Terafab Saves Tesla Trillions - Napkin Math

Kind: captions Language: en The last thing I want to talk about today is a deep dive on Terafab. You've got all these fabs making chips and Elon has said we can't get enough chips and you know I did the math. You know, I'm known for mad napkin math and I did some napkin math here. And on the one hand, you know, Elon wants to make sure he has an adequate supply of chips. They're planning on making 100 million or more Tesla Optimus bots a year. Where do you get 100 million chips from? That's going to be challenging. Plus you might need 40 million for the cars. Plus you might need you know, SpaceX needs them for the satellites. You need them for your data centers. There's a huge demand for chips that Tesla and SpaceX have together. So how do you do that or if you if you're not confident you can get them from your suppliers and I think that's what Elon is reflecting what benefit can you get by doing it yourself? And it sounds like it's a crazy idea, but when you do the math, this looks pretty compelling. So roughly Elon has said that the AI 5 chip or the AI 6 chip is going to cost them about $6,000. If you dive into what's going on in these fabs and the process of getting your chips, an Nvidia Blackwell 200 might cost $40,000. And then Nvidia's making like 70 80% margin. So okay, it's really less than $10,000 that their cost. Now they're paying let's say TSMC to make the chip. Well, TSMC is getting a 50% margin. TSMC is buying the electric the lithography machines from ASML. They're getting 50% margins. ASML is buying optics from Zeiss. They're getting 50% margins and you end up with you know, this what's the real cost of this $40,000 chip if you're able to knock out some layers of cost. I think the game plan for Tesla and SpaceX is to build their own fab which is a huge endeavor, but the value you get out of it long term is number one you get secure supply of the chips that you want. You don't have to worry they're not going to be available or you have more control over the availability of the chips. But number two, how much money do you save? So let's assume that they need about 150 million chips a year. 100 million for the humanoids, 40 million for the vehicles, 10 million chips a year for the space AI data centers. I think there's going to be some earth level data centers. They're still going to be buying you know, Nvidia chips and other hardware for the data centers, but let's just go with these numbers and say they want 150 million chips a year. Over 10 years it's 1.5 billion chips. Now Elon has said they're paying about $6,000 a chip for the chip that goes in Optimus. If you're able to just save some of the money. I'm not saying they're going to get down to like the $1,000 cost level. I think it's conceivable that with what Elon's talking about doing with Terafab, they're able to knock out a couple of layers of cost and really drive their car they're able to vertically integrate a little bit more and maybe knock the lithography player out of the game or maybe they make their own optics or something that that knocks some of that margin out and lowers their cost. If instead of paying $6,000 because they're paying 50% margin to let's say Samsung to make their chips, if they're able to get their cost down to $3,500. And I think this is if you know Elon, I think this is pessimistic. This is a conservative estimate. You're saving $2,500 a chip. So that that's a good start, right? You're saving $2,500 a chip on 150 million chips a year for 10 years or 1.5 billion chips. Well, how does that add up? The total economic value is $3.75 trillion. So even if you haircut demand whatever, you get to about $2.5 trillion in savings. So the investment ballpark $200 billion to build out the three to four fabs that you're going to need. Elon's pretty good on capex efficiency. Maybe he's able to get that number down, but to be conservative, they're going to burn about $200 billion in cash getting these the Terafab built. I think Elon wants to go bigger than these numbers, but let's just work with these numbers for now. So you got total capex of 200 billion. You've got $2.5 trillion in savings. The net value $2.5 trillion in savings over capex of 200 billion, you've you've effectively saved $2.3 trillion. Again, I think these numbers end up being bigger and you end up saving more money, but this is a good picture. You know, this this doesn't include the iterations of chips. You go from AI 6 to AI 7 to AI 8. You've got the D3 chip for the space center space space data centers. Maybe you're going to have a D4 chip. Those are going to be curveballs, but the cost savings are going to be roughly the same every iteration of chip. So you're potentially saving $2.5 trillion. My math maybe you're saving four or five trillion dollars on a $200 billion capex investment. You have to have the demand that Elon sees to do this and they do. So the total economic value $2.5 trillion in savings, the $200 billion capex, a net value of 2.3 trillion, a return multiple of 12x. You're getting a 12x return on investment over 10 years. But 12x return on investment is a really good return on investment over 12 years. You're making your money back in you know, the first couple of years, two three years you're making your money back and then you're just printing effectively savings. So if you think about it at a high level, Nvidia is capturing system level margin, TSMC captures manufacturing margin, ASML captures tool margin, Zeiss captures optics margin. There's a whole bunch of players in the system that are taking margin out and that's increasing the cost. The value is fragmented. If they're able to vertically integrate, Tesla and SpaceX are able to capture the design margin. They've already got that one with their chips. They go straight to TSMC and Samsung without using Nvidia for their specialized chips just like Google does with Broadcom. Then maybe you're able to save some money on the manufacturing margin by building your own fab instead of paying TSMC or Samsung and maybe you're able to save a little bit more in the intermediate layers by knocking out some supplier margins here and there and that collapses and then maybe there's some upside later that maybe you end up with excess capacity. You're able to sell chips externally. You're able to license the process and packaging. There may be some other benefits down the road from this, but I just wanted to give you a sense of Terafab makes a lot of sense if you dive into why that I think $200 billion is a very very large investment, but when you see that you have the demand for all these chips, I don't know that anybody else does. Maybe Google does. Maybe it would make sense for Google to build its own fab. Because they have specialized chips and they work with Broadcom to make them. So if they built their own fab, they're still paying Broadcom on the design margin, but maybe they can save some margin. I think fundamentally Google The other side of this is if you're one of these companies that's making this high-end stuff, your margins are so high that it sounds like it's a lot of savings, but maybe those savings don't really matter compared to the scale of what you're doing. And you want to focus your effort on what you're doing what where you're really adding a lot of value and trying to say instead of trying to save a little bit of money here. But when a little bit of money turns into $2.5 trillion, maybe that matters. That's My my take is that matters enough that that it makes sense that Elon sees it. Securing your supply chain for this critical piece of your game plan and saving a ton of money on it while still buying stuff from Nvidia, still buying stuff from Samsung, still buying stuff from TSMC. They're still going to be buying hardware from them, but their demand is so high they need to make their own.