WEBVTT 00:01:09.440 --> 00:01:10.800 Guess I can kill this. 00:06:28.160 --> 00:06:29.200 Check, check, check. 00:06:30.640 --> 00:06:30.960 Nine. 00:06:31.600 --> 00:06:32.320 That's from the phone. 00:06:32.960 --> 00:06:34.240 That's from the nine choos. 00:06:34.560 --> 00:06:35.520 Okay, nine choos. 00:06:35.920 --> 00:06:36.400 Okay, nine. 00:06:42.240 --> 00:06:43.360 Chess is gay nines. 00:06:44.960 --> 00:06:46.000 Chess is game nines. 00:06:46.240 --> 00:06:47.360 Chest chest. 00:06:47.600 --> 00:06:47.920 Chip. 00:07:09.040 --> 00:07:14.400 Chiek, check, che when I was preparing when I was preparing 00:07:16.640 --> 00:07:19.360 I just visualized was ET. 00:07:19.760 --> 00:07:28.960 So ET going across the moon, you know, with a boy on the bicycle and a contraption to communicate to call home. 00:07:29.200 --> 00:07:29.680 Obviously. 00:07:30.240 --> 00:07:32.240 Predecessor to Starlink. 00:07:32.720 --> 00:07:35.040 So I guess that didn't work. 00:07:38.400 --> 00:07:53.200 So so so Elon has uses a principle you know first principles about everything should be done the right way, not the fastest way, not the quickest way, not the cheapest way, the best way. 00:07:53.360 --> 00:07:55.200 And everything looks at all over again. 00:07:55.280 --> 00:07:59.600 So if someone makes an investment on his behalf for one of his businesses, 00:08:00.000 --> 00:08:04.160 it's always viewed in the terms of is this the best we can do? 00:08:04.480 --> 00:08:08.320 Best we can do, the fastest we can do, the most effective. 00:08:08.640 --> 00:08:13.200 So so one thing I would say is that he's a builder. 00:08:13.440 --> 00:08:17.760 He's probably the only builder I'm aware of in the United States who builds stuff. 00:08:17.920 --> 00:08:24.720 But builds stuff that's not just regular stuff, like ten million square foot buildings, five million, twenty million. 00:08:24.880 --> 00:08:29.680 Now he's got a hundred million square foot building for terapab that he's thinking about or going 00:08:30.160 --> 00:08:50.240 to build and and and so the question and no one else has done this no one and anything like it and so the question is of course and when you're doing such a thing then you get scaling benefits you get to buy things produce things much more effectively than you could 00:08:50.240 --> 00:08:59.760 if you were in a smaller entity so the question is how do you manage to do this and why didn't anyone else ever try to do anything like this we're trying to get silicon to come back to the United States 00:09:00.320 --> 00:09:03.600 Again, we need it, we need power, we need silicon. 00:09:03.920 --> 00:09:05.680 No one's doing this except for you. 00:09:05.760 --> 00:09:06.480 How come? 00:09:06.880 --> 00:09:13.840 Well, that's actually that's not quite accurate in that TSMC and Samsung are building fabs in the US. 00:09:14.000 --> 00:09:20.000 So TSMC is building a large collection of fabs in Arizona, and some of them are, I believe, operational. 00:09:20.720 --> 00:09:29.840 And Samsung has some operational fabs in Texas and is is building, or has built, I should say, a fab for Tesla that will be dedicated. 00:09:30.480 --> 00:09:36.560 To the AI-5 and AI-6 chip designs that Tesla's done. 00:09:36.800 --> 00:09:45.120 So, but relative to what's being done in Asia, the amount of chip application in the US is very small. 00:09:45.200 --> 00:09:59.760 And even if you take the best case assumptions of all the chip fabs that that have been announced that will be built in America, that's still a tiny fraction, perhaps on the order of 10%, optimistically 10% of what's in Asia. 00:10:01.120 --> 00:10:07.360 So as I think most people know, Taiwan does an incredible amount of heavy lifting on chip production. 00:10:07.680 --> 00:10:11.280 TSMC is an amazing company, truly incredible. 00:10:11.600 --> 00:10:16.640 The sheer number of a variety of chips and the magnitude and the volume is incredible. 00:10:17.360 --> 00:10:18.640 That TSCMTC does. 00:10:18.800 --> 00:10:19.920 Same thing for Samsung. 00:10:20.400 --> 00:10:25.360 And I think Micron does a great job too on the memory side. 00:10:25.600 --> 00:10:29.680 So but the but it's overwhelmingly built. 00:10:30.320 --> 00:10:30.880 Asia. 00:10:34.800 --> 00:10:42.080 So my concern really is that there is always like some geopolitical risk in that region. 00:10:43.440 --> 00:10:57.360 And I think that that risk may come to fruition at some point, in which case if we don't have all the pieces necessary to build chips at scale in America, then we may not have any chips at all. 00:10:57.440 --> 00:10:59.760 Or very few, the 10% number type. 00:11:00.320 --> 00:11:00.800 thing. 00:11:02.080 --> 00:11:12.800 And then there's the the the maybe looking ahead you know past any sort of medium-term geopolitical bumps in the road. 00:11:15.360 --> 00:11:23.680 Even if you if you exclude geopolitical crisis issues, the rate at which we're making chips is simply too slow. 00:11:24.080 --> 00:11:29.680 So if if you if you want to get to say a terawatt a year of chip production, which is our goal, that's 00:11:30.000 --> 00:11:31.760 That's the reason it's called the terapab. 00:11:31.920 --> 00:11:36.720 It's sometimes people, when I say terapav, there's you know three possible meanings to terapab. 00:11:36.800 --> 00:11:40.000 One is terror, t-e-r-r or 00:11:41.680 --> 00:11:44.320 which I think sometimes people are like, why would you want a terrafab? 00:11:44.880 --> 00:11:52.880 and then there's tera, which just means earth, but it's actually t-e-r-a, as in terawat. 00:11:53.360 --> 00:11:57.200 so but arguably all three apply. 00:11:57.680 --> 00:11:58.000 So 00:12:02.560 --> 00:12:03.840 It's terrifyingly large. 00:12:05.920 --> 00:12:06.400 difficulty breathing. 00:12:06.640 --> 00:12:13.600 So so the rest of the world is making we'll make this year around twenty gigawatts of AI chips, maybe twenty-five. 00:12:14.000 --> 00:12:24.320 and then next year, depending on who whose projections you believe, it's maybe around thirty-five to forty five gigawatts of AI chips next year. 00:12:25.680 --> 00:12:29.760 and but but if if 00:12:30.160 --> 00:12:38.800 Want to do a a terawatt, that's now you need you need 30 times so good what current global production will be roughly next year. 00:12:39.120 --> 00:12:40.400 For just stock. 00:12:41.600 --> 00:12:42.880 We're building that for us. 00:12:43.280 --> 00:12:44.000 Stock pot? 00:12:44.400 --> 00:12:45.200 Well, yes, we're building. 00:12:45.760 --> 00:12:52.320 I mean, we can tell it to other people too, but basically, we're building 30 times what the annual production is in one plant. 00:12:52.480 --> 00:12:59.840 What you think is going to do as value as big a production as the GDP in the world, as a GDP for the 20th lowest. 00:13:02.480 --> 00:13:02.640 Yeah. 00:13:05.680 --> 00:13:15.520 So no no obviously you know a a plan like this I think if it fails it'll be crazy. 00:13:16.000 --> 00:13:17.280 If it succeeds it'll be audacious. 00:13:17.520 --> 00:13:17.920 Look at that. 00:13:18.160 --> 00:13:20.240 so it it sort of depends on the outcome. 00:13:20.480 --> 00:13:22.720 But I I think it will succeed at least to some degree. 00:13:22.880 --> 00:13:25.840 so yeah. 00:13:26.000 --> 00:13:28.560 Yeah, I mean it's gonna cost a trillion dollars. 00:13:32.080 --> 00:13:33.200 Problem other than that. 00:13:37.520 --> 00:13:38.800 but we won't need that money immediately. 00:13:38.880 --> 00:13:39.920 We'll be building faces. 00:13:40.720 --> 00:13:52.880 and and we are looking at ways to change the manufacturing process for for chips to improve the efficiency of of chip production as we've done in m many other arenas. 00:13:55.280 --> 00:13:59.680 like just one example, like like right now, the amount of time that a wafer in 00:14:00.160 --> 00:14:05.920 Chip app spends simply being transported and waiting is more than half the time. 00:14:06.080 --> 00:14:11.840 So more than half of what a chip is doing in a chip app is nothing. 00:14:12.560 --> 00:14:15.760 It's either being moved around or waiting for a machine. 00:14:16.080 --> 00:14:24.880 So and and I think there's basically a lot of a lot of potential for improvement in chip fabrication. 00:14:25.280 --> 00:14:27.360 Why don't these people do it themselves? 00:14:28.000 --> 00:14:29.680 If they're if they have plans just 00:14:30.160 --> 00:14:32.400 Sit with chips half the time just not moving. 00:14:33.840 --> 00:14:35.920 What's to stop them from improving their own systems? 00:14:37.440 --> 00:14:45.520 Well no, actually if you if you visit a chap a fab, you'll see that the the the the fat the the the wafers are moving, they're zipping all over the place. 00:14:45.760 --> 00:14:47.680 and and waiting. 00:14:47.840 --> 00:14:54.640 But but but the the amount of time that the atoms are being changed on the wafer in a way that is useful is a is a is less than half the time. 00:14:55.840 --> 00:14:59.600 and I'd say sometimes it's maybe twenty percent of the time that that that something 00:15:00.400 --> 00:15:03.120 Useful is happening to the atoms on that wafer. 00:15:05.600 --> 00:15:14.720 The reason that they don't want to try something adventurous or or with chip apps is that these things cost a tremendous amount of capital. 00:15:15.760 --> 00:15:22.080 And if they have a recipe that works, they just they simply don't want to risk something new. 00:15:22.320 --> 00:15:29.840 So generally the preference in the chip industry is to repeat what they did last time with. 00:15:30.160 --> 00:15:34.800 Minor changes and and not take any risks. 00:15:35.680 --> 00:15:46.080 So we we we're building a research chip fab in Texas, just in Austin, at the at Giga Factory, Texas. 00:15:46.240 --> 00:15:51.280 That research fab is going to try many different approaches to chip fabrication. 00:15:51.680 --> 00:15:58.080 and and many of those approaches won't work, but some of them I'm sure will work. 00:15:58.320 --> 00:15:59.840 And then we will translate those. 00:16:00.160 --> 00:16:08.240 improvements to the the Terra Fab itself, the production Terra Fab, which is in Grimes County in Texas, near Texas AM. 00:16:09.440 --> 00:16:15.680 and and that's that's that's where we'll scale up production of our AI chips. 00:16:15.840 --> 00:16:29.840 the I I think we we the one of the things that's gonna be interesting about the research fab in Austin is that under one roof we will have the ability to make d a do do logic memory packaging 00:16:30.560 --> 00:16:32.880 And do the lithography 00:16:34.800 --> 00:16:47.520 so that we can have we can design new chips and and in under one roof make the make the whole chip test it make the whole chip test it with a very fast iteration cycle. 00:16:48.320 --> 00:16:51.120 Something that would normally take a year. 00:16:51.360 --> 00:16:54.320 I think we we we may be able to reduce to a month. 00:16:55.440 --> 00:16:59.600 So if if for any given technology endeavor, the 00:17:00.320 --> 00:17:07.680 The pace of progress is determined by how many iter how many iterations can you do and how much progress can you make between those iterations. 00:17:09.680 --> 00:17:26.960 So if you if you can improve the speed of iteration and the and the r and like how how many iterations can you do per year or per unit time, and the progress between each iteration, in any given technology, you will you will accelerate your progress at a rate far greater than anyone else. 00:17:27.040 --> 00:17:29.680 If you if you are a hit, you know, if you if you're able to 00:17:30.160 --> 00:17:36.640 Iterate an order order of magnitude faster, you as a first approximation will advance your technology of order of magnitude faster. 00:17:37.040 --> 00:17:53.360 So most people are there's a tremendous amount of inertia in business because if someone makes an effort to spend money and it's not successful, then that's a black mark on them. 00:17:53.520 --> 00:17:59.840 And if it is successful and they're trying to get the top job of the business, then it just means that they have to work. 00:18:00.240 --> 00:18:00.720 Harder. 00:18:00.880 --> 00:18:09.200 So if they're successful, they work harder, and if it's if it fails, they're finished for their opportunity for the top job. 00:18:09.360 --> 00:18:26.800 Elon has established a culture in his business, business is, where it's always about come to me with ideas, and then if it's successful, if you think it's successful, and I believe that's the case, then we'll throw away what we've done before and start all over again. 00:18:26.880 --> 00:18:28.800 It's always about doing the best thing. 00:18:29.040 --> 00:18:29.840 One of the things that's 00:18:30.320 --> 00:18:44.720 Interesting to me about your culture is that you introduced me to Key Key Luck, the person who's in charge, Intel, who's now going to be have an interest in Terrafab. 00:18:45.280 --> 00:18:52.720 And so I spent some time with him and he comes to visit me and I said, So so what's it like to work with Elon? 00:18:52.960 --> 00:18:59.760 He says, Well, other people come to visit my facility in Oregon, and they come and they walk around for an hour and 00:19:00.320 --> 00:19:10.960 Half and they leave, and Elon comes and then he sits there, he was there for four and a half hours, and he interviews one engineer after another after another after another, questioning everything. 00:19:11.040 --> 00:19:12.080 That's the whole concept. 00:19:12.160 --> 00:19:14.160 So everything that he has to do is speed. 00:19:14.320 --> 00:19:29.840 And then when I went to Texas to Boca Chica to visit their car factory that they were building there, they introduced while it was under construction, they introduced me to the head of that facility, the manager, giant man. 00:19:30.560 --> 00:19:36.080 And and of course everyone seems like a giant to me, but he was a he was a really giant guy. 00:19:36.160 --> 00:19:39.040 And he had been previously the head of production at Ford. 00:19:39.120 --> 00:19:43.120 And so I said, Well, so what's it like, difference between Ford and Tesla? 00:19:43.360 --> 00:19:51.360 He said, Well, when I was at Ford and I had a change, I had to go through six different levels to get a change approved. 00:19:51.600 --> 00:19:56.240 When I have to have a change here, then all I do is walk upstairs and there's Elon. 00:19:56.320 --> 00:19:59.600 And so when a plant is something under construction, Elon's not 00:20:00.160 --> 00:20:01.840 Some distant guy, he's there. 00:20:02.080 --> 00:20:03.920 He even has an airstream. 00:20:04.080 --> 00:20:05.520 He lives in an airstream. 00:20:05.680 --> 00:20:18.400 Do you know any other executives of any big companies, potentially the largest company in the world, going to live in an airstream to make sure that all the questions that are taking place in making a facility are going to be answered fast. 00:20:18.640 --> 00:20:25.920 So, you know, it's an amazing culture, amazing person, that and amazing people who's recruited. 00:20:26.000 --> 00:20:29.680 So this whole people who work there, they're just brilliant people. 00:20:30.240 --> 00:20:32.960 It's like harder to get a job there than get into Harbor. 00:20:33.280 --> 00:20:36.160 Of course I don't know why people would go to Harbor anyway anymore. 00:20:36.880 --> 00:20:37.280 Yeah. 00:20:37.520 --> 00:20:43.520 Well by the way, S-streams are actually pretty nice, so I wouldn't say I'm like No, it's not that bad, honestly. 00:20:43.840 --> 00:20:46.080 I don't I kinda like my little air stream. 00:20:46.240 --> 00:20:50.880 but it's it's it's it's it's pot it's it's actually parked right next to the power plant. 00:20:50.960 --> 00:20:54.800 that we we so we built this power plant that's producing over a gigawatt. 00:20:55.280 --> 00:20:59.760 and probably it'll you know go to one and a half to two gigawatts over. 00:21:00.080 --> 00:21:06.400 Internal power pamp production in about a fifth of the time that such a thing would normally be built. 00:21:06.560 --> 00:21:11.200 So built in about a year as opposed to normally be about five years for that. 00:21:11.920 --> 00:21:15.680 And it's not it's not as though somebody has to come to me upstairs. 00:21:15.760 --> 00:21:20.960 There is no upstairs, so we have no other floor in in the at our Memphis Colossus II thing. 00:21:21.040 --> 00:21:29.200 I I walk I walk the entire site sometimes t two, sometimes three, sometimes four or five times a day. 00:21:29.440 --> 00:21:29.680 So 00:21:30.240 --> 00:21:33.680 the entire site, the power plant and all the buildings. 00:21:33.840 --> 00:21:42.480 and I also insist on doing at least some of the work myself manually. 00:21:42.720 --> 00:21:47.200 So so I've like I've pulled power cables the size of a an Anaconda. 00:21:47.600 --> 00:21:50.560 Like they're incredibly heavy and long. 00:21:51.440 --> 00:21:56.080 And I've I've done the fiber optic cabling. 00:21:56.480 --> 00:21:59.600 I I've done electrical switches 00:22:00.400 --> 00:22:04.240 I've worked on the on the power plant management software. 00:22:04.560 --> 00:22:13.520 It's so that it I I I try to do at least a little bit of everything so that I'm not numb to the pain of the people who are actually doing it. 00:22:13.680 --> 00:22:17.760 I have some understanding of the difficulty and what can be done to improve the situation. 00:22:18.000 --> 00:22:24.960 so it's not as though I'm sitting in an office and somebody has to come to me, I'm walking the floor, asking them questions actively. 00:22:25.840 --> 00:22:29.840 274 company 74 VMA Medical, 100 mail, 170. 00:22:30.320 --> 00:22:34.720 So let's talk a little bit about data centers, which are now in the news and people. 00:22:35.120 --> 00:22:36.720 Yeah, popular topic. 00:22:46.800 --> 00:22:48.480 Yeah, I'd say. 00:22:48.880 --> 00:22:52.000 so so it but but 00:22:54.080 --> 00:22:57.200 but if you want to have AI, you've got to have data centers. 00:22:57.440 --> 00:22:59.600 And so let's talk about them first on this 00:23:00.240 --> 00:23:00.800 planet. 00:23:01.120 --> 00:23:08.560 So so Elon believed that the the roadblock would be in the way of Didis