Elon Musk 谈 DOGE、Optimus、Starlink Smartphones、与 AI 共进化，以及西方为何正在崩塌

How do you have time? I never understand you.

Yeah, I do work a lot. All right. Good.

All right. Where are you?

Palo Alto.

You're in Palo Alto and not Washington, D.C.?

I'm at Tesla Global Engineering Headquarters in Palo Alto.

So, no more Washington, D.C.? You're back at work. You're focused?

Yeah. I haven't been to D.C. since May.

Okay. That was a hell of a side quest. Any lessons from your time in Washington, D.C.?

The government is basically unfixable.

I support David's noble efforts, and it's good to have talented people in the administration, but at the end of the day, if you look at our national debt, which is insanely high, the interest payments exceed the U.S. Defense Department—I guess, sorry, War Department—budget. Basically, if AI and robots don't solve our national debt, we're toast.

Which is a great segue. Optimus is, I think, going to be the greatest product in the history of humanity. What's the progress like, and how many of your cycles are going specifically to Optimus? What's the timeline? I think you're on version 3, maybe 4. Tell us everything.

Everything would take a long time.

We've got time.

We're finalizing the design of Optimus Version 3, and that really is going to be a very remarkable robot. It will have essentially the manual dexterity of a human, meaning a very complex hand, and an AI mind that can navigate and comprehend reality. It will also be made in very high volume.

Those are the 3 things that are missing. If you see any other robotics company, they're missing those 3 things. Those are the 3 really hard things.

At this point, it might be more of my mental cycles than anything else—more than any other single thing—going to Optimus. That means solving for real-world AI, all of the electromechanical issues of Optimus, and the supply-chain and production challenges. There is no supply chain that exists for humanoid robots, so we have to recreate it from scratch, which requires doing a lot of vertical integration.

None of the actuators in Optimus are available from an existing supply chain. I think it is accurate to say that, if successful, Optimus will be the biggest product ever.

And the cost of it at scale: $20,000, $30,000, $40,000 a robot? What do you think the first wave of them will cost? When will we be able to buy one to work on the ranch?

I think the marginal cost of production, once you hit 1 million units per year, is probably around the $20,000 range. It depends on how much you spend on the AI chip in the robot, and you need to achieve a lot of efficiencies in the actuators.

There are 26 actuators per arm—26 electric motors, gearboxes, and power electronics. The AI chip will be pretty expensive. That might be $5,000 or $6,000 of the bill of materials, maybe more.

At a volume of 1 million units a year, I think the production cost is probably on the order of $20,000, maybe $25,000—something like that. The price will be a function of demand.

Elon, can you explain to everybody why the hand is so important to get right, why the actuator design is so unique, why it's so difficult, why nobody makes it, and why you have to start there almost to build the rest of the robot properly?

Well, it turns out the human hands have evolved to be an incredibly sophisticated machine. Your hand is actually a remarkable thing. Look closely at your hands and think of all the things you can do with them, which is a lot.

I can think of many things.

Yeah, I was just thinking about something.

Your hands are a very versatile instrument. You can give someone a high five.

Very versatile. You can swing a baseball bat, thread a needle, play the piano or violin, and disassemble or assemble a car. The hands are incredibly versatile instruments.

Most of the muscles of the hand are actually in the forearm. Your hand is kind of like a puppet. The muscles are coming from the forearm and pulling the tendons, and human tendon evolution is incredibly good.

You've got this web of tendons. I think the human hand has something like, depending on how you count it, 27 or 28 degrees of freedom in the hand. It's amazing. In order to create a robot that can be a generalized humanoid, you must solve the hand problem.

Yeah, we had—it's got hands, needs hands. Is it literally like when you were first building Tesla, where the supply chain doesn't exist and now you have to go out and find folks to work with, build all this vertical integration, and get support? Is it just nowhere to be found?

Yes. We could not actually buy the actuators for any amount of money. They simply didn't exist. Even though there are, I don't know, 10,000 or 20,000 electric motors out there of various sizes and shapes, we've had to design every electric motor, gearbox, and piece of controlling electronics from scratch—basically from physics first principles.

The good news is you've got a lot of experience with factories over the last couple of decades. How challenging is this versus Cybertruck, Model Y, Model X, and the Gigafactory? The Fabergé egg known as the Model X.

Right. It's harder than any of those things.

Okay. Much harder? Significantly? Yeah, Starship?

Yes. Well, more—no, Starship's harder.

Okay. So, somewhere between a Model X and a Starship?

Yeah.

What's harder, the hardware or the software?

Right now, we're struggling with the final design of the hardware. Like I said, it's primarily the hand. Not to dismiss the rest of the robot—the rest of it is also important—but the hands, inclusive of the forearm, are the majority of the engineering difficulty of the entire robot.

Let's assume you get past the hardware challenges. How much do you get for free based on all the progress happening with LLMs? Will consumers just be able to interact with this, talk to the robot, ask it to do things, and have it understand?

Oh, yeah. Yeah, no problem.

You're spending a lot of time with AI, I noticed online.

Not that long. Maybe I went a little over the top with Bunny Girl in Grok Imagine.

Well, in all seriousness, those characters and these robots—it seems like maybe you could get the embodiment of Ani, I suppose.

Why the human form factor, Elon? You could make something that's maybe better than a human or simpler than a human to do specific tasks, and maybe better than a human to do more things than a human can do. How do you decide to make it just like a human?

If you wanted to do all the things that a human can do, it turns out you need a humanoid robot. If you want to do just a subset, that's much easier.

It turns out humans evolved to the shape and capabilities that we have for good reasons. There is value to having 5—4 fingers and a thumb. Even the pinky is quite useful. Toes are more important for walking, but the fingers—

Humans have also designed the world. We designed it for us. So, if you can make a humanoid robot, it'll be immediately backward-compatible with what we've built the world for.

Precisely.

Elon, there's another part of the robot. There's the LLMs, there's the actuation in the hands, but there's also the silicon that runs it. There was Dojo, and I think you posted on X about AI5 and AI6. It just seemed like you were incredibly excited about the direction in which the silicon layer was also going.

Can you tell us about that and what it is? What are we building here? What is being built? Is it a complement to everything that exists in the world, or is it a potential long-term competitor? What is it?

At Tesla, we basically have 2 different chip programs: Dojo on the training side, and what we call AI4, which is our inference chip. AI4 is currently shipping in all vehicles, and we're finalizing the design of AI5, which will be an immense jump from AI4. By some metrics, the improvement in AI5 will be 40 times better than AI4.

Wow. So, 40 times?

40 times. This is because we work so closely, at a very fine-grained level, on the AI software and the AI hardware. We know exactly where the limiting factors are, so effectively, the AI hardware and software teams are co-designing the chip.

So, a 40-times improvement in the silicon—I think then, as everybody here in the audience experiences it, is that just an almost order-of-magnitude increase in the quality of FST and the safety that you experience as a Tesla driver, and the quality of the robot? Where does it all manifest when you bring it up and actually get it into production?

To be precise, the 40 times is compared to the worst limitation on AI4, which is running the softmax operation.

Yeah.

We currently have to run softmax in around 40 steps in emulation mode, whereas that will just be done in a few steps natively in AI5. The AI5 chip will also easily handle mixed-precision models, so you won't have to manage that—it will dynamically handle mixed precision. There's a bunch of technical stuff that AI5 will do a lot better.

In terms of nominal raw compute, it's 8 times more compute, about 9 times more memory, and roughly 5 times more memory bandwidth. Because we're addressing some core limitations in AI4, you multiply that 8-times compute improvement by another 5-times improvement from optimizing, at a very fine-grained silicon level, things that are currently suboptimal in AI4. That's where you get the 40-times improvement.

You had—I'll keep going.

Keep going, keep going. Now, that said, I am confident that the current AI4 chips in the cars will achieve self-driving safety that is at least 2 to 3 times that of a human, and maybe even 10 times. The software that will enable that is coming out over the next few months.

Version 14 will be the biggest upgrade in Tesla software since version 12. We're increasing the parameter count by an order of magnitude, and there's a lot of reinforcement learning being used. You can think of AI as a way of compressing reality, and some of those compression steps were too lossy. We addressed the lossiness in those compression steps.

These are all software updates that will go out over the air. Your car is going to feel like it is sentient by the end of the year.

It feels that way already, to be honest. I saw in the trades that you spent about $17 billion on some spectrum.

Yeah, some couch change.

To enable your satellites and the Starlink network to connect directly with phones, what will that look like in a year or 2? Are we going to drop our Verizon account and just expand our Starlink account?

Thank you.

We're kind of hoping, because Verizon kind of sucks. How many of you want a Starlink phone? Who wants a Starlink phone? Is it technically possible? I know you can't see it, but it's everyone.

Yeah, all right, cool. This is a long-term thing. It will allow SpaceX to deliver high-bandwidth connectivity directly from the satellites to the phones. But there are hardware changes that need to happen in the phone.

Since these frequencies are not supported in current phones, the chipset has to be modified to add these frequencies, and that probably is a 2-year time frame. The phones that are able to use the acquired spectrum will probably start shipping in around 2 years.

We also need to build the satellites that are going to communicate on those frequencies. In parallel, we're building the satellites and working with the handset makers to add these frequencies to the phones. Then the satellites and the phones will handshake very well to achieve high-bandwidth connectivity.

The net effect is that you should be able to watch videos anywhere on your phone.

Wow. And would these frequencies work indoors, inside buildings, like your phone currently does?

If you're in a building with a thick metal roof, then no.

No, the same types of—

Yeah, normal homes, yes.

Yes. Elon, is your vision for this that instead of having an AT&T account and then roaming when you're in the UK or India, we could have 1 direct deal with Starlink that works all over the world eventually? Not today, but at some point? Is that the end goal—that basically we don't need a regional carrier, we have a global carrier, and that would be you?

That would be 1 of the options. To be clear, we're not going to put the other carriers out of business. They're still going to be around because they own a lot of spectrum.

But yes, you should be able to have Starlink, like you have AT&T, T-Mobile, Verizon, or whatever. You could have an account with Starlink that works with your Starlink antenna at home, free Wi-Fi as well as on your phone. It would be a comprehensive solution for high bandwidth at home and for high-bandwidth direct-to-cell connectivity.

Could you buy some carriers to have more spectrum? Maybe you could buy Verizon.

Not out of the question. I suppose that may happen.

Let's talk about Starship. You just had what appeared to be a phenomenal launch. How close is it to being predictable and ready to go in a commercial setting?

I think we'll recover the ship next year. We've got 1 more launch of the Starlink Version 2 stack. There's only 1 booster and ship left in the Version 2 design, and thereafter it's Version 3, which is a gigantic upgrade because it has Raptor 3. Pretty much everything changes on the rocket with Version 3.

Version 3 might have some initial teething pains because it's such a radical redesign, but it's capable of over 100 tons to orbit, fully reusable. Unless we have some very major setbacks, I think SpaceX will demonstrate full reusability next year, catching both the booster and the ship and being able to deliver over 100 tons to a useful orbit.

What does the best rocket in the world do now in terms of tonnage to space?

In terms of commercial rockets, there's Falcon Heavy, which with side-booster reuse will do about 40 tons.

Yeah. So this is 5 times bigger.

Well, 2.5 times bigger, but Starship would be fully reusable.

Got it. So everything comes back.

Elon, after the explosion that happened with the failed launch—

Sorry—

Which failed—

Oh, the more recent one? The more recent Starship with the big boom?

The big boom on the base. There were a lot of proclamations that there were going to be environmental and FAA reviews and all these other sorts of questions. The recovery back to the launchpad was incredibly fast. How did you get back so fast—not just technically and work-wise, but regulatory-clearance-wise? They said there were going to be all these questions and reviews. How did you manage that?

There were a lot of questions and reviews. We got through them all. Credit to the SpaceX team: They worked incredibly hard, got the next ship and booster tested and on the pad, and flew them. Huge credit to the SpaceX team. I'm very proud of them for—

Doing such a great job recovering.

Creating a fully reusable orbital rocket is 1 of the hottest engineering problems ever, and certainly a candidate for the most difficult engineering project ever. It's on the podium, at least.

That has been the goal of SpaceX from the beginning, from 2002. Here we are, 23 years later. It's a long journey, with a super-talented group—by far, I think, the most talented group of rocket engineers that has ever been assembled. Finally, next year, I think we'll be able to achieve full reusability.

Elon, what are the big technical blockers that you're focused on between now and full reusability? Are there some showstoppers that you're literally obsessing over trying to figure out, or is it more about getting through a laundry list of your learnings and integrating them into the next launch?

For full reusability of the ship, there's still a lot of work that remains on the heat shield.

So, no one’s ever made a fully reusable orbital heat shield. The shuttle heat shield had to go through 9 months of repair after every flight.

Right?

So, no one has ever made a fully reusable orbital heat shield.

And is that a materials science problem, an engineering problem, or both?

Yeah, it’s a materials science and engineering problem. But we’re really looking at the fundamental physics here—again, physics first principles—and trying to figure out how we make something that can withstand the heat, is very light, and doesn’t transmit the heat to the—

Yeah.

—primary structure. And then as you descend, if you hit some rain, the tiles don’t dissolve in rain. There are a lot of different issues, and then you really need to know that these tiles are working. You can’t go through this laborious inspection. It really needs to be that these tens of thousands of tiles all work and don’t need to be refurbished or checked one by one, as was the case with the shuttle.

Can we maybe switch now? We’ve talked about Tesla, then you go to SpaceX. Now I’d like to ask you some questions about Grok and xAI. Do you want to just give us an update? I think you kind of talked about where the next Grok model is, and you said something incredible. I still don’t think people really understand it, which is that there’s going to be a next training run where you expect not to start from the common web and Common Crawl, but where you expect an enormous amount of synthetic data. Just tell us about how the evolution of Grok is going, and this innovation and why it’s so important.

Yeah. We’re using a lot of inference compute and reasoning to look at all of the source data, which is really the corpus of human knowledge, and then thinking about each piece of information, adding what’s missing, correcting mistakes, and removing falsehoods from that training data. It’s like if you take Wikipedia as an example, but this really applies to books, PDFs, websites—every form of information.

Grok is using heavy amounts of inference compute to look at, for example, a Wikipedia page and say, “What is true, partially true or false, or missing in this page? Now rewrite the page to correct it, remove the falsehoods, correct the half-truths, and add the missing context.”

Well, Elon, by the way, could you just publish that? Could we create a Grokipedia?

Yeah, especially for our bio pages, which are a disaster.

Wikipedia is so biased, and it’s a constant war. If something gets corrected, 5 minutes later, there’ll be an army of people trying to change it. I mean, it’s become hyper-partisan, and there are activists all over it. So, if you do fix Wikipedia, for example, as a source of truth, it’d be great to publish that just so the world has it.

All right, I’ll talk to the team about that—like Grokipedia or whatever. Here’s the Grokipedia version.

It’d be interesting, yeah, to just have it out there for a few minutes. In terms of training Grok 5, you’re scaling up your supercluster, Colossus, in Memphis.

Yeah, we can have a second one.

Yeah. Could you give us an update on that? And then also, as part of that, where are we in the scaling laws? If you scale up to a bigger cluster, do you get a more powerful AI model? Is there a point of diminishing returns? How much more compute—if you throw twice as much compute at it, do you get a 10% better model, or do you get a 100% better model? Is it log-linear? How much more juice is left in scaling hardware, do you think?

I think there’s a natural logarithmic function associated with the amount of compute. For argument’s sake, 10× more compute will double the intelligence. Maybe that’s a rough rule of thumb, but that still means that you go from 100 IQ to 200 IQ. That’s still a pretty big deal.

I think we’ll see intelligence continue to scale all the way up to where most of the power of the Sun is harnessed for compute, and then ultimately most of the power of the galaxy—sort of Kardashev Type II, Kardashev Type III-scale compute.

Once you think of artificial intelligence not as a destination that you reach, but really as part of the overall escalation of intelligence that we’re aware of, human intelligence has also scaled as the population has increased and we’ve been able to store more and more information. Human intelligence, because of population declines and low growth rates, is somewhat plateauing and will actually decline. My guess is that we might have AI smarter than any single human at anything as soon as next year.

Wow.

And then probably within 5—say, 2030—AI is smarter than the sum of all humans.

Do you think humans are on the decline because AI is evolving? Do you think there’s this evolution of the ecosystem on Earth that’s underway that we don’t really understand—the structure of what’s going on?

Maybe. Yeah, maybe we implicitly know that it’s coming.

Yeah.

I hope the birth rates turn around. I’m a big proponent of increased birth rates, obviously.

Well, are you doing anything about it?

Yeah, I’m trying to set a good example.

We had a big conversation at this conference that we didn’t expect, which is “suicidal empathy” in the West and this declining birth rate. I noticed you’ve been pretty active about it and open borders.

And open borders is like, “Let the invaders in.” Could all 3 of those be the same thing? It seems like there are a number of symptoms of the West being suicidal.

The most obvious one is that the birth rate is not at a replacement level. If that continues indefinitely, then the West will literally not reproduce enough to replace itself. But there are other things too. The borders were totally opened to the point where Western culture and the social fabric start to come apart, and you see this especially in Europe, where the indigenous cultures of the UK, France, or Germany are starting to potentially be taken over by cultures of people who are brought in and aren’t assimilating. You have crime where—we have this case on social media right now—this young woman, Ire Ina, was just killed in a senseless way on a subway.

Which is horrific enough in and of itself, but then, in addition to that, the elite media, just for whatever reason, refused to cover it, like it didn’t exist. So you have this issue of crime that’s not being addressed or even acknowledged, and no acknowledgment of this. It’s almost like we’re trying to deny the reality of the spiral.

Yeah. So, you have all these data points that seem to suggest that the West is suicidal or doesn’t seem to want to defend itself or propagate itself. Look, I think everyone in this room thinks that life is awesome, right? It’s pretty great, and I think worth living.

Yeah. And when Alex Karp was here earlier today defending the West, that got some of the loudest applause at the conference. So, I guess we probably don’t really understand what’s going on. We don’t really—

What’s your take, Elon? What’s your take on the suicide of the West?

I’m very worried about it.

Yeah.

I’m very worried about it. I think the actions of the West are indistinguishable from suicide. But at least in America, there’s generally a sense of optimism. When’s the last time you talked to someone from Europe who lives in Europe who’s optimistic?

Not for a while.

Decades—like, even one.

It’s rare.

So, I think unless people have a sense of optimism and purpose about the future, suicide might be just what happens. Having a child is an act of optimism about the future. If you’re not optimistic, they’ll be less interested in having kids. So, I think we need to give people a sense of optimism and excitement about the future, and a belief that the future will be better than the past, and they’ll be more interested in having kids.

Did religion play a role in the past, Elon, to make folks feel that way?

Yeah, I think so. Nature abhors a vacuum, and if you take away religion, then I think you get something in its place that’s actually worse than what was there before. I mean, it’s destructive, basically. You get the woke mind virus filling the hole that religion used to have, taking the place of religion. You get these dystopian de facto religions that are very, very self-destructive.

So, I think perhaps some sort of revival of religion—or at least what we need is some coherent philosophy that people can get excited about. For me, it’s a philosophy of curiosity.

I'm curious about the nature of the universe, and I want humanity to be out there exploring the stars. Maybe we'll meet alien civilizations. Maybe, in some cases, we'll see the ruins of a long-dead alien civilization, but they were very strong for 10 million years. You know, the kind of stuff that you see in Star Trek, in a nondystopian science-fiction book, movie, or show.

I have a philosophy of curiosity. I just want to know what's going on. In order to know what's going on, we must have an increase in the scope and scale of consciousness; we must expand consciousness. We must grow humanity and extend humanity in order to comprehend and understand the universe—or even what question we should ask about the answer that is the universe.

Douglas Adams's book The Hitchhiker's Guide to the Galaxy is actually a deep book on philosophy disguised as humor. The point he was trying to make in that book was that the questions are really the hard part. The answer is the universe. The answer is everything you see around you. But one of the questions is one that we don't know to ask.

Yeah.

Now, some of the questions I do know I'd like to know are: Is the Standard Model of physics correct about the origins of the universe? Are we actually 13.8 billion years old? How does the universe end? Does it end in a heat death or in some other way?

A black hole.

We might be.

Elon, can you talk about the whole simulation question? Are we a simulation? Maybe. Where do you think we find the answer first: in AI or in the stars? Because you're pursuing both, obviously.

Yeah. I don't know. I hope more people can get behind a philosophy of curiosity.

Yeah.

I think it's very exciting and inherently optimistic. There's this amazing sense of wonder about the nature of the universe. When you uncover some secret in the universe, that's amazing, and a whole world of understanding is opened up.

We used to not even know where all the continents were. The map would say, “Here be dragons,” and all we knew was that when they sailed in that direction, they didn't come back.

I mean, the moon base—

That's all they knew. I kind of feel like the moon base, or just going to the moon for real this time, would be a big step in the right direction.

You still have the moon planned. What's the status of that? Is that still on the agenda?

Yeah. I think we want to try to reach new heights as a civilization.

Yeah.

I think it's fine to go to the moon, but we should go to the moon in order to establish a lunar base, like a lunar research base. There are parts of the moon that are perhaps older than parts of Earth, and we might understand more about the nature of the universe if we had a science base on the moon.

That would be very cool. Then we obviously want to go beyond the moon to Mars and build a self-sustaining city on Mars. I do think that there is a fork in the road of human destiny. If we can establish a self-sustaining city on Mars, with the key test being whether Mars continues to prosper or dies out if the resupply ships from Earth stop coming for any reason, then at the point at which Mars is able to prosper and grow on its own, the probable lifespan of consciousness is dramatically greater.

We are no longer dependent on everything going right on Earth. There's always some possibility of self-annihilation on Earth with World War II or a supervirus, or a meteor like the extinction event that destroyed the dinosaurs. We know from the fossil record that there have been many mass extinction events. The question I was wondering about is whether civilization—whether the civilizational arc—will continue to ascend such that we can make Mars self-sustaining before the civilizational arc descends.

The window of opportunity to make life multiplanetary exists now for the first time in the 4.5-billion-year history of Earth.

Yeah. Elon, let's assume that we get there and you're there. You'd be the elder statesman. You'd have the moral authority of Mars. How do you run Mars?

I want to emphasize again that it's more important than the form of governance on Mars or who's there in the early days. What really matters is that Mars is self-sustaining, that we are truly a multiplanetary species, and that we've achieved planetary redundancy. We should do everything possible to make sure life on Earth is great, but there's always some risk of an annihilation event on Earth.

Yeah.

Like I said, that could be self-annihilation or some natural disaster. The probable lifespan of consciousness increases dramatically as soon as we are a multiplanetary species, with the key test being whether Mars can survive if the resupply ships stop coming.

The first missions to Mars are not that important. What matters is whether you can get sufficient tonnage to Mars such that Mars can prosper on its own. That means it has to have all of the ingredients of civilization. It's not just that you need to build, for example, a chip factory, or chip fab, on Mars; you need the ability to build.

Do you have a sense of the timescale? Let's assume Starship is at a starting state in 2026. Then there's going to be a bunch of testing. Obviously, there are going to be early tests, and we only have certain launch windows, so there are a bunch of time constraints. Is this a 50-year thing in your mind? Is it a 150-year thing? Is it something that's for our generation or our children's generation? Where do you see that point, if it's optimally possible—if things break our way?

I think it can be done in 30 years.

Wow.

Provided there's an exponential increase in the tonnage to Mars with each successive Mars transfer window, which is every 2 years. Every 2 years, the planets align and you can transfer to Mars.

I think in roughly 15, but maybe as few as 10—10 to 15 or so—Mars transfer windows, if we're seeing exponential increases in the tonnage to Mars with each Mars transfer window, then it should be possible to make Mars self-sustaining in roughly 25 years.

Amazing. That's incredible.