54,000 per kilogram. We're going from 54,000 to under 200. Those economics sound like they are from a freaking different universe. Get it? Rockets universe.

[00:12:44]Every single obstacle we just talked about gets dramatically more solvable the moment that number lands. Layer two is chips. On March 21st, 2026, Elon got on stage at a decommissioned power plant in Austin, which is kind of hilarious if you really think about it about this whole electricity discussion. And he announced Terraab. This is a joint chip manufacturing factory between SpaceX, Tesla, and XAI. Phase 1 will cost about $55 billion with a full buildout up to $120 billion. The exact quote from Elon was, "We either built the terrafab or we don't have chips and we need the chips, so we built a terafab." This is actually a very classic Elon quote. He looked at the global chip supply chain and he thought it was someone else's problem and he thought there wasn't enough chips and then he made it his own problem on purpose to solve for those chips. The reason why is most of Teraf's output is earmarked for space, not for the ground.

[00:13:36]So this isn't even about competing with your TSMC's, your other chip makers on Earth. It's about building chips for a data center network that doesn't exist yet in space. Now, write that down because it's about to mean something completely different in about two minutes. I promise. Now, for layer three, you have compute. SpaceX already has a genuinely insane amount of compute running today on Earth with Colossus in Memphis, Tennessee, which is sitting at around 555,000 GPUs with 18 billion invested. Then Anthropic just rented all of Colossus 1 with 220,000 GPUs and 300 megawatt, which is enough to power 250,000 homes. Isn't that crazy? The deal specifically mentioned orbital compute.

[00:14:17]So, the company that needs to go orbital just partnered with the only company that can actually do that today. That's obviously not a coincidence. And layer four, you have satellites with over 10,000 of them already functional in orbit with Starlink, which is a part of SpaceX. SpaceX did more than 80% of all global rocket launches last year. And let me say that again, one company, 80% of the launches. So there is a massive precedent that SpaceX getting to a million satellite launches is not only probable at this rate, it is inevitable.

[00:14:48]Now stack all those together, you have launches, chips, compute, and satellites. There's no other company in history that has ever owned that full stack. So SpaceX is literally the only company that's capable of bringing this reality to life in its present state.

[00:15:02]There are other companies out there trying to compete with SpaceX. You have Blue Origin, you have Rocket Lab, and a host of different Chinese companies. But as of today, SpaceX stands alone by a country mile and then some. Now we have to ask ourselves the question of why?

[00:15:18]Why does this make so much sense? This already sounds freaking insane, but what if I told you that physics makes this an inevitable outcome? And it's because of that power problem. The reason why is in orbit, solar panels collect energy at roughly five times the efficiency of ground systems because there's no atmosphere and no night cycle. You cannot replicate that on Earth. You can put these satellites up there in space that will orbit around the Earth and then it will always point itself towards the sun, literally catching the sun 24/7. But the question becomes, how do you get rid of the heat that your chip will generate while you use the power from the sun? You're just going to let your chips melt in space? How do you solve for cooling? Let's look at how we do it on Earth today. Google's Oregon data center uses 2.3 billion gallons of water every year just to stop GPUs from melting. a gigantic amount of it in a contained recycled loop. But in orbit, you point a radiator at deep space at -270° C and your cooling bill just drops to zero. You just dump the heat with radiative cooling instead of using complex water cooling that could destroy your entire rack in case of a leak.

[00:16:27]Meanwhile, down here, we're building freaking nuclear plants to run air conditioners for data centers that run AI. Hilarious. Now, here's where it gets genuinely insane. SpaceX knows this is inevitable. They are going public with a target date of June 12th, 2026, targeting a $75 billion raise at an over $2 trillion valuation. It will be the largest IPO in stock market history by far. Saudi Aramco held the record last at 29 billion. SpaceX is going for over 50 times that, 50 times larger than the previous record held by a company going public. This is absurd. But buried inside the IPO documents is one line that tells you everything. Elon gets 60 million additional shares if SpaceX deploys space-based data centers with a 100 terowatts of computing capacity, which is equivalent to, are you ready?

[00:17:18]85 billion average US homes. Yes, you heard that right. And what's even crazier is that if Elon were to deliver that and hit the SpaceX valuation target of over 6.6 trillion, the payout will be equivalent to $1 trillion just for that one goal alone. The orbital compute future is written into the largest IPO in financial history as a core performance metric. This is inevitable. But there's one more player nobody is talking about. Now, before I get back to something I mentioned at the start. If you are a content creator or always wanted to be one, I've made my entire content creation tool available to everybody at farzza.ai, which I use for this video. This tool has helped me reduce the time I spend creating content by 80% and it has 10xed my views. If you're interested, I have the link for that in the description in the comment section below where you can use Farza 10 for 10% off. Thank you so much. So, everyone assumed this race was between SpaceX and maybe Google, maybe Microsoft, maybe a bunch of players in between, but nobody was seriously watching Amazon. And I get it. You're probably like, "Okay, Amazon, the e-commerce store and the AWS guys that run like freaking half the internet, right? How is anyone not watching Amazon?" But here's what people are getting completely wrong. They're watching the wrong part of Amazon because it's not really about Amazon.

[00:18:39]It's about Jeff Bezos. Bezos got on stage at a conference in Italy last October and said, quote, "These giant training clusters will be better built in space because we have solar power there 24/7. There are no clouds and no rain and no weather." I hope Jeff had some nice pasta while he was in Italy.

[00:18:57]I'm kind of jealous. Then when he came back to the US, he filed for that same exact idea through three completely separate companies simultaneously. First was Amazon LEO. As of April 2026, they have 302 production satellites in orbit.

[00:19:11]And I know what you're thinking. SpaceX has over 10,000. That gap is enormous.

[00:19:15]And I'm not going to pretend otherwise, but everyone is missing the actually important thing about those 302 satellites. Every single one is natively integrated with AWS. That's a massive difference and it matters a lot. AWS already runs the cloud infrastructure that powers most of the internet. So, Amazon doesn't have to build customer relationships from scratch the way SpaceX does. They can just integrate their AI directly in hundreds of thousands of customers, perhaps even millions. Now, remember that enthropic deal we walked through earlier with the 220,000 GPUs and the 300 megawatts.

[00:19:48]Remember that thing? The deal that specifically mentioned orbital compute, that API traffic, the actual inference requests people are sending to Claude right now is already running through Amazon's infrastructure today before a single orbital compute node exists.

[00:20:03]SpaceX has to earn that customer relationship from scratch. Amazon is already the landlord and it has been there for years. That's a massive advantage. And if you're not convinced yet, check this out. On March 19th, 2026, Blue Origin filed with the FCC for something called Project Sunrise with up to 51,000 satellites specifically designed as orbital data centers for AI computing. They're connected to a separate constellation called Terra Wave with 5,400 laserlength satellites providing ultra high-speed connectivity between the orbital data centers and the ground. The deployment for that project is starting in Q4 of 27. Then there's New Glenn, which is Blue Origin's heavy lift rocket. Its third flight was April 2026 and they successfully reused the booster for the first time, making Blue Origin only the second company in history to do that after SpaceX. The payload ended up in the wrong orbit.

[00:20:54]Oops. So, the mission wasn't perfect, but the reusability piece worked perfectly. Now, here's why that matters more. We spent a lot of time earlier on Starship hitting under $200 per kilogram, and why that number changes everything. Every problem gets solvable.

[00:21:10]every obstacle gets smaller. And Bezos just proved his rocket can do the one thing that unlocks that same math. He doesn't need to match SpaceX on cost tomorrow. He just needs to stay close enough that he never has to write a single check to Elon. That independence alone is worth more than most people are pricing in right now. Not investment or financial advice. And the competition between these two is going to be absolutely epic, which means all of us win in the long term. So, here's what I actually want you to watch over the next 12 to 18 months. Watch the SpaceX IPO documents closely, specifically around the orbital compute milestones. That one line tells you more about where this is going than anything else. Then watch Bezos. When someone moves that quietly, that quickly, pay attention. And if you're a Tesla investor, the amount of stuff happening in tech and AI right now is genuinely overwhelming. Every week there is a new news cycle that supposedly changes everything. Welcome to the channel. And the psychology of holding through this kind of compounding information is actually harder than the analysis itself. The mental game is what trips most people up. Bradford Ferguson and his team at Rebellionaire have an entire video on this called why most people could never hold Tesla stock. It is not about the numbers. It is about the mental game about sitting through years of doubt while a thesis takes time to play out about not selling on every news cycle. I will link that video in the description below. Even if you're not a Tesla investor, I highly recommend you watch that. It's unbelievably helpful. And these guys are the same guys who called Elon selling his shares a month before that famous Twitter post, so they know what they're talking about.

[00:22:45]Thank you so much for watching. I hope this video was informative and helpful.

[00:22:48]And we'll see you on the next one. Take it easy, everybody. Uh, bye-bye.