Why Tesla & SpaceX Want Their Own AI Chip Factory
Description
Tesla and SpaceX want to build Terafab, a massive semiconductor factory designed to produce chips for artificial intelligence, cars, robots and even future data centers in space.
But why build a factory this big? And more importantly, how realistic is a project like this?
In this video, we start with an old Z80 processor and end up exploring one of the most ambitious industrial challenges in recent years.
We'll take a look at what Terafab is supposed to become, why Tesla and SpaceX want to make their own chips, what challenges they're facing and what computers in space have to do with any of it.
We'll also talk about EUV lithography, semiconductor manufacturing, satellite cooling and the first milestones announced for the project.
Because imagining a gigantic chip factory is one thing. Actually getting it to produce millions of working chips? Well, that's a whole different story.
Terafab is still a project in development. The factory's proposed size, investments and timelines dis
Transcript
Read auto-generated transcript (2911 words)
Kind: captions Language: en Now, take a look at this little thing right here. It's an old processor. And would you believe Elon Musk wants to build a cheap factory 10 times bigger than Tesla's Gigafactory to make processor for cars, robots, and even computers up in space. Yep, you heard it right. Computers in space. And I'm not talking about ordinary satellites that collect information and send it back down to Earth. The idea is to have actual artificial intelligence systems working up there powered by enormous solar panels. Now, I don't know about you all, but ideas like that absolutely fascinate me because just look at what I'm holding. Sure, it's old, but even back then, making a processor like this took factories, precision machinery, and a whole lot of complicated steps, and today's processors. There are whole different level of complexity. We already need chips for our computers, our cars, our robots, and now we're talking about filling space with computers, too. So I figure somebody must have started thinking well if we are going to need that many chips why not just build our own factory a really big factory it's called Terraab and it's a project involving Elon Musk Tesla and SpaceX and no they're not planning to build the factory in space thankfully that part staying right here on Earth but I'll tell you What really caught my attention when I went through the project's presentation, I came across a picture. One picture that made me stop and look twice because hearing somebody talk about a gigantic factory is one thing. Actually, seeing just how gigantic that factory is supposed to be. Well, that's something else entirely. On one side, we've got Tesla Gigafactory in Texas. And let's be fair, that ain't exactly a little shed out back. On the other side, we've got Terapab 10 times bigger. They're planning a complex covering roughly 100 million square ft. That's over 9 million square meters. Now, keep in mind, those are the announced dimensions of a planned industrial complex. We are not talking about a factory. that's already been built. When I saw a number like that, I had one pretty simple question. What in the world are they planning to put in there? Because making a processor isn't as simple as grabbing a piece of silicon, printing a couple of circuits on it, and sticking it in a box. You start with silicon wafers. Then you build layer after layer of circuits using incredibly precise manufacturing processes. After that come inspections, cutting the chip apart, making the connection, packaging, and more testing. And for the most advanced semiconductors, those steps are often spread across different factories and different companies. Terrafab wants to bring much of that process together in one enormous complex. Logic chips, memory, advanced packaging and testing all close together. All designed to produce enormous quantities. But here the thing, why go to all the trouble? Well, according to Musk, Tesla and SpaceX are going to need so many chips that the manufacturing capacity available today might not be enough. They'll still be buying chips from other manufacturers. That isn't going to change overnight. But they also want manufacturing capacity of their own. Well, that's a little like running out of bread at home and deciding to open yourself a bakery. except this bakery would have to make some of the most sophisticated chips in the world. And there's Intel to consider too. Back in April 2026, Intel announced its involvement in the project. The details of that collaboration haven't all been worked out yet, but it sure gives you an idea of just how complicated this challenge really is. But right about here, I started wondering something else. all the manufacturing capacity. What kind of chips are they actually planning to make? And that's where this story starts heading in two very different directions. In the Teraphar presentation, two names come up. AI5 and AI 6. AI5 is designed for Tesla's driving systems and Optimus robots. AI6 is presented mainly in connection with robotics. Now to understand why they need specialized processor, let's look at something really simple. Picking up a cup, we reach out and grab it. And if it's heavier than we expected, we adjust our grip without even thinking about it. But a robot, well, he's got to recognize the object, figure out where it is, calculate the movement, decide how much force to use, and keep making adjustments. along the way, all in a very short amount of time. Because if that robot has to wait for some fairway server to send back an answer every single time, that cup might just end up on the floor. So, it needs plenty of computing power right here on board. And so far, I can see the reasoning. Cars, robots, and chips designed specifically to make them work. But then another name shows up D3. Now this is the chip the project is presenting for computing in space. And we can't just say well let's take a powerful processor and put it in on a satellite because up there the rules are different. We've got radiation, power limitations, thermal conditions and there's reliability. If something breaks, we can't exactly open up the case and swap up the bad part. Now, we still don't know enough about the tree to say exactly how it's going to handle all those problems. And believe me, it's even more ambitious than the factory itself. When we ask an artificial intelligence system a question, all we see is a little window and an answer. It all seems so simple, almost like there's nothing physical behind it. But somewhere that answer has to be calculated by actual hardware. And a whole lot of the time that means buildings packed with servers, processors, memory network connections, power supplies and cooling systems, real physical equipment. And I'll tell you, all that equipment use a whole lot of electricity. So SpaceX comes along with an idea. What if we moved some of that computing work up into space? And the first thing I wondered was why in the world would we make things that complicated? Well, the answers starts with the sun. Outside Earth's atmosphere, sunlight reaches you without being weakened by the air. And in certain orbits, solar panels can receive that sunlight for most of the time. Now, picture satellites carrying huge solar panels, processors, and memory. Some satellites already use artificial intelligence to process information right there on board. But this idea, it's a good bit more ambitious. The goal is to move some of the work we currently do in massive data centers here on Earth up into space and maybe even connect a whole network of satellites so they can work together. That's quite something when you think about it, isn't it? Down here we are discussing where to build the next data center and how much electricity is going to need. and somebody looks up at the sky and thinks, "Well, why don't we just put it up there?" Now, of course, there's a mighty big difference between imagining something like that and actually making it happen. SpaceX is counting of Starship to bring down the cost of sending large amount of equipment into orbit. But cheaper launches aren't enough. You've still got to build those satellites, maintain them, and make sure they can keep working reliably. And then those computers have to talk to each other. Here on Earth, large AI systems can exchange enormous amounts of data between servers using incredibly fast connections. But up in space, satellites are far apart and constantly moving. And for certain tasks like training large AI models, that communication could become a serious problem. But there's another issue. And when I figure out what it was, I couldn't help but smile. Because at first glance, it looks like we've already solved it. When a computer's working, it produces heat. We all know that. Here on Earth, we can use heat sinks, fans, and liquid cooling systems to carry that heat away. A fan, for example, moves air around, and that air picks up some of the heat and carries it somewhere else. Works just fine. Now, let's send that computer up into space. Most of us think of space as being freezing cold. So, problem solved. Nope. Because out in space, there isn't any air. That fan can spin all it once, but in a vacuum, it's got nothing to move. And meanwhile, that processor just keeps getting hotter. Somehow, we've got to get rid of all that heat. So, how do we do it? Through surfaces that release heat by thermal radiation. Basically, radiators designed to work in space. But the more computing power we use, the more heat we're going to get rid of. And those radiators have mass. They take up space. And we got to carry every bit of that equipment into orbit. Now, this is the kind of paradox I find fascinating. We're going into space partly because we want to take advantage of the sun's energy. But the more of that energy we use for computing, the more heat we have to figure out how to get rid of. One solution that creates a whole new problem. And that's not the only one. Back in April 2026, the US Government Accountability Office, the agency that oversees how the federal government spends public money, published a report on space-based data centers. Among the challenges it identified were cooling, power supply, communications, and costs, not mentioning radiation and managing satellites in orbit. So, we're not talking about something we've already figured out how to build on this scale. We're talking about a system that still has to prove it can actually work. But if we're ever going to build infrastructure this big, we're going to need an enormous number of chips. And wouldn't you know it, that bring us right back to our gigantic factory. Because manufacturing all those processor is a challenge that starts long before we ever send them into space. So let's head back down to Earth. Now take another look at this little thing right here. We can hold it between two fingers. But making today's most advanced processor takes an industry that's incredibly complicated. We're talking hundreds of manufacturing steps tightly controlled environments and machines so sophisticated they took years to develop. Take EUV lithography for example. It uses extreme ultraviolet light to create incredibly tiny structures on silicon and the industrial machines that made this possible. They're produced by just one company. ASML based in the Netherlands, one single company. But here's where things get tricky. A factory doesn't start producing advanced chips just because you've bought the equipment. and flip the power switch. You've got to install everything, get all those machines working together, and fine-tune the manufacturing processes. And then there's one very important word, yelled. Because making one working chip is an achievement, but making millions of them with enough good chips to make the whole operation financially sustainable, well, that's a whole different story. If too many chips on a wafer turn out effective, manufacturing becomes too expensive. And right there is where a gigantic promise runs into its first very real challenge. In April 2027, Tesla started construction on a semiconductors research facility in Austin near the Gigafactory. But that's not the big terap just yet. For that larger project, SpaceX announced the site in Grimes County around the Gibbles Creek area also in Texas. And for the first phase, the planned investment was estimated at around $16.8 billion. Now, that's an enormous amount of money. And yet, building an industrial operation on this scale could take a whole lot more than that. All of it just to manufacture little objects we can hold between two fingers. After looking through the presentations, the dimensions, and all those plans, I finally came back to one pretty simple question. When are we actually going to see the first working chip from the terra project? On October 20th, 2026, Musk gave us his first target date. Sometime between May and June 2027, they're hoping to produce their first working experimental chip at the Austin Research Facility. Most likely an SRAM based test circuit. All that work just for a test chip. Yep. And I'll tell you that would already be a pretty important achievement because the goal at this stage isn't to have a processor for space-based data center rolling off the production line. It's to prove that the manufacturing process can actually create working circuits that the machines, the materials and all those complicated steps can finally start working together. But now here's the thing. Making your first working chip doesn't mean you can suddenly make millions of them with a good manufacturing yield. For large scale production, the announced target is the third or fourth quarter of 2028. And Musk himself has acknowledged that these deadlines are extremely ambitious and could sleep. So, I wouldn't go marking those dates on the calendar as guarantees because even if they manage to meet every single target, that still wouldn't mean we'd immediately have massive data centers running up in orbit. There would still be launches, power supply, cooling, communication, and the economics of the entire system to figure out. But at least now we've got something we can actually keep track of. a research facility, an announced location for the big factory, planned investments, and the first deadline we can check. Now, that doesn't prove Terafab is going to succeed, but it does mean this incredibly ambitious project is starting to have real milestones we can measure. And the very first one, that chip, will they actually manage to make it? You know what really gets me about this whole story? We started out with an oil processor that fits right in the palm of my hand. And somehow we ended up talking about an enormous factory. Robots that can move around on their own. Satellites and computers up in space. But here's what I find most interesting. We still don't know how any of these is going to turn out. We don't know if Terrafab will manage to produce those chips on schedule or whether huge data centers in orbit will ever become a practical affordable solution. There might be problems nobody's even thought of yet. Or maybe somebody will come up with solutions we can even begin to imagine today. But you've got to start somewhere. And for Terafab, that first big milestone is going to be making a chip, an experimental chip that actually works. And I, for one, I'm mighty curious to see if they can pull it off. I'm aid mom and I'll see you all soon at the next