Elon Musk Was RIGHT All Along

Kind: captions Language: en SpaceX has already drawn up an AI satellite with wings wider than a jumbo jet. And on October 30th, the idea behind it, passed a big test in orbit. The best part is that Google, one of Musk's biggest AI rivals, did it for them. And to go big, they'd need SpaceX to hit a launch number nobody else is even crazy enough to chase. SpaceX's own plan starts with a problem Musk called out back in January at Davos. He said making AI chips could end up easier than finding the power to run them. A chip can be built and shipped, but a power plant needs land, permits, fuel, and years. Then he tied cheaper solar and cheaper launches to moving computers into orbit. To a lot of people, that sounded like a Musk fantasy. The numbers say he had a point. Every time you ask a chatbot something, a data center somewhere burns power to answer you. The IEA, the world's big energy watchdog, says data centers used about 1.5% of all the power on Earth in 2024. Its main forecast has that more than doubling by 2030. And AI is one of the big reasons. That's getting close to what all of Japan uses in a year just for data centers. The harder part is just getting plugged in. The IEA says about a fifth of planned data center projects could get delayed by grid problems. New power lines can take 4 to 8 years to build in places like the US and Europe. That means the chips can show up long before the power does. That's exactly the wall Musk was talking about. Tech giants have even started signing deals for whole power plants, which tells you how tight power is getting. It's the same grid your house runs on. So, this is a problem for all of us, not only the tech giants. Space has the opposite problem, which is way too much sun. There's an orbit that rides the line between day and night, following the sunrise around the planet. A satellite there sits in sunlight almost all the time. One big tech company did the math. A solar panel in that orbit can collect up to eight times as much sunlight over a year as a panel down here. It skips the nights, the clouds, and the low winter sun. So, it barely needs a battery. So, why isn't everyone already doing it? The doubters always come back to two questions. Can a normal data center chip survive a launch and the radiation up there where charged particles can flip bits and wear chips down? And how do you cool a hot chip with no air around it? Data centers down here spend a fortune pushing air and water through their racks. If the answer to either question is no, the whole idea falls apart. SpaceX's version included. SpaceX is building like it already knows the answers and it's building big. The power part is the easiest to see. A recent SpaceX listing for an iron worker spells it out. Building on-site power for the Starbase Supercomputer. The job covers steel frames, pipe racks, and platforms for gas turbines, plus the generators and piping around them. The turbines come on trailers and skids. They're the same kind of machines power companies use, just built to move. That setup means they can show up and start working fast instead of waiting years on a utility. Another listing wants a senior manager to build a power plant for, in SpaceX's words, launch and compute. That person would run the whole job from placing turbines to heavy lifts to handing the plant over for startup. The listings don't say how big the plant is or when it opens. But a supercomput stuck waiting on the local utility is just a pile of expensive chips doing nothing. So that's Musk's power argument showing up at his own rocket base. I'd say you don't hire iron workers to bolt down turbines for a computer you're only daydreaming about. Then come the lasers. A listing for a lead engineer on Starmind, SpaceX's AI satellite program, is all about building optical transceivers. Those are the parts that turn data into laser light and back again. The job title even says mesh optical, as in a mesh of laser links wrapped around the planet. The work means robotrun production lines, lining tiny parts up just right and testing them. The goals are yield, speed, and uptime, and yield just means how many parts come out good. A handful of lasers is a science project. Thousands of them is a business. SpaceX wants to make them by the truckload for Starlink gear and for AI compute. And SpaceX already has a head start because it flies thousands of laser linked satellites right now. It says it Starlink satellites carry three lasers links each at up to 200 Gbits a second per link. Those links pass internet traffic between satellites, which is how Starlink can reach places with no ground station nearby. Satellites doing AI work would need to talk to each other even faster. But SpaceX isn't learning lasers from scratch. It's also published its own AI satellite design called AI1. On paper, it's a giant with wings wider than a jumbo jets. Its computer is built to draw up to 250 kilowatts at peak, roughly what 200 American homes use on average. Whatever it works out would get sent home through Starlink's lasers. And it can take chips from more than one maker. So SpaceX isn't stuck waiting on a single supplier. Those are design numbers, not a satellite in orbit, but they tell you how big Musk is thinking. Put the pieces together and you get something nobody else has under one roof. the rockets, the satellite network, the lasers, and now the power. Amazon has satellites and a giant cloud business, but it buys its rides to orbit from other companies, SpaceX included. The last piece is the chips themselves. SpaceX is hiring managers to lock down chip supply with multi-year road maps, supplier capacity, and long-term contracts. That means buying years ahead so it doesn't end up stuck in the same line as everyone else. It's the boring part of building something huge. And it's usually the part that decides who actually ships. For now, SpaceX buys its chips like any other customer. But Musk is already pouring concrete to change that. Recent photos of the research fab going up near Giga Texas show huge steel trusses across the frame. There's more foundation work, fresh concrete, and more steel arriving. This is the smaller research site right by Tesla's Big Texas factory. SpaceX said it broke ground there in April. It's meant to feed the much bigger Terapab plan for Grimes County, Texas. The plan for that one covers about 100 million square ft, which is more than 1,700 football fields. Texas announced that main project back in August. The state says its first phase brings almost 17 billion and 3,000 jobs. It also handed SpaceX a $30 million grant. The plan puts processors, memory, and final assembly all on one campus. It names chips called AI5 and AI6 for use on Earth and one called D3 for space. D3 is the one that tells you where Musk thinks a lot of these chips end up. The long range goal is chips adding up to a terowatt of computing power every year. That's about what a thousand big nuclear plants put out. It's a goal, not a building yet. But to me, it's Musk covering both ends of his own argument with power plants to run the chips and a fab to make them. Building things is the part SpaceX controls, though. The rules for its phone network are a whole other fight, and that one just heated up at the FCC. The FCC just voted to propose letting satelliteto phone service use 13 bands instead of five, and the vote was unanimous. That's the service where satellites borrow a carrier's airwaves to reach a normal phone. More airwaves means more phones can connect at once. The FCC also wants looser license areas and sharing deals that fit each company instead of the current rigid lease setup. It's still a proposal and comments open for 60 days once it's published, but it opens a path where a company holding the right ground licenses doesn't have to lean on one big carrier for you. That could mean a signal from space in a lot more places. SpaceX is fighting over the global version of those rules, too. in a filing for the next World Radio conference. It takes aim at T-Mobile's plan for protecting cell towers. If you are enjoying the video, please hit subscribe to support the channel. It genuinely means the world to me. SpaceX argues that plan could create no-go zones reaching more than 240 m out from towers near borders. If you live near a border, that's the difference between having satellite coverage on your phone and having none. And SpaceX's next batch of phone satellites is planned to start going up on Starship in 2027. So these fights matter right now. Credit where it's due. The carrier does back opening more airwaves to satellites. It just wants carriers to keep the keys and it says its limits simply protect the networks already there. Nothing's decided, but you can tell who's nervous about whom. All of that still leaves the doubters two questions hanging. Can a real data center chip actually work up there? And can anybody keep it cool? On October 30th, it happened. A satellite in orbit answered a question from Earth using a real data center AI chip. And that satellite belongs to Google. The team sent a prompt up to an AI model running on the satellite's own chip. As it passed over a ground station on Spalbard, way up in the Arctic north of Norway, the answer came down as a hi coup. Blue marble below data streams in silent grace. Starlight feeds my core. Honestly, that's the most robot thing a computer could ever say. And it's also a little creepy. The project is called Suncatcher, and it's behind that sunlight math from earlier, too. The chip is a TPU, the same kind of AI chip Google runs in its own data centers. The model was Gemma 4, also from Google, and it's the kind anyone can download and run at home. That first answer came after weeks of switching the hardware on and testing it. Since then, the team has tried Limmerick and even asked it how it's holding up. The thinking happened in orbit. The chip took a request, worked it out up there, and sent the answer home. So, the doubter's first question just got answered. The chip survived the shaking of a launch, switched on in space, and did real work. Google actually started this project before Musk's speech. That means two very different outfits looked at the same problem and landed on the same answer. Google isn't even the first either since a startup called StarCloud flew an Nvidia H100 chip back in late 2025. And here's the part that should make every SpaceX fan grin. Google's satellite got to orbit on a SpaceX rocket. It launched at the start of October on Transporter 18, a SpaceX ride share out of Vandenberg in California. That's a flight where dozens of small satellites share one Falcon 9, basically a carpool to space. It rode up with about 20 satellites from Planet, the firm Google built it with. Planet said that was its 40th successful launch, and it confirmed first contact right after the satellite reached orbit. So, the satellite that just backed up Musk's argument hitched a ride on his rocket. The second question, keeping it cool, is still the hard one. Space has plenty of free sunshine and zero air conditioning, so you can't just blow a fan across a hot chip. Google's testing heat pipes and cooling panels that let the heat glow off into space. Radiation isn't fully beaten, either. Google blasted its chips with a proton beam, and the memory took it harder than the parts doing the math. Now, the team knows where the weak spot is. If you ask me though, heat is the sneaky hard part because more AI work always means more of it. That's why Google's taking it slow. Its test satellite reportedly carries a handful of these chips and makes up to 1 kilowatt of power running one chip at a time. SpaceX's [clears throat] AI1 is designed to draw about 250 times that. Google's next step is a pair of satellites flying together in 2027. Its computer models picture whole clusters flying within about a mile of each other because a laser link gets weaker over distance. In the lab, Google's pushed 800 gigabits a second each way over a laser link. Real AI work up there means lots of satellites talking to each other over lasers. That's exactly what Starlink does every day. And then comes the number. In Google's research paper on scaling this up, the price of launching hardware is one of the makeorb breakak numbers. AI satellites are heavy with all those solar and cooling panels. Google's math has launch prices falling under about $90 a pound around 2035 if costs keep dropping the way the paper models. And it ties that to Starship flying about 180 times a year. That's a Starship launch about every 2 days, and it's the number only SpaceX is even trying to hit. Starship's flown 14 times in its whole life. But Falcon 9 already launches about every 2 days, and the more a rocket flies, the cheaper each flight gets. So, Google's own engineers wrote a business case for AI in orbit, and a Starship flight schedule sits right in the middle of it. I don't know about you guys, but to me, that's where this stops being Google's story. Google can own the chips, the model, and the satellite, and SpaceX could still get paid to carry all of it. Starite feeds my core. Sure, something still has to haul the core up there. And right now, that something looks a lot like Starship. Google even ran the numbers on what it would cost. At those prices, launching a Starlink style satellite works out to about $810 per kilowatt of power each year. Data centers down here pay about $570 to $3,000 per kilowatt a year just for power. That's only one piece of the bill, but it puts orbit in the same ballpark. And if launches ever get that cheap, everything from weather satellites to space stations would get cheaper to put up, too. The doubters said the math would never close. Google's own paper shows a path where it does if Starship flies the way SpaceX says it will. There's just one problem. The whole industry is already running low on rockets. A space research firm figures the world will come up about 150 medium and heavy launches short over the rest of the decade. In its model, 2028 is the worst year. It lined up the missions people want to fly against how many launches each big rocket can actually handle. It reportedly has Blue Origin's New Glenn topping out at 16 flights a year. And it says Starship could change the picture. For the rest of us, that means new GPS, TV, and internet satellites could all wait longer for a ride. Take Telesat. Its CEO just said it doesn't want to burn a whole Falcon 9 on only the first two satellites of its Lightseed network. They'll ride on someone else's Falcon 9 around June 2027 instead and full global service slides back about 3 months. Lightseed wants to compete with Starlink and it's waiting on a ride from the company that owns Starlink. The CEO even said getting on to Starship later might mean building satellites to fit the same rack SpaceX uses for its own Starlinks. Meanwhile, Falcon keeps grinding. On October 31st, it flew its 700th launch with one booster making its 18th flight. I think that's the real point. The world's about to want way more rides to orbit. And the company with the most proven rocket is also building the next one. Even Starlink is feeling the squeeze. And you can see it in the bills. Starlink is testing prices that move with how busy your area is, especially in parts of Missouri and Kansas. In one example there, the top home plan jumped 20% to $156 a month. One Kansas town got offered the same plan for 104. Customer emails reportedly say prices will be reviewed about every quarter, and some go down as well as up. If your neighborhood's crowded, you pay more. And if it's quiet, you might pay less. Business customers get hit harder. Starlink's biggest global business plan gets a new name and goes up nearly 40% on December 1st. Its regular business plan also cuts the time you can use it abroad in half. That's what demand looks like when the sky is the warehouse. More satellites fix it, but they need more rockets. And the rockets need more ground. And down on the blue marble, SpaceX is running into the edge of the map. On October 31st, a draft agenda for the Texas Parks and Wildlife Commission was reported to include a possible land sale. It covered about 1300 acres in Cameron County, where Starbase sits. The item got pulled before the final November list came out. The agency is also said to have tried to keep the draft from being released. Nothing's been signed and Texas Governor Greg Abbott said recently the board wouldn't take up a sale anytime soon. The state park itself is about 1,60 acres, so the two numbers don't line up exactly. Starbase annexed more than 7,000 acres in September. That gives the city more control, but it doesn't hand SpaceX any state or federal land. Starbase builds, tests, moves, and launches rockets. And each of those jobs need space. That land is also a public park and a beach people love, which is why this one's touchy. Honestly, it says a lot when a rocket company starts running out of beach. Back in January, Musk said the chips would be the easy part and power would be the wall. Now, a data center chip is working in orbit on sunlight, and the company behind it is leaning on Starship to make it big. That's about as clear a told you so as Musk could ask for, and it came from one of his biggest AI rivals. So, what do you think?