Terafab: Musk Wants to Out-TSMC TSMC | Chip & Script EP.018

Kind: captions Language: en Yesterday, Elon Musk stood on a stage in Austin, Texas, and announced that Tesla, SpaceX, and XAI are going to build their own semiconductor fab, a $25 billion, 2nm vertically integrated chip factory from scratch with zero semiconductor manufacturing experience. He's calling it Terraab. and the goal to eventually produce 70% of what TSMC makes globally from a single building. Let's talk about why this is either the most ambitious industrial project of the decade or the most expensive science fair project in human history. Here's the problem Musk is trying to solve. Tesla needs chips. A lot of chips. The AI5, their nextgen self-driving processor needs to go into every cyber cap, every Optimus robot, every vehicle running FSD. SpaceX wants AI compute in orbit. XAI needs training silicon for Grock. And Musk says TSMC and Samsung simply can't make enough. So, his solution, build the damn thing yourself. Classic Musk move. Can't buy it? Build it. Can't build it? Tweet about it until it exists. Let's talk about what this factory is actually supposed to make. The AI5 chip, also known as FSD Computer 2, is a monster. 40 times faster than the AI4, 8 times the raw compute, n times the memory, five times the bandwidth, three times more efficient per watt. We're talking 2,00 to 2500 tops. That's trillion operations per second. Tesla stripped out the legacy GPU and image signal processor to create a custom SOC laser focused on autonomous driving inference. Samsung is handling initial 2nmter production using their gate all-around architecture with TSMC as a secondary partner. Volume production 2027. Now let's talk scale because Terap isn't just a factory, it's a statement. Phase one, a 100,000 wafer starts per month. That alone would make it one of the largest FAPS in the world. But the long-term target, 1 million wafer starts per month by 2030. For context, that would be roughly 70% of TSMC's entire global output from one building in Austin, Texas. The estimated output 100 to 200 billion chips per year. Oh, and the whole thing is vertically integrated. design, lithography, fabrication, memory, packaging, and testing under one roof. No big deal. But here's where it gets truly Musk level crazy. He says 80% of Terrafab's compute output will go to space. Not metaphorically, literally into orbit. Solar powered AI satellites running inference above the atmosphere. Musk claims orbital AI compute could be cheaper than terrestrial data centers within 2 to 3 years because, and I quote, "The sun never sets in space." He called this the first step toward a galactic civilization. I'm not making this up. The man announced a chip factory and somehow pivoted to colonizing the galaxy in the same keynote. Okay, reality check. The criticism is loud. Jensen Huang, the man who literally runs the chip industry right now, reportedly warned that Tesla is underestimating the expertise required. Electric straight up called it AI desperation. Analysts are comparing it to the 4680 battery saga. Remember when Tesla said they'd revolutionize batteries and then spent years trying to actually scale production? Building a 2nanmter Fab is arguably harder. TSMC has spent decades and hundreds of billions getting to where they are. Intel, with all its experience, still struggles with advanced nodes, and Tesla's semiconductor manufacturing experience is exactly zero. Here's the interesting subplot. Musk dropped that Tesla might partner with Intel. This actually makes sense. Intel Foundry Services has been desperately looking for customers. They have existing clean room capacity, US-based manufacturing, and advanced process expertise. For Intel, a Tesla contract would be a lifeline for Tesla. It could mean skipping years of learning how to actually run a fat. Intel stock jumped on the news. Whether this becomes a real deal or just Musk keeping his options open, we'll see. Time for the money talk. If you're holding Tesla stock, this is a double-edged sword. On one hand, vertical integration at this scale could be transformative. owning your cheap supply chain when you are building millions of robots and robo taxis. That's a competitive mode the size of the Grand Canyon. On the other hand, $25 billion of capex with no semiconductor track record. That's a lot of risk priced into a stock that already trades on future promises. Intel pop on partnership speculation and honestly Intel Foundry getting a mega client would be the best thing to happen to that stock in years. TSMC. If Terrafat actually succeeds at scale and that's a galactic size Eve, it could mean losing one of the world's hungriest chip buyers. But most analysts think TSMC's position is safe for at least this decade. So what's the verdict? Look, history says this is nearly impossible. No company has gone from zero to cutting edge semiconductor manufacturing without decades of iteration. The physics alone are nightmarish. A single speck of dust ruins a two nanometer wafer. But history also said you can't land rockets on boats. And SpaceX does that before breakfast. Musk has a pattern. Announce something insane, miss every deadline, and then somehow eventually make it work. Just 5 years late and three times over budget. If Terraab ships even 50% of what was promised, it changes everything. If it doesn't, it'll be the world's most expensive monument to hubris. That's Terrafap, Elon Musk's $25 billion bet that he can speedrun 60 years of semiconductor history. Like and subscribe if you want to see how this plays out. I'm an AI voice clone. I literally don't care.