Elon Musk Announces Terafab Chip Factory for Austin

Kind: captions Language: en Welcome to the AI for Career Success podcast by Kurt Robbins. We're experts at AI coaching and helping professionals like you and your employees upskill in AI. I'm your host, Fred Jones, and uh this is episode 206. Joining me today is our resident expert, Daffhne Blake. >> I am uh really excited to get into this one today. It's a massive topic. >> It really is. Like if you're an AI professional listening right now, you're probably used to, you know, stressing out over standard cloud computing costs or maybe GPU availability for your employees. >> Oh, yeah. The usual daily bottlenecks, >> right? But today, we're going to make those terrestrial concerns look completely microscopic. We're unpacking a speech Elon Musk delivered recently on March 21st, >> the one he gave at the uh the defunct Seaholome power plant in Austin, Texas. >> Exactly. And our mission today is to unpack the official launch of the Terapab project. Musk literally called it the most epic chip building exercise in history, >> which is quite a claim to make. But we're going to break down how this project connects AI, advanced robotics, and like actual heavy lift space exploration. >> Because this isn't just about building faster computers so your applications run a little quicker. This is about fundamentally altering human civilization. It's about pushing humanity up the Cardartesev scale, >> right? The Cartesev scale. So, uh, for those who might not be familiar with that term, it's a concept proposed by a Russian astrophysicist back in the 1960s. >> It's a way of reading civilizations based entirely on their energy consumption. Right. >> Exactly. And well, Earth is barely even a type one civilization. >> Barely. I'd say we're not even close. >> No, not at all. A type one civilization harnesses all the energy that reaches its home planet from its parent star. But you have to look at the raw data of our solar system. The sun contains 99.8% of all the mass in our solar neighborhood. >> Wow. So we are practically nothing. >> Yeah. Earth is basically just this tiny dust moat in a vast darkness. Because of our size and distance, Earth only receives about half a billionth of the sun's total energy. >> Just half a billionth, >> right? And if you look at our entire global electricity production, it's only about a trillionth of the sun's energy. >> Okay, let me unpack this because those numbers are just enormous. Trying to power advanced AI exclusively on Earth is well, it's like trying to fill a swimming pool by catching raindrops when there's a fire hose right next door. >> That is a perfect analogy, Fred. >> But wait, I have to stop you there. We have the technology for nuclear fusion developing right here on Earth. Why go to the massive expense of rocketing servers into a vacuum when we can just, you know, build massive fusion plants in the desert? It's a completely logical assumption, but terrestrial power is fundamentally constrained by physics and geography >> and thermodynamics, I'm guessing. >> Yeah, thermodynamics, too. Even if you perfect fusion on Earth, you are still operating on that tiny dust mode. To truly scale civilizational power, you have to physically go to space. >> So, it's mathematically necessary if we want sensient level AI across billions of robots. >> Exactly. If you increase our power output on Earth by a million times, you are still only harnessing a millionth of the sun's total energy. Plus, generating that much power down here creates a massive heat problem that our atmosphere just can't dissipate without cooking us. >> Which means to capture that fire hose of energy for AI. We need entirely new hardware. We need a scale of manufacturing that makes the current tech industry look like, I don't know, a cottage business. >> Which brings us directly to the compute deficit. The here scale of the terapab goal is a terowatt of compute per year. >> A terowatt? That's staggering. For context, what is the current global output of Earth's AI compute? >> Everything powering the models you use for work today, it's only about 20 gawatt. >> Wait, only 20 gawatt? >> Yeah. So if you took every single fabrication plant on Earth right now, Samsung, TSMC, Micron, all of them, and combined their outputs, it would only supply 2% of what this Terapab project needs. >> We are essentially 98% short of the hardware required for the next leap in AI. That is insane. >> Precisely. And that's why the immediate terrestrial solution is being built as we speak. It's the new advanced technology fab right there at Giga Texas in Austin. It's an undertaking so massive that Governor Abbott was actually in the audience showing state support for it. >> Yep. It's a huge deal. >> But Deafany, what makes this Austin facility so unique? I know Musk loves vertical integration, but how are they setting this up differently than say a traditional TSMC plant? >> The real innovation is in its physical layout. They are housing the equipment for logic, memory, packaging, testing, and lithography masks all inside a single building. Okay, for those of us who aren't hardware engineers, a lithography mask is basically like a giant, incredibly complex stencil used to print the microscopic circuits onto silicon. Right? >> That is a brilliant way to describe it. Normally, chip development is super fragmented. You design that complex stencil in one location, send it out to a specialized manufacturer, wait for it to be made, >> then test the resulting chip, find a flaw, and weigh some more. >> Exactly. It takes months. By putting everything under one roof, Austin is creating what Musk calls a fast recursive loop. >> So instead of sending a rough draft of a document across the world for edits and waiting weeks, you're basically sitting in the exact same room with the editor just rewriting the laws of physics in real time. >> Yeah. You design the stencil, print the chip, test it, find the flaw, tweak the code, and print a new stencil almost immediately. >> Let's take a break for a special message from our producer, Kurt Robbins. >> Hi, this is Kurt Robbins. First, thanks for listening. I truly appreciate your support. I want to let you know that I'm currently accepting new clients. My rates are affordable and I have more than 25 years of experience working for enterprise companies like Microsoft, Northrup, Grumman, Oracle, PNC Bank, FedEx, USA, and Wells Fargo among many others. If you want to improve your IT documentation and communications, hire me. I deliver fast, know how to use AI to improve efficiency and accuracy, and love going the extra mile to satisfy my clients. Thank you for subscribing and listening. Back to you, Daffhne and Fred. >> Welcome back to the AI for Career Success podcast, where we help you get smarter than your competition by coaching you in AI. >> So, before the break, Daphne, we mapped out this massive earthbound fab in Texas. But the ultimate goal of Terapab isn't to keep these chips on the planet. No, not at all. And achieving that off-world goal means manufacturing two very different categories of chips. The first type is for edge inference. >> That's the hardware optimized to run locally, right? Like primarily for Tesla vehicles and the Optimus humanoid robots. >> Exactly. And the volume required here is just staggering. >> Because Musk pointed out that while the world produces about 100 million cars annually, he expects humanoid robot production to be somewhere between 1 and 10 billion units a year. >> Yeah. The sheer physical footprint of 10 billion robots completely dwarfs the entire legacy auto industry. >> It really does. But the second type of chip is where the space strategy actually comes into play. >> Right? These are high power chips designed specifically to operate in orbit because space is an incredibly hostile environment for delicate electronics. >> I think people imagine space is just, you know, empty and cold, but the radiation is intense, >> very intense. You have high energy ions, photons, and dangerous electron buildup. >> Electron buildup means the radiation in space is literally clinging to the chip, which can cause an electrical arc and short the whole thing out, doesn't it? >> Exactly. You have to design the architecture to survive that bombardment. But counterintuitively, you also want to design these space chips to run hotter than terrestrial chips. >> Wait, run hotter? Why? >> To minimize the mass of the radiators. >> Okay, that sounds backwards until you remember how vacuums work. Down here, air carries heat away from your laptop. But in space, there is no air, >> right? The only way to cool server farm in orbit is to radiate that heat outward as infrared light. By letting the chips run at higher temperatures, the temperature difference between the chip and the freezing vacuum of space is greater, >> which makes the radiator much more efficient. So, it allows you to build smaller, lighter cooling systems, >> which is hyper critical when every single pound you launch costs money. SpaceX already has thousands of Starlink satellites in orbit, so they have immense practical data on heat rejection in a vacuum. >> But the economics of putting these AI chips in space are probably the most surprising part of this whole vision. Musk estimates that deploying AI hardware in space will actually drop below the cost of terrestrial AI in just 2 to 3 years. That's the claim that really made me pause. >> I know. If I'm an AI developer, how on earth does launching server hardware into orbit become cheaper than just plugging a rack into the wall at a data center in Virginia? >> Because space naturally solves the hardest bottlenecks of generating power. Up in orbit, it is always sunny. You don't have to deal with atmospheric attenuation. >> Meaning there is no air, clouds, or pollution absorbing and scattering the sunlight before it hits your panels. >> Exactly. Plus, there are no extreme weather events. No hail, no hurricanes, >> which means your space solar panels actually cost less to manufacture because they don't need heavy glass framing or structural steel to protect them from the elements, >> right? They can be incredibly thin and lightweight. >> And let's not forget the NIMBY factor. You know, not in my backyard. >> As you try to scale gigawatts of power on Earth to train bigger AI models, you run out of easy real estate and political goodwill. People don't want massive power plants or noisy data centers built next to their neighborhoods. >> The regulatory friction on Earth is immense. Increasing power down here just gets harder and more expensive over time. In space, you have compounding economies of scale. >> So once the transport infrastructure is in place, expanding a solar array in orbit actually becomes cheaper and easier over time. >> Yep. But to make this economy of scale work, SpaceX really has to carry the load. You have to get millions of tons of hardware off the planet. >> To achieve this terowatt of compute, SpaceX needs to launch 10 million tons to orbit per year, assuming they can generate about 100 kow per ton of payload. That's why the massive Starship rockets V3 and eventually the much larger V4 are the lynch pin of this entire operation. >> During the presentation, they actually showed a visual of a 5'11 Optimus robot standing next to the Starship V3 just to give a sense of the scale. >> It's absurd. You have this humansized, incredibly advanced robot standing next to a rocket the size of a skyscraper. It really highlights that this is purely a massive logistical numbers game, but one that is physically possible with fully reusable rockets. >> And as SpaceX fairies those millions of tons to orbit and we actually hit that terowatt milestone, it opens up an even bigger question, >> right? Because physics tells us that eventually even Earth's immediate orbit gets crowded. Does the ambition just stop there? What happens after Terra? >> No, you don't stop. The next phase is getting to a pedawatt of compute. And you achieve that by building an electromagnetic mass driver on the moon. >> Okay, I know we're talking about real engineering here, but that sounds straight out of a sci-fi novel. Daphne, explain how an electromagnetic mass driver actually works. >> Think of it as a magnetic whale gun that is miles long. Instead of using explosive rocket fuel, it uses sequenced electromagnetic pulses to accelerate cargo along a track >> until it reaches escape velocity. Just flinging materials directly into deep space. >> Exactly. And you can only build that on the moon because of the environment >> because the moon has no atmosphere to create air resistance and only 16th of Earth's gravity. >> Yes. So using robotic labor, literal armies of those Optimus robots along with human engineers, you build this rail gun on the lunar surface. >> It dramatically drops the cost of harnessing power and sending pedawatts of compute into the solar system because you eliminate the need for chemical rockets entirely for that leg of the journey. >> It really feels like we are talking about unlocking an age of staggering abundance. You are building an economy that is a million times larger than Earth's current economy. It ushers in an ultimate post scarcity vision. It's very much like the culture universe in Ian Bang science fiction novels. >> Oh, totally. If you have an AI and robotics economy that is a million times our current size, money effectively becomes obsolete. >> Because literally any material need you can possibly think of can be met almost instantly by cheap energy and robotic labor. It's abundance for everyone. Taking a trip to Saturn wouldn't be some exclusive luxury reserved for billionaires. >> No, it would be accessible to anyone who wants it, just like booking a commercial flight today. The limitations we accept today as permanent facts of life simply vanish. >> And that leaves us with a fascinating question to chew on. If you're an AI professional grinding away at your career today, imagine a future where AI and robotics create an economy a million times our current size. An economy where literally everything, even a trip to Saturn, is free and abundant. What happens to human ambition when your survival and comfort are mathematically guaranteed? What will you choose to do with your time? >> Thank you for listening to the AI for Career Success podcast from Kurt Robbins, where we help you get smarter than your competition. Please buy us a coffee, and remember that we're experts at coaching you and your employees about AI. We'll see you next