← All videos

What Happens Inside Elon Musk's $120 Billion Terafab?

Future Atlas 2050 Published Oct 15, 2026 Added 1w ago 24:50 804 views Open on YouTube ↗

Description

What if the most important part of a $119 billion chip factory isn’t the factory at all—but a mysterious circle at its center?

Terafab is being planned on a scale almost impossible to imagine: roughly 100 million square feet, designed to eventually produce up to one million silicon wafers every month. But behind the staggering size lies a far more ambitious idea. A single facility that could bring chip production, memory, advanced packaging, testing—and even its own power generation—under one roof.

Then there is the clue that changes everything: three letters from Elon Musk. “FEL.” A free electron laser could, in theory, provide the extreme ultraviolet light needed for advanced chipmaking from one central source, feeding multiple machines across the complex.

But can such an extraordinary system actually work? And can decades of semiconductor expertise be compressed into a new industrial machine of this scale?

Subscribe to Future Atlas 2050 as we explore the technologies, megaprojects,

Transcript

Read auto-generated transcript (4429 words)

Kind: captions Language: en Look closely at the plans and you will notice something strange. Surrounded by a grid of buildings so large it is hard to say where one ends and the next begins. There is a circle, a perfect deliberate circle drawn right at the heart of the complex. It is not a courtyard. It is not a fountain. And for a while almost nobody asked what it was for. Then someone did. Online a person asked Elon Musk a simple question. Could there be a particle accelerator hidden beneath that circle? Musk did not write a long explanation. He wrote three letters FEL and then three words for the win. FEL stands for free electron laser. It is a machine that as far as we know, no chip factory anywhere on Earth uses to make chips today. If that reply means what it appears to mean, the circle at the center of this project could become the beating heart of an entirely new way to build the most advanced chips ever made. The project is called Terraab. The estimated price is $119 billion, not for a fleet of rockets, not for a constellation of satellites, for one building. A building designed to cover 100 million square ft. a building planned to produce up to 1 million silicon wafers every single month. An output that would place it alongside the largest chip makers on the planet, companies like TSMC and Samsung. But the size is not the real story. Neither is the price. The real story is why a man who already runs car companies, rocket companies, satellite networks, and an artificial intelligence lab would decide that the only way forward is to build a chip factory larger than anything humanity has attempted. And why at the very center of it, he might place a machine borrowed from the world of particle physics. To understand that circle, we first have to understand the wall that every one of his companies is about to hit. At the same time, this is future Atlas 2050. On the surface, Elon Musk's companies have very little in common. One builds electric cars. One launches rockets and thousands of satellites. One trains artificial intelligence models. One is teaching a humanoid robot to walk. But strip away the logos. And underneath every one of them sits the same small, flat, fragile object, a chip. Consider the cyber cab. The vehicle Tesla is developing to drive itself with no steering wheel and no pedals at all. Picture a busy intersection at rush hour. A cyclist drifts toward the lane. A pedestrian hesitates at the curb. A delivery van breaks without warning. A human driver handles this almost without thinking. Years of experience let the brain read the scene and react before a conscious thought has even formed. A machine has no such shortcut. Every frame from every camera must be read. Every object must be named. Every movement must be predicted. And a safe decision must be made in a fraction of a second again and again for the entire journey that requires a powerful chip. And every single car needs its own. Now consider Optimus, the humanoid robot standing upright on two legs, carrying a box up a flight of stairs, pushing open a door. For us, balance is invisible. For a robot, it is a constant stream of calculations, every joint angle, every shift of weight, every placement of every foot. Again, a powerful chip and again one for every robot. This is where the arithmetic begins to break. Musk is not describing a few thousand cars or a few thousand robots. He is describing millions, tens of millions. If even a fraction of households one day owned a robot like Optimus, the number of chips required could dwarf what today's entire smartphone industry consumes. And that is before we add the enormous computing clusters his AI company XAI needs to train its models or the ambition to make SpaceX satellites smarter in orbit. Musk has put a number on the gap. By his estimate, the world currently produces roughly 2% of the AI chips his companies will eventually need. Not half, not even a quarter, 2%. Here is why that number matters so much. Most problems can be solved with money. This one cannot. Imagine walking into a small bakery that makes 200 loaves a day and ordering 10,000 for tomorrow morning. Your money is perfectly real, but the ovens are not there. The flower is not there. The bakers are not there. No amount of cash can conjure up capacity that was never built. The chip industry works the same way. The world's existing factories are already committed, often years in advance. So Musk faced a choice. He could keep asking other companies for a larger share of a supply that would never be enough. Or he could build the supply himself. He chose to build and not just a chip factory. Terrafab is planned to bring the entire process under one roof. Computing chips, memory, advanced packaging, testing. In the chip world, these steps are usually spread across different companies, sometimes on different continents. Terrafab would fold that whole chain into a single place, which raises an obvious question. How big does a building need to be to hold an entire industry inside it? 100 million square ft. It is the kind of number the human mind simply refuses to picture. So let us approach it slowly, one step at a time. Start with Gigafactory Texas. It is already one of the largest buildings on the planet. Walking from one end to the other is not a stroll. It is a journey. Standing inside, most visitors describe the same feeling. They feel small. Now take that feeling and multiply it by 10. Terrafab is designed to be roughly 10 times larger. Still abstract? Try this. Converted to metric, 100 million square ft is about 9.3 million square me. That is roughly the area of,300 soccer fields laid edge to edge. Now imagine replacing every blade of grass on every one of those fields with ultra clean rooms, testing labs, and machines that print patterns smaller than a virus. seen from above, it would not look like a factory. It would look like a city. A city with exactly one job. And like any city, it would need a way to move around. At this scale, walking from the memory production zone to the testing zone on the far side would be impractical. The plans suggest an internal transit system. Small electric vehicles traveling along indoor roads. neighborhoods, each devoted to a different stage of chipmaking, each with its own climate. One zone held at one temperature, the next running under entirely different conditions. Crossing from one to another might feel like stepping from summer into winter without ever going outside. It would have its own rhythms, its own traffic, its own daily routine, and its own population. Terraab is expected to create more than 3,000 jobs from construction crews and machine operators to repair technicians, logistics staff and security teams. People who will need places to eat, to rest, and perhaps to live nearby, but the size is not vanity. It is strategy. Terraab is designed to grow in phases. Rather than breaking ground on a brand new factory every time demand rises, Tesla could simply open the next section of the building and switch on the machines inside. The early phases might run at a fraction of capacity. The long-term target is that figure from the beginning, 1 million wafers a month. Reaching it demands clean rooms on an extraordinary scale. In a clean room, the air is filtered so intensely that a single speck of dust is treated as a threat because a single speck of dust can ruin a chip. Workers move in full body suits so that not one flake of skin, not one strand of hair ever reaches the air. Imagine painting a portrait smaller than a grain of sand while standing in a gentle breeze. That is the level of care involved. And then there are the machines. The printing machines at the core of modern chipm are roughly the size of a school bus. Each one can cost hundreds of millions of dollars. A factory chasing a million wafers a month would need rows of them running day and night. Picture hundreds of school buses, each worth a fortune, lined up in a sealed and spotless hall. Suddenly, $119 billion stops sounding like science fiction. It starts sounding like a budget. Yet for all its size, the most important part of Terraab may not be the halls or the clean rooms or the transit lanes. It may be the light. Every advanced chip begins as a pattern of light. The process is surprisingly similar to old-fashioned photography. Light shines through a stencil called a mask and projects its pattern onto a silicon wafer coated with a light sensitive material. Layer after layer, those projected patterns become billions of microscopic switches. The smaller the pattern, the more switches fit on a chip and the more powerful that chip becomes. But there is a catch. You cannot draw a line thinner than the light you are drawing with. To print smaller, you need light with a shorter wavelength. Ordinary visible light is far too coarse. So the most advanced chips today are made with something called extreme ultraviolet lithography. usually shortened to EUV. It uses light with a wavelength of about 13 and a half nanometers. Light so short that it is absorbed by almost everything, including air itself. And producing it is one of the most extraordinary feats in modern engineering. Inside an EUV machine, tiny droplets of molten tin fall through a chamber. A powerful laser strikes them about 50,000 times every second. Each impact turns a droplet into a flash of glowing plasma. And that plasma releases a burst of extreme ultraviolet light. That light is then guided by a series of mirrors. Not ordinary mirrors. These are among the smoothest surfaces humans have ever made. If one of them were stretched to the size of the earth, its tallest bump would still be shorter than a person. It is breathtaking and it is fragile. Every explosion of tin throws off debris. That debris threatens to coat or scratch the very mirrors the system depends on. When a mirror degrades, the machine has to stop for maintenance. The whole process draws enormous amounts of electricity. And each light source is complex, temperamental, and expensive to keep running. For an ordinary factory, that is a cost of doing business. For a factory chasing 1 million wafers a month, every hour a machine sits idle is a hole in the plan. Now multiply that problem across every machine on the floor. Because in a conventional chip plant, every EUV machine carries its own light source, its own tin, its own laser, its own small storm of plasma. Which brings us back to the circle and to those three letters Musk typed. What if the light did not have to be made inside every machine at all? A free electron laser begins with something very small, moving very fast. Electrons are pushed to nearly the speed of light. Then they are sent racing through a long corridor of magnets arranged with alternating poles. Engineers call this corridor an undulator. As the electrons pass through, the magnets push them left, then right, then left again, forcing them into a tight zigzag. And every time an electron is forced to change direction, it gives off light. Thousands of tiny swerves, thousands of tiny flashes. Together, they build into a single intense beam. Here is the elegant part. By adjusting the energy of the electrons and the spacing of the magnets, the wavelength of that beam can be tuned with great precision. Tuned in principle to exactly the extreme ultraviolet light that chipm needs. No molten tin, no exploding droplets, no cloud of debris drifting toward those priceless mirrors. In theory, a powerful free electron laser could produce clean EUV light steadily from a source that never has to vaporize anything. That alone would be remarkable. But it is not the boldest part of the idea. The boldest part is where the light comes from. Think about how electricity reached homes a little over a century ago. Before power grids, a factory that wanted electricity often had to run its own generator. Every building its own noisy, costly machine in the basement. Then came the central power station. One large plant feeding many buildings through a network of wires. Cheaper, more efficient, and far easier to maintain. Now apply that logic to light. In a conventional fab, every EUV machine makes its own light. 50 machines means 50 separate light sources, each with its own lasers, its own tin, its own maintenance schedule. The circle at the center of Terrafab suggests a different approach. One powerful free electron laser in the middle of the complex, producing a vast supply of EUV light and sending it outward through beam lines to dozens of machines at once. One source, many printers. Suddenly, the layout of the entire building makes sense. The circle is not decoration. It is not architecture for the sake of architecture. It could be the power station of light with the rest of the factory arranged around it like a city around its river. If it works, it would not simply be a better chip factory. It would be a different kind of chip factory, one that no one else on Earth has built. But it is important to be clear about what we know and what we do not. What we have is a circle on a plan and a short reply from Musk. Building a free electron laser powerful and reliable enough to feed an entire production floor would be a staggering technical challenge with enormous risk. And even if the light works perfectly, Terraab faces another problem. One that has nothing to do with lasers at all. A chip factory can survive a great many things. A delayed shipment, a broken part, a difficult quarter. What it cannot survive easily is darkness. If the power drops even for a few seconds, the wafers inside the machines can be ruined in the middle of production. Equipment can be damaged. And because a single wafer can spend weeks or months moving through hundreds of steps, the loss is not just a few minutes of work. It is everything that came before. Imagine a potter shaping a delicate vase on a wheel. The clay is almost finished, the walls impossibly thin. Then for one second, the wheel stops. The vase collapses. Now imagine that vase took three months to make. That is what a brief outage can mean inside a fab. Now scale that up to 100 million square ft. The clean rooms alone consume enormous amounts of energy just to keep the air pure. The lithography machines draw heavy power every time they fire. Then add cooling systems, transport vehicles, lighting, and everything else a city-sized building needs. A facility like this could consume as much electricity as a medium-sized city. And it needs that electricity every second without a single interruption. The power industry has a name for this kind of supply. Firm power. Electricity that arrives steadily no matter the weather, the season, or the time of day. So, where will Terraab get it? You might expect the answer to be solar. After all, Tesla sells solar roofs, power wall home batteries, and mega pack systems for entire utilities. But the answer reportedly being prepared by SpaceX is something else. According to reports, the plan is to keep Terraab off the Texas public grid altogether. Instead, natural gas power plants would be built right beside the factory, paired with large battery storage systems. The factory would in effect produce and control its own electricity. Why gas from a company so closely tied to solar? The answer is one word, reliability. Solar panels are brilliant while the sun is up and produce nothing at night. To run a factory this size on solar alone, you would need a battery system large enough to keep a small city running for around 12 hours every single night. The cost would be staggering. There is also the matter of waiting in line. Across the United States, technology companies are racing to build AI data centers, and each one can consume as much power as a small town. The grid is already stretched thin. A new facility asking for a massive connection could wait years. With its own power plants, Terraab would not need to wait for anyone's permission. The batteries play a different role. A facto's demand is never perfectly flat. When machines start up or shut down, power use spikes and dips. The batteries absorb those swings and smooth the flow. And if a gas turbine falters, the batteries can step in instantly, bridging the gap until the problem is fixed. So look again at what Terrafab really is. A chip factory, a power plant, a battery farm, a private utility, all folded into one site. The race for AI has quietly changed shape. It is no longer only about who can buy the most processors. Processors need chips. Chips need factories. And factories need a river of electricity that never stops flowing with its own power supply and perhaps one day its own source of light. Terraab begins to look almost self-sufficient. Almost because there is one ingredient it cannot generate on site. cannot order from a supplier and cannot buy at any price. Tesla knows how to stamp, steel, and assemble vehicles at astonishing speed. SpaceX knows how to design rockets and fix problems in real time. XAI knows how to train models. None of that is chipm. Advanced chip manufacturing happens at a scale that brushes against individual atoms. The air inside a FAB must be cleaner than a hospital operating room. The chemicals must be so pure that a single trace of contamination can change how a circuit behaves. The printing process is so sensitive that the vibration of a truck passing on a nearby road can ruin a chip. Some fabs sit on massive shock absorbers simply to calm the faintest tremor rising from the ground. And behind all of this sits one number, the number that decides whether a factory thrives or quietly bleeds money. It is called yield. Yield is the share of chips that come off the line actually working. When a new fab starts up, its yield is usually poor. Many chips fail and are thrown away. At 50%, half of the precious silicon and half of the machine time are simply lost. To make real financial sense, a factory needs a yield above 90%. Climbing from one to the other can take years. Think of a chef trying to perfect an extraordinarily complicated recipe. Except this recipe has thousands of ingredients. Everyone must be measured perfectly at the same moment every time. And every failed attempt costs a fortune. The biggest chip makers spent decades learning how to raise their yields. And that knowledge does not live in textbooks. It lives in people. engineers who spent years standing beside the machines, watching what happens when a temperature drifts by a fraction of a degree and learning one problem at a time how to fix it. Inside the industry, raising yield is often described as a dark art. Money cannot compress decades of that experience into a few months, which is why one detail in the Terrafab story may matter as much as the laser or the power plants. According to reports, a former Intel leader with 17 years at the company is in line to run the project. This is someone who oversaw Intel's 18A work, one of the most advanced chip manufacturing efforts on the planet. Intel has been making chips for more than 50 years. It has lived through yield collapses, equipment failures, and problems that took months to untangle. That kind of hard one experience is not something a consultant can simply explain in a meeting. It is closer to muscle memory. Now the pieces begin to fit together. Tesla brings the ability to build at enormous scale. SpaceX brings the habit of solving hard problems quickly. XAI brings a hunger for chips so deep that the factory would never be short of a customer. And the veterans from Intel bring the one thing none of these companies has ever had. real experience taking an advanced chip process and making it work in production. If that experience transfers, Terrafab might avoid at least some of the long, painful learning curve that nearly every new fab has to climb. If it does not, the most spectacular building in the world could spend years producing very expensive silicon that does not work. Step back and look at the whole picture. A demand for chips so vast that by Musk's own estimate, the world currently makes only about 2% of what his companies will need. A building of 100 million square ft designed to grow section by section toward 1 million wafers a month. A circle at its center and a threeletter reply that hints at a free electron laser. A single source of light feeding an entire floor of machines. power plants and batteries built beside the factory so that it never has to wait for the grid and never has to blink and a leader from Intel carrying 17 years of the kind of knowledge that cannot be bought. Each piece on its own would be ambitious. Together they describe something that has never existed. A single place that would control nearly the entire chain from the electricity in the wires to the light in the machines to the finished chip inside a self-driving car. a humanoid robot or an AI data center. Today's chip industry is spread across companies, countries, and continents. Terraab is a bet on the opposite idea. That one enormous, tightly integrated facility could move faster than the scattered system the world relies on. Now, if it works, its impact could reach far beyond Tesla or SpaceX. It could change how the most advanced chips are made and where. If it fails, it will become one of the most expensive lessons in industrial history. Either way, the question it raises will not go away. In a future built on artificial intelligence, robots, and autonomous machines, the limit may not be ideas or software, or even money. The limit may be how fast we can turn sand and light into thought. $119 billion says Elon Musk believes that limit can be moved. So, what do you think? Is Terrafab the blueprint for how the future gets built or a gamble too large even for him? And which part surprised you most? The scale, the circle, or the power plant hidden beside it? Tell us in the comments. And if you want to keep exploring the machines and mega projects shaping the world of tomorrow, subscribe to Future Atlas 2050. The future is being drawn right now. We will keep mapping.