Elon Musk's New $119B TERAFAB Building Is Shocking Everyone
Chapters
Topic clips curated from this video. Click to jump in.
Description
Inside Elon Musk’s $119B TERAFAB Building — Nothing Like This Exists
Elon Musk is preparing to spend $119 billion on a single building called Terafab a chip factory spanning 100 million square feet that could produce one million silicon wafers every month.
The circle at the center of Terafab is reportedly a free electron laser a technology that could replace the exploding-molten-tin EUV light sources every other chipmaker depends on, eliminating the debris, the mirror damage, and the crippling per-machine electricity cost that keeps production bottlenecked.
This video shows you exactly why Musk has no choice but to build this, how the 100-million-square-foot layout actually works section by section, what the circle in the middle really does, where the power comes from to keep it running without a single blink, and why a 17-year Intel veteran is the one person who can make any of it produce a working chip at scale.
#TechInnovation #FutureOfTech #AdvancedManufacturing #TechSecrets #
Transcript
Read auto-generated transcript (4194 words)
Kind: captions Language: en Elon Musk is getting ready to pour $119 billion into a single building. One, it's called Terraab. It isn't a car plant and it isn't a rocket pad. It's a chip factory. And if it goes up the way it's drawn, nothing else on Earth comes close. The plans cover 100 million square ft. Running flat out, it could turn out as many as a million silicon wafers every month. A number big enough to make TSMC, Samsung, and Nvidia sit up straight. But the size isn't the most interesting thing on those drawings. Dead center in the whole complex sits a giant circle. Musk has hinted that what goes inside it could change how the world's most advanced chips get made and leave the method every other chip maker uses behind. What's planned for that ring caught almost the whole industry offguard. Stay for it because you'll learn what it is and how the rest of this place is supposed to come together around it. To see why anybody would spend that kind of money, you first have to see the wall Musk's companies are headed for. Without Terraab, they may simply stop growing. And the trouble starts with four businesses that look nothing alike on the surface, but share one enormous headache underneath. Why Musk needs his own chip factory. Musk has a lot of big projects going at once. And each one of them runs on computing power. Lots of it. Far more than anybody is making today. Over at Tesla, there's the Cyber Cab, a self-driving car built without a steering wheel or a single pedal for a person to use. There's also Optimus, a robot shaped like a person meant to walk around, lift boxes, and lend a hand at home or at work. At SpaceX, Starlink keeps spreading internet service across the whole globe. And the company wants smarter computers up in orbit, too. And then there's X AI, training giant artificial intelligence models that burn through computing power the way a furnace burns through coal. Cars, robots, satellites, AI. On paper, four different businesses. In practice, one problem. Every one of them needs advanced chips. And in numbers most of us can't picture. Start with the cyber cab. To drive itself safely, it has to make sense of camera video the moment it comes in. It has to spot people on foot, read the signals, find the lane markings, and keep up with streets that change second to second. Think about driving downtown at rush hour. You're watching for a kid chasing a ball, a car ahead, breaking hard, rain turning the pavement slick. Now take the driver out of the seat and every one of those calls lands on a computer that has a fraction of a second to decide and no second chance if it's wrong. Optimus is harder still. A robot on two legs has to stay upright, run dozens of motors, recognize what's in front of it, and hold a conversation without fumbling. You do all of that without a thought. A robot has to figure out every step, every shift of weight, every grip, one calculation after another. It's a video game on the hardest setting. Only the floor is real and so is the fall. Now, if Tesla only planned to build some thousands of these machines, it could probably keep buying chips from the usual suppliers the way it does today. That's not the plan. Musk talks about millions of smart machines, tens of millions. Add in what XAI and SpaceX need, and chip demand shoots past anything the world has ever seen. put an Optimus in every American home, and that alone would call for more chips than every phone and every laptop in the country put together. By Musk's own estimate, the world's entire output today might cover only about 2% of the AI chips his businesses will eventually want. 2%. So, this isn't fixed by sending TSMC or Samsung a bigger purchase order. When a plant can turn out a 100 chips a month, you can't buy 1,000 from it, no matter how thick your checkbook is, because money isn't the limit. The equipment to make the other 900 simply isn't there. It's like calling a machine shop that runs one lathe on one shift and asking for 10,000 parts by Friday. You can wave all the cash you want. That spindle only turns so fast. Musk's own way of putting it was blunt. Build their own giant chip plant or go without. they chose to build. And Terrafab isn't meant to stamp out simple chips. The idea is to do the entire job in one place, start to finish, without shipping half-built parts around the world. It will make specialized computing chips, high-speed memory, and advanced packaging, then test the finished parts and get them ready to ship. That's no longer a factory. It's closer to a city built for artificial intelligence. The Cyber Cab and Optimus are the big reasons this place has to exist. Those two will eat more chips than any product in history. And without Terraab, they could sit in a parking lot for years with empty sockets waiting on parts that nobody can make fast enough. If you like seeing how these things actually get built, hit like and subscribe so the next one finds you. Knowing the demand only tells you why. To get the how, you need a feel for the size of this thing. Next to it, every other factory on the planet looks like a model on a train table. 100 million square ft. 100 million square ft. Your mind slides right off a number like that. So, let's put it in terms you can walk around in. Start in Texas. Terrafab is drawn at close to 10 times the size of Gigafactory Texas. That plant already ranks among the biggest buildings on Earth. Thousands of electric cars roll out of it every week and walking at end to end eats up hours. Multiply it by 10. Now try Washington. Finished. Terrafab would be about 50 Pentagons. The Pentagon is such a big office building that the postal service gave it a zip code of its own. You could park 50 of them inside this one plant. In metric, that's roughly 9.3 million square m. Call it 1,300 soccer fields or better than 1,700 American football fields with the end zones counted. From an airplane, you wouldn't see a factory at all. You'd see a whole town laid out for one job. Picture those football fields buted edge to edge in one big grid. Then pull up the turf and put down clean rooms, machinery, and test labs instead. That's the view from above. Say you started at one end and wanted to reach the far wall. Bring comfortable shoes. Better yet, don't walk. The place may need its own transportation inside with vehicles and indoor roadways and whole sections will feel like a small town rather than a shop floor. Getting from memory production over to testing could mean catching a ride on a cart. Musk has even said it'll dwarf some of the country's most famous landmarks. He isn't after one more chip plant to add to the list. He's after the biggest thing anybody has ever built. Full stop. All that space is also why Terrafab is expected to bring in upwards of 3,000 good jobs, and engineers are only the start. A plant this size needs construction crews, operations people, shipping and receiving, maintenance, security, and plenty more just to keep the doors open. It's a small town with a roof on it. The folks working there may need their own cafeterias, break rooms, maybe even homes nearby. Here's what matters most about all that floor space, though. It isn't built around a single production line. Terrafab is laid out as a system with zones for making, testing, and finishing, all sharing the same walls. So those 100 million square ft can grow a piece at a time. When Tesla needs more chips, it doesn't break ground on a new plant. It opens the next section. The first phase may be modest, but the target is a million wafers a month. And hitting that takes room for clean rooms bigger than anything the industry has built. Those are rooms where the air gets filtered until not one speck of dust is left. Because one speck can kill a chip. On something that small, a fleck of dust is a boulder landing on a freshly poured driveway. Workers suit up head to toe so not a hair or a flake of skin gets loose. The plant also needs space for machinery, storage, packing, and testing, plus inside roads with traffic that barely lets up. The machines that print the chips are about as big as a school bus. Everyone runs hundreds of millions of dollars, and Terrafab will need several hundred of them, lined up in perfect rows like cars at a dealership. Fill 100 million square feet that way, and the $119 billion price starts to make sense. This isn't a building meant to impress. It's meant to keep growing for years, maybe for decades, one section after another as the orders come in. It's a machine that gets bigger right along with the world's hunger for AI. Still, the size is only half of it. Terrafab's real secret isn't how big it is, it's what's sitting in the middle, the circle in the middle. Spread out the plans, and one feature grabs you before the rest. In the middle of that sprawling complex is a giant circle. It may be among the costliest, toughest pieces of the whole project to build. And yet, almost nobody covering this story has said a word about it. Somebody online asked Musk whether a particle accelerator might be buried under that spot. His reply was a handful of letters, Felftw, free electron laser, for the win. Let that settle. If that hint is right, the ring could solve one of the biggest headaches in making advanced chips today. That's the surprise I promised you at the start. Here's the short version. No engineering degree needed to make chips smaller and stronger. Chip makers print incredibly fine patterns onto thin silicon discs called wafers. And they do it with light. It works a little like an old slide projector throwing a picture on the wall. Only instead of a picture, they're laying down circuits so tiny you could fit thousands on your fingertip. The smaller the feature, the shorter the wavelength of light you need. Today's best chips are made with a system called EUV. That's short for extreme ultraviolet lithography. And the light it uses measures about 13.5 nanome. Making that light is a wild business. A powerful laser hits little droplets of molten tin, 50,000 shots every second. Each hit turns the tin into a blazing plasma, and the plasma gives off EUV light. That light bounces off a string of mirrors so smooth they rank among the flattest things people have ever made. Blow one up to the size of the earth and its highest bump still wouldn't be as tall as you. It's amazing engineering. It's also a headache. It draws a staggering amount of electricity and the machines are so fishly complicated that keeping just one of them running is a full-time job for a whole team of specialists. And there's always a chance that stray bits of tin will fly off and damage those mirrors which run hundreds of millions of dollars a piece. A free electron laser goes at it from a completely different direction. No tin at all. It takes electrons and speeds them up to nearly the speed of light. Then they race through a long row of magnets, a device called an undulator. And as they wiggle back and forth through those magnets, they shed a beam of light. That beam can be steered with great precision and tuned to exactly the wavelength you're after. In theory, a strong enough machine could deliver a clean, powerful EUV beam with nothing exploding along the way. No debris, no danger to the mirrors. But the new light source isn't the whole trick. The bigger idea is making that light a shared utility for the entire plant. Right now, every EUV machine and a chip plant carries its own light source built right in. Think of a shop where every workstation has its own little air compressor chugging away underneath it. It works, but it's expensive, it's loud, and it eats up floor space you'd rather use for something that makes money. Now, picture one big compressor in the back room, piping air out to every station on the floor. That's the difference. A plant aiming at a million wafers a month will need a mountain of these machines. Giving each one its own light source would be a monster of a job. One free electron laser in the center could make a flood of EUV light and pipe it out to a whole row of machines at once. 50 machines, one light source sitting in the middle and feeding everything around it. That's why the circle matters so much. It isn't decoration on a site plan. It's the beating heart of Terrafab. It's also a big gamble. A machine that powerful is no small risk. And a chip plant is nothing like a car plant. It demands a level of care that's close to impossible to reach. That's exactly why Musk has to reach for help from a very unexpected corner. But first, there's another problem lined up. Power that never blinks. Say Terraab gets the chip design right. And say it gets the extreme ultraviolet light right too. It still has an even bigger hurdle ahead of it. A plant covering 100 million square feet with giant machines running around the clock needs power that never stops. Not for a second. A chip plant can shrug off plenty of delays. Losing power mid production isn't one of them. A drop out of even a few seconds can wreck every wafer sitting in the machines at that moment. Damage equipment that takes months to replace and cost a fortune before anybody can even reach for a phone. It's like losing power. or halfway through a weld on a pressure tank. You don't pick up where you left off. You cut it out and start over. Now, say that batch was worth $50,000 and carried 3 months of work. That's a blackout inside a chip plant. So, the real question isn't only how many chips come out the door, it's where the electricity comes from to keep a $119 billion machine humming day and night. What SpaceX is reportedly planning may surprise you. According to reports, SpaceX doesn't want Terrafab hanging on the Texas grid alone. The plan calls for natural gas power plants paired with big battery banks. In other words, a power station right next door to the factory. That may sound odd from companies famous for solar and batteries. Its sister company Tesla puts solar roofs on houses, power wall batteries in garages, and giant mega pack systems at utilities. So why wouldn't it run Terraab on its own products? One word, reliability. A plant like this needs what utility people call firm power. That's electricity that keeps coming whether it's sunny, pouring rain, or pitch blackout, in July heat, or a January ice storm. Solar makes plenty during the day and then goes quiet after sundown. Running a plant this size on sunlight and batteries alone right from the start would take a jaw-dropping battery bank and a mountain of money up front. You'd need enough stored power to carry a small city for 12 straight hours night after night. Gas hands the project one big advantage and that advantage is control. With its own plants on site, Terrafab doesn't have to sit around until the utility upgrades its lines. And it stays out of the shoving match. With every new AI data center going up around the country, those data centers are being built as fast as crews can pour concrete. Each one pulls as much electricity as a small town, and the grid is already straining. If Terraab had to wait in that same line, it could sit for years just waiting to get connected. A gas plant runs day and night and delivers a steady flow of power without asking anybody's permission. The batteries then work as a shock absorber. A factory doesn't pull the same power every second. Machines kick on, machines shut off, and the load swings. The batteries soak up those swings and keep the power smooth as glass. If the gas plant hiccups, they take over instantly until it's back. So, Terrafab isn't only a chip project, it's a power project, too. The AI race has moved past a contest over who can stockpile the most graphics processors. Those processors need chips. The chips need factories, and the factories need a river of steady electricity that never runs dry. So, if Musk wants a chip supply of his own, he has to own the power behind it as well. At $119 billion, Terrafab is shaping up as an industrial city that very nearly supplies its own electricity. But power and light still won't make a single working chip. For that, you need a very particular kind of knowhow. The peace money can't buy. A car plant and a chip plant might as well be on different planets. Tesla is very good at building cars with machines, faster than almost anybody. SpaceX is great at rockets and at changing a design on the fly. XAI is great at software and training AI models. None of that gets you ready to make advanced chips. Chipm takes rooms cleaner than an operating room. Chemicals that are nearly perfectly pure and control right down near the size of atoms. One small slip anywhere along the way can wreck an entire batch of wafers. Weeks of work gone in an afternoon. Laying down those microscopic lines is so touchy that a truck rolling by outside can spoil a chip. Some chip plants rest on huge shock absorbers just to keep the ground from shaking the machines. The air gets scrubbed so hard that one lone particle of dust is treated as a real problem and the chemicals are held to a purity where a single stray impurity can change how the chip behaves. Then comes the number that rules every chip plant. Yield. That's the share of chips that actually work when they come off the line. A brand new plant can start out with a terrible yield and nobody is surprised by it. Most of what it makes goes in the scrap bin, and it can take years of tests and fixes before the yield climbs high enough to earn money. At 50%, half your expensive materials and half your time goes straight in the trash. To turn a profit, a plant needs 90% or better. Getting there is among the hardest jobs in all of manufacturing. The biggest names and chips needed decades to learn it, and they paid for every lesson in scrap wafers. You can't write a check for decades of hands-on experience and have it show up next week. That's why Terrafab needs people who've been there. According to reports, the person set to run it is a former Intel leader who spent 17 years there. This person used to head Intel's 18A work, some of the most advanced chip making anywhere. Intel has been in the chip business for more than half a century. It's been through every problem you can dream up. From yields falling off a cliff to breakdowns that took months to untangle. Knowledge like that doesn't live in a binder on a shelf. It lives in the people who were standing on the floor at 3:00 in the morning when something went wrong and had to figure out why. Tesla brings the muscle to build at enormous scale. SpaceX brings speed and quick problem solving. XAI brings a bottomless appetite for computing power. So, the plant will never be short a customer. But the Intel veterans bring what none of those three has ever had. Real experience taking a cuttingedge chip process and getting it to run at huge volume day after day. Raising the yield is closer to a craft than a science. You have to know how each chemical, each temperature, and each machine setting plays against the silicon. No manual teaches that. You live in the plant, you watch the machines, and you run down every problem as it shows up, one at a time. A small change in temperature can decide whether a chip works or goes in the scrap bin. The Intel veteran heading Terraab spent 17 years learning those secrets, bringing in somebody who's already paid that tuition gives Terraab a shot at skipping the long, painful stretch of trial and error most new plants go through. And that's the final piece. The demand is there. The building supplies the room. All 100 million square feet of it. The laser in the center supplies the light. The gas plants and batteries supply the power. And the Intel veteran supplies the knowhow to make it all actually run. Terraab is an attempt at a whole new kind of industrial world. One that controls everything from the power station out back to the AI software running on the finished chip. If it works, it won't only change Tesla or SpaceX. It'll shake up the entire global chip business and show that one giant centralized plant can outdo the way chips have always been made. So, think back to those drawings. A 100 million square ft laid out like a small city. And right in the middle, that circle, every road, every clean room, and every machine on the plan points toward it. It's a $119 billion bet that robots, self-driving cars, and artificial intelligence are going to want more computing power than the whole world can supply today. The scale of it is almost too big to take in. But piece by piece, it's lining up. If you like this one, hit like, subscribe, and tap the bell so the next one reaches you. There's another story waiting for you right after this one. And here's my question for you. Can Elon Musk really pull this factory off? Or is $119 billion more than any one project can carry? Tell me down below. I'm Dale Harg Grove. The next one's already on the bench.