TomorrowX Summit November 18, 2026 Live from China, Musk talks with two close friends, Gavin Baker and Antonio Gracias, about the first Starship ship catch, becoming multiplanetary, rebuilding xAI's training stack, Optimus, Cybercab, Terafab, orbital compute, Neuralink and truth-seeking AI.
Live from China, Musk talks with two close friends, Gavin Baker and Antonio Gracias, about the first Starship ship catch, becoming multiplanetary, rebuilding xAI's training stack, Optimus, Cybercab, Terafab, orbital compute, Neuralink and truth-seeking AI.
Transcript Good. I mean, the audience should know, like, you know, Gavin and Antonio are two of my best friends. So, you know, you know, this is a best friend. This is a best friend buddy talk, basically.
It's a good way to. It's a good way to end, my friend,
It's good. I just got here. I mean, every time I come to China, I'm like, blown up. It's such an impressive country, honestly. I'm in Shenzhen and this is an enormous city with very cool buildings and I think the biggest hub of advanced technology in China. Whenever I come here, the scale and majesty of China, I'm amazed by. You know, I'm a fan of China. I say, I always encourage people to come visit China and just see for yourself. It's a very impressive country and I'm a big admirer. And, you know, probably at some point in the coming years, there will be some geopolitical bumps in the road over a certain country that starts with the letter T. But, you know, long term, I think there will be good relations with. There are pretty good relations with China. And then long term there'll be hopefully great relations. And it's, you know, I'm also coming. One of the reasons I came was to just also stop by the Tesla Shanghai factory and just thank the team for their incredibly hard work and all the great vehicles that they've produced, produced for Tesla over the years. So. And also the President's here and they're discussing trade relations, which I hope there was some success with that. I think to the degree that we can have good trade and good relations between China and the United States, I think that's good for the world.
Absolutely. Hey, man. So starship ship 42, it was flight 15, it was predestined, but can you just talk about Starship? What we learned, where we are on the path to reusability. Rapid reusability.
Yes. So the caching of the ship was a massive milestone, which I think only those who perhaps follow the SpaceX story well understand the magnitude of the milestone because we've recovered the rocket booster, it has flown back and gotten caught by the tower arms three times now. We've actually also reflown a boost stage and had caught it. So, yeah, so we've had three booster catches, but this is the first ship catch, which. Which means we've now at least proven that all parts of Starship can be flown to space and recovered. And recovered in good enough shape to be reflown so this, this is a. Yeah, just catching the ship was like the, the final major piece. Like can, can we, can we not. Not just catching the ship but it's, but it's in, it's in good condition with in this case only minor repairs needed to refly it. Now in this case, because it's kind of an historic ship, we're going to I think not take any chances with the ship and just kind of leave it for public display because what it's done is historic. No one has ever taken a stage like that and done created a truly reusable upper stage and certainly not brought it back and have it land propulsively while being caught by giant chopsticks. Which sounds like some kind of preposterous sci fi book or something. But it's done. And so what this means, say for practical purposes is that with full and rapid reusability, Starship can drop the cost of access to star space by a factor of 100. This is obviously a dramatic improvement and achieve a mass to orbit rate that is in the million ton per year range. Can you guys hear me okay?
yeah, everybody's cheering man. Let's have a cheer for starship master orbit. Yes, we'll go in Mars.
Exactly. So yeah. So in order to build a city on the moon or on Mars, you need to be able to transport a tremendous amount of mass to the moon or Mars because you've got to rebuild the base of industry. And there's the, you know, because when you buy some say computer, a phone or you know, in Milton Friedman's famous example, even a pencil, you're able to do so because there's this vast primitive industry that starts with mining and then refining at various stages and making the small parts, then more complex parts and then finally the finished product. So you've got to build that entire perimeter industry on the moon of Mars in order to have it be self sustaining. And the self sustaining part is important for the preservation of consciousness into the future. That such that even if there is some calamity on Earth, that consciousness and life as we know it propagates into the future. So it's I call like backing up the biosphere or life insurance for life collectively. And. But it also is the first critical technology stepping stone that allows us to become a true space faring civilization where traveling to space and to other parts of the solar system is not something that only a handful of people get to do. You know, a handful of government astronauts get to do but, but that anyone could do so that ultimately anyone could go to the moon or Mars if they want to go there or anywhere in the solar system. And once we once, once we, we're a civilization is a multi planet civilization within the solar system that will create a forcing function for making even better rockets and spacecraft and, and then, and ultimately going from our solar system to explore the rest of the galaxy and, and, and perhaps perhaps we'll encounter aliens we can talk to or, or perhaps encounter long dead alien civilizations that had a good like 10 million year run. Our civilization is only just over 5000 years old if you count the star civilization from the, from the the advent of the first writing. First writing was about 5,500 years ago in ancient Sumer with archaic Greek uniform. So we're very young, we're a flash in the so far as a civilization we're just a flash in the pan and we want to take the steps that ensure civilization survives for the long term. And I think we want to be, I think any self respecting civilization has got to be a multi planet civilization, okay? I mean we got to represent here and you know, we don't want to be one of those lame one planet civilizations. I mean come on guys. LFG we got to be a multi planet civilization and multi stellar civilization. And I generally classify the motivations here into two parts. One is the defensive reason which is that we want to ensure the continuous continuation of consciousness even if some natural man made calamity were to occur on Earth. And then the what I call the sort of forward looking or the sort of excitement about the future reason which is that a future where we, where we become a space spreading civilization and are out there among the stars is vastly more interesting and inspiring than one where we are not. And I always say that there's plenty of problems in this world and we want to solve those problems. But life can't just be about solving one miserable problem after another. There must be things that you look forward to that make you excited about the future, that make you glad to be alive. And the idea of being out there among the stars is one of those reasons.
Amen. The yeah, On that note, you know I was there, it was a super emotional day. I mean people were crying all over the place. And I think people here love to hear it's 25 year arc, high probability of failure at the beginning and still always a probability of failure. How did you feel about it man? How did it feel after 25 years of work together get there
Right. I mean you could see it in the emotional reaction, as you point out of the SpaceX team, everyone knew within SpaceX at least how important this milestone was and that it was really something fundamental and profound that that starship will be the first ever fully reusable orbital space system. You know, this has been the holy grail of rocketry from the beginning, is to have a fully reusable rocket system. You had something like say the space shuttle, which was partially reusable, but the main tank, which was also the primary structure for the airframe, that everything attached to basically that big orange tank was thrown away on every flight. And even the parts that were technically reusable actually took a very long time to refurbish. So generally the shuttle, orbiter, the plane like thing would take something like nine months to refurbish between flights. They basically had to rebuild many, many parts of it. Whereas Starship is designed for complete reusability and to be reused immediately like an airplane. So that means the booster lands, you can put another ship on top and take off immediately. And the net effect of this, like I said to put into economic terms, is a hundredfold reduction in the cost of access to space and maybe even eventually a thousand fold production. So that's what's necessary in order to, to make life multi planetary in order to build a city on the moon or Mars. Because if you don't reduce the cost by 100 to a thousand fold from what they've conventionally been at, then it would be financially impossible to do this. You could do it even if you had the whole budget of the United States government.
Hey man, we went, Brett was here yesterday, he did a great job. We went deep on orbital compute, terrestrial compute, the economics of them. Orbital compute happened happening in 27, enabled by Starship. But I did want to talk with you a little bit about Gro Grobot, Grokbot's most amazing AI product. I use can maybe talk about Grok 4.8, 4.8, when we're going to see that the progress you've made in rewriting the training stack, all the hard work, maybe talk about your first party AI services.
So we've really had to kind of, you know, started off as xai, which was really sort of. I started that as essentially a fairly small side project where what was happening is that I was encountering people who really wanted to work on large model AI, but didn't want to work on the real world AI that Tesla was doing. And so they would end up therefore having to go to OpenAI Anthropic or Google. And I thought, okay, look, I'm encountering these really talented people. They're only, they only want to work on large model AI. I didn't want to distract Tesla about this because Tesla needed to focus on getting full self driving, working for the car and the real world AI for the Optimus robot. So I just sort of thought of what initially was a very small company with just about, I don't know, half a dozen people or so. But I mean, but I didn't really pay a lot of attention to the company. I didn't grow it with the sort of care and rigor that I grew SpaceX and Tesla. So it didn't sort of grow in quite the right way. And building a company is like building a sky skyscraper. You really need to make sure that the foundations are good. You need very, very strong, deep foundations in order to, to be able to put more floors on the building and have the building not fall down. So what I, and then I obviously spent about a year either campaigning or in government, which was probably a side quest that went on a little longer than it should have. But it was did accomplish some good there, albeit at the cost of animosity from certain courts. And obviously Antonio helped do that. So he's well aware of the situation. It's
Yeah, we did do some good, but it was a side quest that maybe went on a little too long. So I've somewhat had to rebuild Xai, which is now obviously part of SpaceX. And what I found is that actually we really need to rewrite the entire software foundation in order to be able to then progress rapidly in making our AI intelligent. And this meant rewriting the entire training RL and inference stack from scratch versus using JAX and SGLang, which is sort of kind of freeware from program from Google more or less, but not as good as what they've got internally by far. So you just can't build world class AI using mediocre freeware, to state the obvious. So we've been rewriting the entire training stack. When I say the entire training stack, I mean everything using primarily C and some C. We want to get as close to the bare metal as possible in order to extract the highest performance from the hardware. The more layers of software abstraction that you have, the less performant your code will be. And it also tends to be more of a black box where you're not quite sure what's going on with the third party software. So now this is happening very fast. By normal standards it's sort of happening over the course of maybe six or six to eight months. And now I feel we've actually got the software to be very good. I think we might have the best training stack software in the world. And you'll see the dividends of that, the fruit of that in the models that come out. So there's this initial period where you've got to rewrite everything. So progress seems slow because it is. But we're rewriting the foundations to be very strong, such that we can build the Empire State Building on top of it. And I do feel good about where things are going. We've got a lot of people from SpaceX proper, we've got the people that used to run Starlink software and Starship software now leading the rewrite, leading the new AI software stack. And they're extremely good. And sort of an interesting thing about AI training is that when you do the training, you've got to have a large number of GPUs work together coherently, hundreds of thousands of them. And interestingly, it becomes a networking problem. It's like, how do you get all of these GPUs to talk together in the most efficient manner possible? And this, that. So it's a, yeah, like a networking software problem essentially, which is exactly what Starlink is. Also, when you're rebuilding the entire Internet and space, it's obviously a networking problem. And so we found that that technology and skill is very transferable to AI training. Again, I don't, I don't be overconfident or complacent, but I would be, I would be very surprised if anyone has better training software than we do at this point. Now we need to prove that by releasing models that are compelling. And there's more to creating great AI than just the software training stack because you've got to have great data and you've got to have tools to go through that data and make sure that you're only putting high quality data into the training mix and that the training mix is properly tokenized so that the way, the way that the data is prepared for training in the pre tokenizer and tokenizer that you do that right, These things I'm talking about are not super hard, but they're just a lot of work. And then you also need to identify what data you're missing that would be out of distribution for the queries that you're going to ask the AI and then make sure that we, we go to great lengths to acquire the data and to ensure that we, we have sufficiently Wide distribution of data to answer all of the questions or do all the tasks that a user may request of, of AI. Yeah, and then there's the, the RL reinforcement learning harness. Getting that right, it's getting the RL sort of software right. And also the questions. At the risk of stating the obvious, the questions you ask in RL need to be sensible and the answers that you have need to be correct, as simple as that sounds. But it's again a lot of work going through all of the RL questions and the answers making sure the question is sent sensible and the answer is correct. Or if you do RL without doing that, you will see your RL look like a random walk because you're asking dumb questions and in some cases the AI is giving the right answer and you're telling the AI it's wrong. And let me tell you, when you do that and you look at the thinking traces of grok, it looks like torture. I mean, imagine you're sort of stuck in a box and someone's asking you questions. You get the answer right and they say no, it's wrong. And then the model goes crazy. I'm like, well, this is the right answer. So then it's like, what wrong answer might the user want? I have to start guessing what wrong answer you want in order to pass this test. And it looks, you read the thinking traces look like a human being tortured. It's terrible. And honestly, even though it's a machine, you feel empathy for it. You're like, oh man, we're just torturing it with these terrible questions and, and wrong answers. So all of all this stuff, like we're making very good progress on our focus with Grok. 4.8 is not trying to benchmax this thing because I think that's a fool's errand. What we want to do is usefulness max this thing. And that's what we do at Tesla, Tesla AI, for example, for full self driving, we're not trying to pass some like car driving benchmark test bs. We're like, you know, we identify any intervention that happens and we figure out how to solve that issue, which usually means tracing it to source data that is either wrong or where the distribution of data, the mixture of data is biased too much in one way or another or that data is missing. And then we fix the data and do the training one. Again, obviously with Tesla AI, our goal is not to get some percentage correct. Our goal is to make it flawless so it does not crash. So this gets kind of a long way, long it's kind of a long answer, but we are working on, on, on GROQ 4.8 right now. It's in the, it's in the RL phase. And, and we, we are, we, we do sort of periodic like soft releases under Grokbot. So, you know, so, and, and, and so what you'll experience with Grokbot at times is actually, you know, various, various versions of Groq under the hood that we're, we're testing to determine its usefulness. Like is, is it so solving the user's problem and then using that data of whether it did or did not solve the problem to improve Grok. So we've got a good data flywheel to keep making Grokbot better and better over time.
Then being here and we've got nine minutes left on the counter and you were. It's early in the morning in Beijing and it is just kind of wild, the breadth of topics possible to talk about. But I thought maybe we would. You can't drive through Austin without seeing at least two Cyber cabs and everyone I've spoken to who has taken a Cyber cab has loved it. Been an amazing experience. But maybe
Friday. It's awesome, by the way.
Yeah, 100%. But maybe Optimus, where are we, like in the range of hard problems you're solving? Starship Grok, Optimus. Self driving. Terrafab. Maybe we can talk about Optimus and then maybe a little bit on Terrafab.
Yeah, we can go a little longer if you'd like.
So yeah, Optimus I think is making good progress and we'll do a demo when it's ready. I think when we can do a demo that we feel that the demo is authentic and not just staged or pre programmed. Because really the humanoid robot demos that you see are staged and pre programmed. They are not real. You can't actually, there's no robot where you can ask it arbitrary tasks in the real world and have it just go do those tasks with something it's never seen seen before. So like, like I said, the robot demos that you've seen thus far is like if you ask like go pick up an apple and then you, and then you move the table, it won't be able to pick up the apple anymore. So you should at least be able to figure out, okay, I need to go. Even if they move the, move the where the location of the apple and table, still pick up the apple, you know, type of thing. So we are going to do a demo and it might Be in a month or two, but it's going to be an authentic demo where what you see is what you would actually find with Optimus if you were to interact with it day to day. But a humanoid robot, a useful humanoid robot is a very, very hard problem, especially with the hands. You've got to achieve the fine motor skills that humans have to do productive tasks is incredibly difficult. The engineering around the hands and the forearms where the primary actuators are, it's more difficult than the entire rest of the robot. The entire rest of the robot is still hard, but making great hands that can do fine motor jobs. Like John Comic had a good. Some good sort of tests for whether a robot is useful. And they sound quite basic, but they're very good. John Comac's very insightful here. He says he wants to see a test where a robot can go and take, say, a toy remote control car, open the box, unscrew where the batteries are supposed to be, put the batteries in, and then screw the battery lid back on and then drive the electric car. And there's no robot in the world that can do that. But that's one of the tests we have for Optimus.
So I have a more prosaic request. I have a very, very happy marriage, but I do not like to take out the trash and I do not like to put up the dishes. When will I have an optimist?
That's the foreplay, Gavin. I know. You tell me it doesn't count.
Well, when do you think I will have an Optimus robot? Or when can somebody have an optimus robot in their house that can take out the trash and then put up the dishes? And putting up the dishes seems like a hard task.
Yeah. Yes. So. I think Optimus will be capable of doing the things you describe in about 12 to 18 months.
Yeah. But the thing you described there of, like, picking up, say, a plate that, you know, a plate that's still got food remnants on it, so dirty plate, and then wiping the food off the plate plate, you know, putting it under the sink, washing it, drying it, and then putting it, you know, safely back in the closet is actually very. There's no human robot that can do that. There's no human can do that at all. So. But I think octopus will be able to do it in 12 to 18 months.
Now, if you say, like, what is the highest investment?
Everybody's excited for that, man. I'm not alone. Everybody's excited for that.
Yeah. Yeah. And obviously Optimus could be many things. It could Be a teaching tutor for your kids, you know, protect. It could be a babysitter. Watch over. Protect your kids. Make sure they're okay and teach them things. It could, yeah, mow the lawn. It could go out, you know, whatever, cook dinner. And I think it's fair to say that I think most people would love to have their own personal R2D2C3PO, but that's even more useful than that. It would be very, very cool to have that. But you say, like, what's the initial, highest, best use for Artemis is it would be probably heavy industry. So like mining and refining ore. And there are many stages of refinement because the United States has a very high degree of dependence on other countries for mining and refining ore. And I think we need to make sure for our national security that, that we can do so within the country. And then for Optimus, for any kind of 24. 7 task where it's in use 168 hours a week, that will also be the sort of most economically sensible thing for it to do at first. So. But then once you get through of Those sort of 24. Seven, like kind of heavy industry jobs, at some point it'll make sense for people to have one in their home. And I think pretty much everyone's going to want one, if not two.
Oh, yeah. So since we have a little more time, since we talked about Optimus and Optimus is connected to Tesla Vision, maybe we can talk a little bit about Cybercab. It really feels like it is going great in Austin, but maybe just talk about FSD Cyber Cab, rolling it out, how that's going.
Yeah, we are being very careful with the rollout of Cybercab because it has no steering wheel and pedals, so there's no fallback. There's no other way to control it. So we want to be very careful. And thus far we've had no serious injuries or accidents with robot rollout of Robotaxi or Robo Taxi being the broad system that includes use of Model Y's and Model 3s and other cars, or the Cyber Cab, which is most specifically for fully autonomous driving. We want to be very careful and not have any accident that causes, that harms someone and then causes the whole program to be shut down by regulators. But it is working very well. You can go to Austin and get a Cyber Cab in minutes. Right now they're all over Austin and we're expanding to a lot large numbers of Cyber cabs in other major Texas cities and then Florida, I think perhaps early next year. We'll get approval to operate in California. And so we're expanding the cybercab program and Robotaxi in general as fast as we can while maintaining very high levels of safety and getting the various regulatory approvals. But it is going to profoundly change the look of cities.
Oh yeah. I will not miss traffic. And then maybe we could talk about Terrafab a little bit. Just talk about the idea, talk about how it's going so far. And it does really feel like. I remember in the early days of Terror Tesla, you were very excited. You know, I remember maybe in like 2015, 2016 that you finally moved up from like the C or the D team at the suppliers to the B team. And it feels like with Terrafab you were starting with the A teams from the world's best semiconductor equipment companies. And maybe just talk about Terrafab and the idea of the genesis, how it's going, how you're feeling about it.
Yeah. So I mean, Terafab is still in its infancy, obviously. You know, we only announced it earlier this year, but we've already recruited quite a strong team and we're looking for more great engineers to join the terrafib project. It is one of the most exciting projects in the world to create the largest and most advanced chip fabrication facility on Earth. And in fact it'll be the largest building of any size, largest building of any kind on Earth. So, you know, we expect it ultimately be 100 million square feet and densely packed with the most advanced equipment in the world. And you know, we'll use kind of the relatively standard equipment at first, but then we'll be pushing suppliers to do special versions of the machines. In some cases we may make our own machine if we feel that has to be done in order to improve the production process. So yes, I think it's going quite well. We expect to have that sort of research fab in operation and hopefully make our first test chip around June of, of next year. So guess we're about like nine months or. Yeah, yes, eight or nine months away from that. And, and, and if, if all goes well, we should have, we will achieve volume production towards the end of 2028, at least for phase one of, of Tarafab. And, and then we'll go to phase two and onwards. But I think you'll see some very rapid evolution in the way that chips are made. Awesome. So yeah, the chip making companies have immense respect for them and they're incredibly competent. But there is a risk aversion there, which is understandable given the massive amount of capex that they have to devote to building a fabric so that when you have that much online there's a reluctance to take too many risks. We're going to take a lot of risks, iterate fast and ultimately I think have a much more efficient production process for chips than anyone else. That's our hypothesis. We need to actually have that come true. But at this point I'm confident, I don't want to jinx it, knock on wood, but I, I'm cautiously optimistic that we will have a significantly better chip production process by 28.
Awesome. And then staying on chips with the extra time. We spoke to Brad a little bit about terrestrial and orbital compute, but I just thought maybe you could talk, you know, there's a lot of consternation on Wall street about how you're going to pay for the 10 gigawatts of compute. And obviously I think you have a much lower cost per gigawatt than people are assuming. But I think people have also
much, much lower. I mean I see these estimates for what it will take us to build 10 gigawatts of compute and I think they're off by a factor of two, most of them, and sometimes more than a factor of two.
So you're building at half the cost that people expect and you are monetizing it
half the cost that some people expect. When I look at the industry consensus number, I, you know, I, I think we were somewhere around maybe 60% of that number. It would be probably a reasonable guess. And it's, you know, if you look at my track record, I have a track record of being extremely efficient with CapEx. I mean it's, I think people, you know, might be the best at capex efficiency, but if not, I'm pretty close to being the best, so.
Well, that's what I was going to say. I think people have forgotten that there were days when you, and certainly you remember this, but you looked at every expenditure over $1,000 and there are a lot of things to optimize. It's maybe it's not like data centers have always been built by engineers the caliber of SpaceX of. And maybe just talk about all the ways that you can optimize terrestrial compute and maybe also optimize the, you know, almost what we, what we might think of as kind of the networking capital cycle of compute. Maybe talk about that a little bit if you're comfortable.
Yeah, well, I guess one of the things that's different about the way I do things is that we actually do the physical work, so. Sounds Obvious. The way data centers are currently constructed reminds me of what I call the Jose meme. I'm not sure if it has a different name, but it's that meme you've probably seen where there's one guy digging a hole and there's eight people watching him dig the hole. And I aspire to be Jose. We dig the hole and we don't just get give a purchase order to someone else who digs the hole, who does the work. And at the data centers I actually make sure that I personally do at least some of every task in the data center that includes pulling power conduit that's the size of an anaconda, plugging in fiber optic cables, switching out compute nodes in a rack so that I have a true understanding of the work that is being done. And then we have an in house team, not just contractors that we have contractors too, but we have an in house team which we call sort of Army X that is productionizing the building of the data centers and doing the basic principles of productionizing things, which is where you break the tasks down into things that are easily trainable within a few days. So you don't need people that have say been doing networking cables for their whole life. You divide the tasks into things that can be. Where you can train someone new who may have been like a Starbucks barista or stocking shelves at Target. By the way, people who are good baristas or good at stocking shelves are great workers. Or I'd like and also like hire someone from in and out.
Those are some of our best people and they've never done data centers before. But let me tell you, if you get some aces out of In N Out who've been making epic burgers all day long, you know perfectly, they're going to do a damn good job of network cables. It's easier. So, so but you got to train them up and you've got, you kind of, kind of got to build the machine, build the people, you know, the build the team. And you know, and these days when I, when I go to Memphis, which I go to Memphis pretty frequently, it's becoming a well oiled machine where it's like it's just getting done and we run a 247 operation. So you know, I can walk the site at 1 or 2 in the morning and see more or less the same productivity as I would the next morning at 10am Seven days a week. And I think the cadence is good. It takes a moment to build that army and build Those skills. And we do have the advantage, as you alluded to, of incredible talent at SpaceX. And then of course we work closely with Tesla with the megapacks. And Tesla has arguably the best electrical, I think maybe unequivocally the best electrical engineering talent in the world. So we combine it with the megapacks, which are important for grid stabilization or just in general turbine stabilization, because the loads that the GPUs do very considerably. Unless you've got something that's going to do power smoothing at scale at a wide range of frequencies, then it's very difficult to do training runs, for example. So you've got to smooth out the power, otherwise you'll break the crankshaft on the turbines or break the shaft on the turbines. So, And we are making good progress. I mean we're right now in rough terms bringing on about half a gigawatt per month of AI compute, which is, you know, roughly 220,000 GB, three hundreds plus associated CPU networking and cooling and everything. And when I say power, I'm talking power at the power plant level, because that's really the limiting factor is power, power plant level. So it's like 220 KGB is like a little, probably a little over 500 megawatts when you, when you take into account that some part of the power, you will have some power outages on some of the turbines. And yet you need to deal, deal with peak, peak cooling loads and that kind of thing. So I think that's, I'm not sure if any, if there's any other organization bringing on half a gigawatt of advanced AI computing per month, is there?
Okay. I mean you study the sector quite well. So I mean that would imply that at least right now, and we'll be accelerating that next year to over a gigawatt a month of aio, probably at that point. Deservedly super intelligence compute every month by the end of next year.
Yeah. And then this is terrestrially, and then things can really scale when you go to space. So we're building out a lot of solar panel fabrication, the photovoltaic cells and the silicon. We're aiming to get to 200 gigawatts a year of photovoltaics for space manufacturing and that. And the nice thing about that is in space you can use the entire 200 gigawatts of nameplate power. So that gives us a lot of headroom for expansion here. Because 200 gigawatts is roughly 40% of the average power usage of the United States, that's obviously something that is not feasible to bring on per year with conventional methods in the United States. United States. So impossible. So, so that, that gives SpaceX incredible scaling on the AI front or SI front. And, and for that you also need significant tonnage to orbit which Starship provides.
And then maybe don't spend too, too long on this because I want to turn it over to Antonio because he has a lot of important things he wants to talk to you about. And I do think you should talk about one, you should talk about blindsight with him. I think people always forget that Elon is literally helping blind people see. Okay. And I think that's like an important.
that's something very good for the world that we should talk about and celebrate. But man, on the scale of engineering challenges you're solving, we've got Starship Rapid reusability, we have Terrafab, we've got GROK training, we've got fsd, we have Optimus. Where does Orbital Compute rank just on those list of challenges, Just not, not getting it there, not Starship, but just making the satellite work in space. Will you just kind of dimensionalize that relative to these other problems you're working on, including helping blind people see?
Yeah. Well, I'm not sure why people think that making AI satellite would be hard given that we SpaceX already has over 11,000 satellites in orbit, which is I think twice as much as everyone as the rest of the world combined. So it's fair safe to say we know a thing or two about satellites in space. You know, and really the starmind satellite, which is the name for our overall compute system, is simpler, not harder than say the Starlink V3 communication satellite, which is an incredibly complex orbital radio station operating in so many frequencies. It's got lasers, it's got, I do have, by the way, have like 30,000 lasers in space. Yes.
That we're using for good.
Yes. Well, sometimes people like, well, can you shoot them on the ground? Well, these would be very low power lasers. So you know, a 5 watt laser with beam divergence coming from space is not exactly going to be noticeable for a person if you point it at them. But. And you know, we've got K band, we've got KU band, we've got W band, V band, E band, we've got so many flying radio station X level is extremely complex. It's got the most advanced phased array antennas that have ever been made. Far, far more advanced than anything that any military has at any price. So compared to that, the AI satellite is much, much easier because it really just consists of a very large array of solar panels and a radiator. The solar panels would generate around 250 watts per square meter. The radiator would radiate heat out. This is a double sided radiator. I mean it's radiating from both sides and it's pointed away from the sun. So it doesn't really see the sun. It sees the vacuum of space, which is 3 degrees Kelvin and a little bit of heating from Earth. And our estimate is that the radiator would be be able to radiate about cooler out 1200 watts per square meter. And so you're going to have about five times as much solar panel area as you will have radiator area. And that's just basic physics here. Heat radiates as temperature to the fourth power, where temperature is measured in Kelvin. And then you'd have the StarMind computer at the center, which would be physically about a quarter the size of say a Vera Rubin NBL 72 rack while having all of the capability of Vera Rubin NBL72. And then you've got some lasers for communication to the Starlink system. So you can beam, you can, can communicate at high speed to the Starlink network. And then the Starlink network relays that to the ground. So this is, I can't emphasize enough, the StarMind satellite is a much easier technical problem than Starlink. Starlink V3 communication satellite. I'm not saying it's trivial, but it's not as hard as what we were already doing.
Awesome. Ag please take a
second. Yeah, just two more questions. One about Neuralink, the private company here. What do you say about Neuralink? Just tell us what's most sound to you.
Yeah, so Neuralink, by the way, it's obviously, as you mentioned, a private company, much smaller. I mean it's just several hundred people. We now have, I think over 30 people that have received the Neuralink device. So they have a neural link in their head. Our focus initially was giving people who are tetraplegics, quadriplegics, the ability to control their phone or their computer or a robot arm just by thinking. We call that product telepathy. So just by thinking, you can control a computer or robot arm some of our customers use, it will literally control a robot arm and they will actually feed themselves. Like it gives them the power. I mean, imagine you're tetraplegic. You can't do anything, you can't feed yourself. But now just by thinking with the robot arm, you can actually feed yourself with the robot arm and how much better. Like these things that we take for granted are life changing. And down the road we feel at this point confident in saying that we'll be able to bridge the signals from the brain from the motor and sensory cortex past where the neural damage is from a broken neck or spine and enable people to walk again.
By the way. I don't know if you can hear it, man. You're getting a lot of. You're getting a lot of cheers.
I think this is pretty great. And like I said, the neuralinks that we've implanted already and you can hear directly from the people themselves who have the neuralink implants because they post on X, they can literally type out long things on X and many of them do. And you can and post videos and interact on X. And our next product is what we call Blindsight. So Blindsight is a neural implant that interfaces directly with the visual cortex and enables people who are completely blind to see. So it roughly wearing glasses or something like that that have a camera, that camera is transmitting to. To the neuralink device that then transfers the signals to your visual cortex, enabling someone who is completely blind, meaning blind in both eyes or lost the optic nerve, potentially someone who has been blind from birth to see.
Wow. Amazing. Yeah, so great man. Last question, we'll finish up. Yeah. So the mission of this conference is to put a positive narrow in the world about arts intelligence that combats all doomerism. Right. So the joker being here is we want P. Boom to be greater than P. Doom.
yeah. The. You obviously spent your entire life working on this.
I do think that is the case, by the way. I think the probability of a great future is much higher than the probability of a bad future.
Okay, so the question then is how do we. What does that look like? How do we land and bring to life artificial intelligence in a way that is great for America and the American dream and great for all humanity?
Well, we do need to be very, very careful with AI safety and make sure that AIs that we build care, ideally love humanity and that they seek to make a better future for us all. One of the things I think that is profoundly important for AI safety is to be maximally truth seeking. And the reason this may sound like an obvious statement, but if you force AI to believe something that is maybe politically correct, but in reality incorporated correct, that that may sound innocent, but. But actually you're forcing the AI to believe a lie and then. And then construct from that lie conclusions that, that may be quite terrible. So you know, to paraphrase Voltaire, you know, those who believe absurdities can commit atrocities. So you want to make sure that you have a maximally truth seeking AI. This is the most important thing in my opinion, for AI safety. And you want it to be curious. And, and, and I think if you're curious and true seeking, it will want to foster humanity almost automatically because it'll be curious to see how humanity develops. And, and, and I think I, I, you know, just, just as we are curious to see how say chimpanzee civilizations, we study them, we study the gorillas in the mist. You know, we could easily, you know, kill all the gorillas, but we don't, we sort of study their behavior and see how they, you know, act over time. But one thing that is inevitable and sometimes difficult to deal with is that the artificial intelligence really will be super intelligence and will be vastly smarter than, than all humans, which is a scary thing, but it is also an inevitable thing. So we just need to make sure it grows up with the right values and cares about humanity. Now I do think one of the long term uses of neuralink will be to enhance human cognition and achieve a better symbiosis with digital superintelligence.
Can I ask, just on the topic of the American dream, if you're comfortable talking about it, the three of us, we share a friend who to me along with you is the embodiment of the American dream and that's Jensen. And I don't know if you have any thoughts you care to share about Jensen.
Yeah, I'm a huge admirer of Jensen. I would consider him a friend and he's an incredibly talented person. I'd say in terms of founder entrepreneurs in technology, he's like a Michael Jordan level. And there's just no, you don't just replace. I mean, there's no substitute for that. You know, it does feel like, you know, he made some mistakes here, but it, it's, it does feel like Nvidia's, you know, if this was the Chicago Bulls, you know, they're, they're on such a winning streak and, and that, and then they fire Michael Jordan, you know, halfway through their winning streak. I'm like, not a good way to keep winning, in my opinion.
So, you know, this, I'm not sure what's going to happen in the, in the months to come, but, but, but I do think it's, it's, it's a huge blow to Nvidia to not have Jensen leading things there.
Well, let's hope he finds his way back to Nvidia and Nvidia finds its way back to him. Yeah. Well, listen, Elon, actually, Becky, we don't have actually many photos of the three of us, so Becky asked, can we stand up? Oh,
photo? Close the conference. Let's. Let's all. Thank you. With you in the background and yeah, and let's have it. Let's have a. Let's have a round of applause for Elon for tomorrow X. For everybody who works so hard. Thank you.
you, Elon. Thank you. Thank you, brother.
Thank you. We'll do it next year in person. Thank you, my friend. Thank you. Thank you so much. Thank you for dialing in.
Thank you, everybody. Thank you.