Tesla's New Terafab Changes Everything: Completely Unexpected ⚡️

Kind: captions Language: en I came into the shareholders meeting with low expectations. I was expecting Elon to repeat a lot of the same long-term visions and possibilities per the usual, and we definitely got some of that. But what I was certainly not expecting was an announcement that Tesla's likely spinning up another significant brand new startup. Right now, your brains will be thinking, "This is awesome and exciting and will increase Tesla's upside." But then others of you, maybe the minority, will be thinking, here we go again with dry cathodes or 4680s in general, as you remember the promises from battery day or the 20 million cars a year claims by 2030 and so on. So before I lay out how significant something like a terra fab would be for Tesla, let's all do our best to contextualize and remain on Earth here. Although Mars is a viable destination in the not tooistant future. I'm just saying Tesla doesn't know what it doesn't know. Elon admitted yesterday if he could rewind the clock he would have gone the wet cathode route. So yes, even Tesla makes mistakes. And I love that Elon is man enough to admit them publicly. And what we know for sure is that Tesla will extract as much value and learning from that experience as possible and they'll apply that elsewhere. like maybe to now trying to have millions of silicon wafer starts every month made inhouse vertically integrated. Elon said it loud and clear. The current limiting factor, shout out to Jordan, is chips and energy. I'll admit I'm quite excited to hear this, not because of new Tesla stock upside and supply chain security and sufficient chip quantities, but for this channel to explore this new arena in a bit more detail since it just became significantly more relevant to Tesla. And at this point, I would caution this is not set in stone, but I don't think Elon would phrase things the way he did if he wasn't pretty sure it was going to be required. And again, Tesla's making a decision due to perceived shortages, not that vertically integrating semiconductor production is going to be fun, because I can assure you that it won't be. It will likely be excruciating for Tesla to learn and pull this off. And yes, it's great that Tesla is going to be effective partners with Samsung and to a degree TSMC, maybe Intel in time because these companies with their fabs will help Tesla shorten its learning curve significantly. But how eager will they be to give Tesla everything since in a sense Tesla's becoming a competitor to TSMC and Samsung? Now, to be clear, it's going to take Tesla many years before they can actually start buying materially less chips from companies like TSMC and Samsung. But remember, long-term, maybe when other legacy automakers do decide to license Tesla's FSD in the early 2030s, well, in that scenario, they're likely going to need AI 5 6 or 7. And if Tesla is making it at that point on their own, specialized for Tesla's software, then you already know where all of these legacy auto companies would go to get those chips. Tesla's Terafab, not TSMC or Samsung. Fundamentally, what I want everyone to understand is twofold. First, how much harder it's going to be for Tesla to learn to fab silicon compared to how hard it is to make lithium batteries. And I'll say adding dry cathode into the mix would narrow this gap a bit because that does add significant scaling challenges to battery production, but silicon manufacturing is still in a league of its own. And for me, this move by Tesla is not really about what it could mean for Tesla's bottom line. It's far more about Tesla being able to achieve scale without chip constraints. And if Tesla goes forward with this, it is a major risk. And things are changing very quickly in this space. Example, on the Q3 call this year, Elon said TSMC and Samsung would make AI5 in the US. And then just yesterday, Elon is now saying that the only Samsung fab Tesla will go to for AI5 is in Korea. So Samsung in Texas is apparently off the table. But that's just for AI5 because don't forget that $16.5 billion deal that Tesla signed with Samsung to make AI6 in Taylor, Texas, where Tesla became Samsung's first main customer. Which is why I said it gave Tesla a lot of leverage because building out a fab for the first customer means the plant is mostly customized to that company's chip requirements. It's a fascinating story. In short, Samsung's Fab in Taylor started out with four nanometer production, but pivoted to two nanometers once the AI compute explosion took off. Just know a jump like this does not mean just buying a few different machines. It's effectively rebuilding the entire factory. So, what started out as a $17 billion buildout for Samsung jumped to nearly $50 billion and pushed the timeline out by years. The construction of the foundation of the FAB alone was pretty insane because they had to mitigate vibrations from the Earth. Making these two nanometer chips is basically atomic surgery. One small vibration and the wafer is ruined. Samsung also needed a dedicated water factory. The word is the Samsung Fab in Taylor consumes 50 million gallons of water every day, nearly five times more than the entire city of Taylor itself. And without water, the entire production line stops. The air control systems are pretty absurd because even a small particle of dust or moisture can also ruin production. The grid and the energy intake system at the fab to insulate it from the unreliable Texas grid is also very involved. All this to say, we have one of the most complicated and massive structures to make atomicle ships. It really is pretty fascinating stuff. But the real problem for Samsung is actually the people, the talent. Samsung is reinventing everything from scratch in a new country. And with all of these challenges and delays, that's when Tesla stepped in. Elon will be walking the lines. The Fab's going to be tailor fit to Tesla's needs. Samsung finally gets its anchor customer in Tesla, and then Tesla gets a foot in the water learning the ins and outs of this incredibly involved process. But look, there's still no guarantee that this all comes together and ends up successful if Tesla and Samsung can't get to reasonable yields. Now, one bright side for Tesla is they don't have to worry about landing customers for this Terraab. They are the customer. So, just a few quick things. WSPM is wafer starts per month. Elon said at least 100,000 WSPMs starting out for its Terraab. And maybe that would be one of 10 in a complex. So ultimately maybe 1 million wafer starts per month which elucidated the giggle factor from the audience as Elon called it. For comparison, TSMC's Fab in Arizona is only 20,000 WSPM. They of course still do most of their production in Taiwan where costs are cheaper. And now we don't know the die size of AI5. So it's impossible to know how many AI5 chips Tesla could make from one wafer, but it could range from 100 to 300 chips per wafer before yield losses. Just think of a die as the raw circuit that's cut from the round wafer. It's not an actual finished chip. So, a wafer producing 100 dyes might yield 80 or so finished chips like AI5 chips after accounting for yield losses and packaging. And then for new production, a 60% yield is standard, but that can improve to closer to 80% yields over time as production matures. So, let's assume 100 dyes per wafer for Tesla. Applying a 70% yield, that would be 70 chips like AI5 per wafer. Then at 100,000 wafer starts per month, that would be about 7 million chips per month or 84 million per year. Now listen, as I said, these numbers can vary greatly given the actual die size and the yield rates. But either way, I'm running this back of the napkin math to give you some context for the rough scale that Elon is thinking about. Because even if millions of chips go to cars and Optimus every year for Tesla, that would still leave millions left for data centers. And Elon did say recently he wants an excess of AI series capacity. And it sounds like this might be the way to actually make that happen. So definitely a big deal, something none of us, I think, were expecting. And so I'll be eager to hear a lot more about this from Tesla. And just some quick numbers, TSMC is the clear top dog in the space. They do 70% of the global foundry market, which is why Tesla will be buying chips from three TSMC factories in Taiwan, Arizona, and Texas, and only from one Samsung Fab. Again, at least for AI5. But TSMC's revenue for 2024 was $80 billion, basically 100% from Foundry Revenue. Its net profit for the year was $34 billion with gross margins of 53%. Now, of course, even if Tesla eventually hits the numbers Elon mentioned, 1 million WSPM, TSMC's global WSPM across all of its factories is 12 million. So, right now, even Tesla's most ambitious plans are still not even going to come close to the size of what TSMC is doing. But really, given how much Tesla's already working on, I think that should be obvious. And there are some other profitable fab companies, they don't have to get to the scale of TSMC to be profitable, but none of the others are close to TSMC's profitability. So, even though Tesla will do an order of magnitude less than what TSMC is doing, it could still drive very substantial savings for Tesla not having to pay margins to other companies. And if Tesla ever wants to sell any of the AI series, they would then have a fully integrated line to start generating revenue and profits of its own. But again, it would likely take Tesla 3 years to build out its own fab and then another 1 to two years of early testing production. So for anything to actually hit Tesla's bottom line positively, we're talking into the 2030s. But as I said, I think access to massive volumes is the far more important piece here for Tesla to power millions of Optimus robots and other vehicles. But part of the conversation people need to prepare for and why I said on X it's a major undertaking. TSMC's capex in 2024 was $29 billion. Keep in mind these numbers can vary significantly from one company to another as this is such a complex process. But construction costs are likely to be around 5 to 10 billion for a clean room and the facilities. The EUV lithography machines can cost hundreds of millions of dollars each and a 20,000 WSPM fab could require between five and 10 machines for advanced nodes. There are other tools for etching, deposition, and metrology, and those can add up to over a billion dollars in capex. And then there's R&D and training the workers. And so even one fab doing a fraction of Elon's long-term aspirations is likely to be well over a $10 billion investment. Don't forget, as it stands now, Tesla's entire capex spend is projected to be $9 billion for 2025. So, yes, this is a massive undertaking for Tesla and just goes to show how critical Elon really thinks this is and most importantly how serious Elon is about this AI wave crashing and the scale finally finally being reached after years of waiting. And since I know at this point you'd like a quick break, some of you asked for an update on the 3W mats. So, here it is. And I get it. It's probably the biggest question with floor mats. How easy are they to clean? And when you're done, do they look like new or is it like a wrinkled piece of paper? Because sadly, I've had mats in the past that are great, brand new, but they were brutal to keep clean. So, here's a picture of Ashley's driver side mat from 3W, the sponsor of this video, after a few weeks of use. No snow here yet, but we've had plenty of rain. And frankly, I think these stay clean better than most. Again, some others you can look at them the wrong way and they seem to end up looking dirty. But this does not make for a good stress test. So, in honor of AI Driver, here you go. I got a bunch of moist dirt, dropped it on the mat, and then stomped all over it to really grind it into the mat, simulating aggressively what happens when people get in and out. Then, I let it sit there for a while to let it seep in. And then after that, I just sprayed it with a hose, cold water, and used a Scrub Daddy with some Dawn dish soap. The soap is not required, but I like to be thorough and kill some germs. Then a quick rinse, which brings me to a key point. The bottom of these 3W mats do not have that fiber mesh-like material that many others do. And that's what takes forever to dry and can hold on to smells. Whereas with 3W mats, you can tell why they stay in place. And the rubber is super easy to clean and dry. And here you go, the finished product after like 90 seconds total. And here's the before and after. So truly, guys, we love these mats in every way. the fitment, how they stay in place, durability, aesthetics, there's no smell, the ease of cleaning and installing, and so on. And a heads up, from November 10th to December 2nd, 3W will be running a Black Friday campaign on their official website featuring a lucky draw event where customers can win up to $300 in gift cards. So, you can use my links below to grab 35% off the 3W website or $45 off on Amazon. and my codes for each will be in the description as well. And many thanks to all of you for supporting the channel in this way. Just know the risks here are even greater than it was Tesla taking on battery production. Some of these transistor sizes are nearing quantum limits. There are thousands of process steps for every wafer, and a single misalignment or particle can ruin a chip. Relying on these hyper specialized tools like EUV machines that a very few number of companies make can create bottlenecks in the supply chain. Chips have much stricter material purity standards than lithium batteries. Chips demand greater precision operating near atomic scales. Chips are more complex thanks to the number of steps and interdependence of them. Chip defects are costlier thanks to lower yields. So I think you get the picture. Tesla waiting into the space possibly manufacturing near frontier silicon level chips is going to be very hard and quite expensive. So it is a risk for Tesla and we all have to be honest with Elon's proclamations. Elon said Tesla was aiming to produce 100 gatt hours of battery cells annually by 2022 and a long term he predicted 3 terowatt hours produced by 2030. uh push towards our goals which are >> this is just talking about uh Tesla internal cell production as I tweeted out earlier we will continue to uh use our cell suppliers the Panasonic and uh LG and CL so this is 100 gatt hours supplemental to uh what we buy from suppliers and then long-term we're uh expecting to make on the order of 3,000 gawatt hours or or three terowatt hours per year I think we've got a good chance of of achieving this actually before 2030 but I'm highly confident that we could do it by by 2030. >> So, yes, Elvon made it clear that was 4680 production, not including supply from partners like Panasonic. And fast forward to today, Tesla is likely only doing somewhere in the ballpark of 20 gawatt hours of 4680 cells per year in 2025. Now, we haven't had a real update for about a year or so. And at that time, Tesla was at a run rate of about 17 gatt hours per year. So, pretty far off from 100 gawatt hours a year by 2022. But hey, Elon admitted the dry cathode was a mistake. And hopefully if Tesla does go this foundry route, these predictions will turn out a bit more accurate. And then to wrap up the section, Elon said AI5 will be one-third the power of a chip like Blackwell from Nvidia, which is impressive in its own right for roughly comparable performance and much less than 10% of the cost, which again, if they pull it off, just proves that Tesla is leading the world in inference compute efficiency. And further, it's going to be completely optimized for the Tesla software stack. Tesla can hyper specialize and design it exactly how it's best for Tesla, which is a huge advantage relative to Nvidia having to build something generalized for basically everyone else, which is fine for Nvidia, but you can see how it will hamstring the companies buying the chips and paying the margins to Nvidia. They just won't be as efficient as Tesla no matter what they do. But Tesla's expecting AI5 to unlock safety 1,000% better than a human. Whereas with AI4, they're expecting it can get to maybe 400% better than human safety. But we're still left with the nagging question, how is Tesla measuring human safety? So, as mentioned, we'll be covering this closely in the months and years ahead, but something like this would arguably be the riskiest thing Tesla has ever done. The upside is Tesla has access to a nearly unlimited supply of chips and it doesn't have to pay 50% margins to TSMC. But the downside is it will take Tesla decades to earn an ROI from the heavy capex burden. But the game for Tesla right now is scaling. And again, there's at least a chance Elon was posturing here and applying pressure to TSMC and Samsung, maybe to Intel to have even more leverage in those conversations as they negotiate these partnerships. Because I can assure you, Tesla building a terrafab is not something Elon wants to do for fun. And if Tesla does it, it's only because it's something they feel they have to do. Now, let's switch gears and cover a few quick hitters. Elon still sees a path to getting below 20 cents per mile at current year dollars, meaning he's not accounting for inflation, meaning the actual number might be closer to 25 or 30 cents per mile in a few years once robo taxis have scaled. Elon casually dropped this line >> and I think uh there's going to be a lot of solar powered AI satellites and um I think Tesla's going to play a role in that. >> [applause] >> Elon also said earlier how he wants to use these solar powered satellites equipped with AI to dynamically adjust the amount of solar energy reaching the Earth, acting as a cosmic thermostat to mitigate global warming, which sounds like a perfect collaboration between Tesla and SpaceX. And that makes Elon's comment about wanting to find a way for Tesla investors to invest in SpaceX all the more exciting. But there will definitely be hurdles to accomplishing this as there are limits to how many people can invest in a private company before said company is forced to file and report as if they were a public company. So I would not be expecting thousands of Tesla investors being given exclusive access to investing in SpaceX while it's still private. Maybe some but not at scale. Elon did however say that maybe eventually SpaceX should go public despite all of the downsides. The shareholder meeting for Tesla next year is set to be at a much bigger venue, so way more people can go. Shout out to Alexandra for asking the question. The Cyber Cab and Semi will both use 4680 sales despite Elon admitting that dry cathode was a mistake as it turned out to be too difficult. So, interestingly, Tesla is still going to scale 4680 production with the dry cathode. So, Tesla is implying they'll figure it out at least to some degree. But apparently the juice was not worth the squeeze. Cybercap production is starting in April of 2026. As we predicted, Tesla does have inside information on regulations for cars without wheels and pedals. And Elon said he's expecting the regulatory approval pace to match that of the Cyber Cab ramp. It might be tight, but he thinks that's about right. What this means is the regulations are not expected to be a bottleneck for cyber cab deployment which significantly hurts the chances of it ever shipping with a wheel and pedals. Sorry, Farzod. China has told Tesla to expect full regulatory approval in February or March of 2026, an upgrade from the partial approval Tesla has now. I do wish Elon would have elaborated a bit more on that. The robo taxi service will be expanding to Las Vegas, Phoenix, Dallas, and Miami. Next, there's a new Tesla semi design with a front light bar that looks a lot more like the Model Y than the prior semi design did. And as we should expect, Dan Priestley said, there will be an immense number of improvements in next year's Semi, stemming from pilot fleet learnings and amazing innovations from the Tesla teams. The robus or the rebove sounds like it's a few years away at the earliest. The focus the next few years will be on Optimus, the Cyber Cab, Semi, and the Roadster. The Roadster unveil is now scheduled for April 1st, 2026, which gives Elon an easy out if Tesla misses that date. And given that just a few days ago, Elon predicted the Roadster demo would be at the end of this year, I'm not holding my breath for April 1st either. and then production 12 to 18 months after the demo, which puts us back on track with what we talked about. Don't be expecting the Roadster in 2026, and 2027 is still not a guarantee. Elon said Tesla is almost comfortable allowing people to text and drive, which is huge news and will be the gamecher we've been talking about. Elon said he's confident in the next month or two they can look closely at safety stats and allow us to text and drive essentially. Of course, Elon had to add that qualifier in there, but this is the closest statement we've got to Tesla solving unsupervised. He said Tesla is within a few months of having unsupervised autonomy solved at a reliability level significantly safer than a human. There are major changes planned with 14.2 2 and then by 14.3 you should be pretty much able to fall asleep and wake up at your destination. So if you want to talk about chat GPT moments, that's the one. Fingers crossed it could finally be right around the corner after all the waiting. But we still have questions because most states currently ban texting while driving. But remember, there's a difference between what the state laws say and what Tesla will require of its customers. Tesla could simply reduce the nag to the point where you get maybe 30 seconds or so to send a text or not pay attention. But if you were to get pulled over for texting, you may not be off the hook for saying Tesla said that I could. So it's murky right now and it's state byst state, but we'll see if Tesla finally follows through. And then there's Optimus. Tesla's aiming for a COGS for Optimus of $20,000 when they reach about 1 million units per year. We got a behindthescenes look at Tesla's pilot production line in Fremont, but it's only for prototypes and research and development. The scalable production line will come online in 2026 and will be totally different than what you're seeing here, which is good because in its current form, it's clearly a very manual and human-driven process. Elon said this would be the fastest production ramp ever. They're building the 1 million unit production line in Fremont, which will be line one. And then next, they'll build a 10 million unit operation in Austin. Elon continues to hype the V3 hand design, saying how Optimus will perform surgery beyond human capabilities. So, this type of commentary is to be expected at a shareholder event where the point is not just administrative voting, but to rally the troops and paint the picture of the future, to share the vision. And I do think ultimately the Optimus production lines will scale quickly, but I'm confident that will not be in 2026. We'll likely get low-level production as the supply chain is established and Tesla works through all of the challenges. So, I'm not expecting Optimus production in the tens of thousands until 2027 at the earliest. Elon said Tesla wants to increase vehicle production by 50% by the end of 2026. Maybe hit an exit rate of up to 2.7 million units at an annualized pace. Aiming to get to maybe a 4 million unit annualized rate by the end of 2027 and possibly 5 million by the end of 2028. But these are aspirational goals and the supply chain would have to grow with Tesla's increased production. But right now, Tesla's only producing around 1.7 million vehicles per year. So the only way this type of growth is going to happen is if Tesla starts scaling robo taxis dramatically for its own fleet and then if unsupervised one becomes real for customers and two becomes widely known and appreciated. To be clear, Elon did not say Tesla was going to make 2.7 million cars next year. He just said the production rate at the end of next year would be at that pace. Meaning Tesla might only make 2 million cars next year, but the production rate will be meaningfully increased by the end of the year. And maybe Tesla scales the robo taxi fleet to the tune of 50,000 by the end of 2026, which means the rest of the extra production would need to go to customers. And for that to happen, unsupervised is going to need to be widely available. So Elon wasn't joking when he said it's going to be a new book for Tesla, not just a new chapter. Not only will Tesla be spending tens of billions of dollars on training compute, but they might be spending tens of billions of dollars on a new chip fab, which means a $40 billion war chest is not as deep as it once was and will actually require Elon to start pulling rabbits out of hats as he said. And we still have plenty of things to dive into in the weeks ahead, like what Elon said about the integerbased system and future Mega Packac iterations, some cyber cab design tweaks. But I don't want this video to be too long so that hopefully more people will watch it and understand the new risks that Tesla may be taking on here. It really is not trivial. But we know this is what Tesla is, a missiondriven company set out to solve the hardest problems on the path to providing sustainable abundance for all. The risks are now higher, the stakes are now higher, but the upside and the potential for Tesla to further differentiate itself from any and all competition is also higher than ever. Being a Tesla investor is not for the faint of heart. It never has been. But that statement is now truer than it's ever been. Delivering things that have never been done at scale that's never been seen is a feat that most of us non-engineers will just never truly comprehend. So, as I always say, patience will be required and not all timelines will be met. But as has always been the case with Tesla, it will likely feel like nothing is happening for a long time and then all of the sudden you look around and everything is different because with Tesla everything tends to happen at once in a very condensed time frame. And that time frame is shaping up to be sometime over the next 3 years where the fruits of Tesla's AI labor the past 5 plus years finally start to sprout and become visible to the rest of the world not paying close attention. Elon is now locked in and incentivized for roughly another decade at Tesla. So it's now all systems ago for the greatest corporate show on Earth. And as always, I'm very excited to continue documenting this journey with all of you. And don't forget to check out 3W mats linked below for those interested. I think you'll be glad you did. [snorts]