Elon Musk 与 Gwynne Shotwell 谈 AI 风险与同行评审、Starship、Terafab 及 SpaceX/Tesla 合并

Quick question: What’s our latest timeline on getting to Mars?

Getting to Mars? Oh, I think we’ll put people down within the decade.

Welcome Gwynne Shotwell, the president and COO of SpaceX.

She’s instrumental, a really crucial person to the success of SpaceX.

She’s really the glue. She’s the glue to the tornado.

Fail fast, learn fast, make your design robust.

I would be disappointed if we didn’t have a settlement on the moon.

The space and AI firm has officially made history on its $75 billion IPO, debuting with a market value of $1.7 trillion.

The most efficient place to put inference compute is on orbit.

Manage risk, don’t avoid it.

Please welcome Gwynne Shotwell.

Hey.

Thank you so much for coming.

You’re right here next to Chamath—on the couch. Welcome.

Right there. Right in the center.

Thank you.

Thanks for being here.

It’s good to be here.

Nice to see you.

Nice to see you.

Well, we really appreciate you coming out. How’s it going at SpaceX?

We’re bored. There’s nothing going on.

Yeah. Not a lot happening.

I’ve got all the time in the world.

Elon has more time, by the way.

By the way, Gwynne, you’ve been at SpaceX forever. Are you the longest-tenured employee after Elon?

After Elon, yeah. In fact, my 24th anniversary was September 9, last week.

Wow.

Yeah.

Gwynne, can you take us back to the day when you initially got this call and you thought, “Hey, there’s this thing”? What did the arc of the decision look like, and how did you get convinced to underwrite this business? Just walk us through that.

1. How Shotwell Joined SpaceX

Okay, Elon underwrote it, right? I just worked there.

Yeah.

But the story is a little crazy, actually. I was going out to lunch with one of my best friends, who I was working with and who was going to work for Elon. I took him out to lunch, kind of a going-away lunch, and when I dropped him back off, he said, “Oh, come on in and meet Elon.” I’m like, “Ah, he’s busy. We don’t need to do that.” But we did it anyhow, and I met him.

I said, “Hey, I think you need a full-time person to run your business development shop. You’ve got kind of a part-timer. He’s a contractor. It’s not a great look.” He looked at me very thoughtfully and kind of scratched his head. We chatted a little bit and said goodbye.

Then I got back to my office and got a call from Elon’s assistant, Mary Beth. She said, “Elon wants you to apply for the new vice president of business development position. Can you come interview?” I’m like, “Now? No, I have a job. I can’t go now. Tomorrow? Okay, sure.” Then it took me 4 weeks to decide yes. I was being a total idiot.

At the time, what was the scale of that business? What did you think it needed that it didn’t have? It was still very much an R&D—

It was very R&D, but Elon had an investment and he wanted some return. He wanted cash, right?

Yeah.

He wanted customers, and I was doing that work anyhow in the space industry. It turned out I was pretty good at it. So what did it look like? I was the seventh employee to sign my contract. I gave 2 weeks’ notice and took a week off to remodel a bathroom because I knew I would never have time to do it once I started working for Elon.

I came 3 weeks later as employee number 11, and we just got to it. We sold a rocket about 12 months later.

Without a rocket. There wasn’t a rocket.

There wasn’t a rocket, yeah. What did that sales cycle look like? How did you convince somebody that they could actually take that risk and maybe give you the deposit you needed to start the cycle to build the thing?

The timing was great. SpaceX should never have existed, right? If the other launch providers had been great, then a new entrant couldn’t make headway because they were so entrenched. So we should never have made it.

But it turns out customer service wasn’t great, rockets were incredibly expensive, and so there was that piece, right? The other piece was 9/11. At that time, the military was desperately looking for someone who could come in and do rapid launch. They kind of knew where the bad guys were for about 4 hours after the events of 9/11, and they really wanted to be able to get some capability in theater. It took us a while to do it, but we did what we said we were going to do.

So it really was, out of tragedy, this opportunity gets born.

To some extent, yeah.

That’s interesting.

And it wasn’t specific to SpaceX.

No.

It was just the state of the industry at the time.

Then, obviously, the very famous first chapter, where you guys got almost to the brink and needed this last launch to work, and your role started to evolve meaningfully by that point. Can you walk us through the internal dynamics at that business that put you into that position?

2. Shotwell Takes The Helm

It was very startup-y. I was the sales lady. But then when you have sales, you need to manage your customers and you need a finance function. Even though we were little, I think one of the defense contractors at the time said, “They’re a fly on my big toe,” or something. I was like, “Well, I don’t think you should have flies on your toes.”

You still need a government affairs function. You need to defend yourself. We never go on the offensive in Washington. We’re always in defense mode. So I kept taking on more and more.

When we were getting our largest contract at the time, it was the cargo resupply contract, which we call CRS. This was 2008. Elon was super busy with Tesla, quite busy at SpaceX as well. There was a lot going on, and he needed a partner at the time. I think he was trying to pick between me and maybe one of the other guys who had gobbled up more technical scope.

He asked me to do it on our way to the final negotiations for this $1.6 billion contract with NASA, on the runway. He said, “Hey, do you want to be president?”

“Do I?”

Well, no, not initially. For that nanosecond, I thought, “Well, I love my job. I love the guys, my teammates.” Then I thought, “Okay, I’m being an idiot again. Yes, I want the job.”

Fast-forward to where we are today. The scope and the ambition of this business have expanded dramatically. It almost seems like it happened in the last year.

Mm-hmm.

It’s as much an AI business as it is a space business. I think Elon may have said that. I don’t know if it leaked or they said it out loud, about the scale—

It’s backed by revenue.

And by revenue, yeah. How do you manage the culture of these 2 different businesses that now have to sit under this roof and ideally work together and complement each other?

3. SpaceX Enters The AI Era

The team—there’s been a lot of churn at xAI, for sure. SpaceXers have marched in and helped out where there were significant gaps in that arena. We were all really excited to learn about AI. I’m an AI noob.

We have a bunch of folks who were just really excited about it, and I think we all recognize Elon set the stage beautifully, as he does. If you’re not using AI, if you’re not leading AI, you run a significant risk of being irrelevant. There was no way SpaceX was going to be irrelevant, right? It’s too important to the company, and the work we’re doing is too important.

Yeah.

Do you expect future generations of rockets will be increasingly informed, built, and designed—

100%.

By—

By AI.

Yeah.

Yes.

How do you manage the team? You have these incredible—literally, rocket scientists. They’ve trained their whole lives, in a way, and now you have to help them onboard a toolchain and be comfortable with what that means.

I don’t know if you saw that all these mathematicians got really upset when, I think it was OpenAI, solved this very famous equation—

Oh, the Navier–Stokes equation.

The Navier–Stokes equation. These 20 Fields Medalists essentially said, “Hey, you’re ruining the craft.” I don’t think that’s—

They’re taking away the medals.

Yeah.

They’re going to take—

Yeah.

Away, exactly.

Yeah.

That's the career. It's not about making stuff. It's about getting awards.

Yeah.

And now all of the awards are gone.

Yeah. I'd rather just have the world be a better place, a smarter place, us being better informed and solving problems than worrying about—

Make more stuff.

That. Yeah. Make more stuff.

What's the cultural merger been like? You've got xAI. You've got X. I don't know how many people are involved in X itself. SpaceX. Are these differently managed businesses? How do you run management there? Are people pretty fluid among the units? Do you think about them as units, or how do you actually manage this?

I would say we are not fully integrated yet. And then we just did the Cursor situation—

Congratulations on that one.

Yeah, congrats.

What a purchase that was.

Yeah, yeah, yeah.

Well done.

They're great. They're great.

Yeah.

We just closed that, I think, a month ago today, actually. So we're not fully integrated yet, but I think the teams are gelling really well. They're very different verticals, but, again, because of the churn that we had at xAI, so much SpaceX leadership and engineering went into that—basically went in there. We're getting integrated faster than I thought. It's probably not as fast as Elon wants it.

Yeah.

4. The Capital Scaling Challenge

One of the things that strikes me is the scale and the number of capital projects you have to manage. You probably have more capital projects that are more different from one another than any other organization on Earth.

It's the scale of a country.

No, I mean, it has to be.

It's like a country.

I think about this Louisiana space port. You've got Starship, which is the most ambitious aerospace project of all time. You've got Terafab, which I think is one of the most inspiring and ambitious projects of all time in humanity.

Mm-hmm.

Terafab is so incredible in its vision and the scale of what you're—

Yeah.

—envisioning to do there. And then I start doing back-of-the-envelope math. I think I had a couple of beers, and I was talking with someone about Terafab, and I'm like, "Holy shit."

Yeah.

Those numbers get very big very fast.

Very big. Although—

Yeah.

The AI numbers—

Are also—

Like—

Yeah.

I used to be horrified at our expenditures before the xAI acquisition. It's like, "Oh, my God. We're going to spend—

Reset.

—hundreds—

Now you see the compute bill.

—million dollars—

Suddenly Starship seems cheap.

—on an air separation unit."

Yeah.

Now it's like, oh, a hundred—

Yeah.

—billion.

When you look at the schedule of all of this over the next few years, does the capital for that sit on the balance sheet? Is it going to come from cash flows from Starlink, or do you have to go back to the markets to get more capital? How do you think about where the capital comes from? I don't think anyone doubts your capacity to execute and deliver. Some people might, but I think generally the markets will say yes. But where does that capital come from?

It will come from—I would say the answer to the above is yes. I don't think we're going to release any more stock, and, by the way, I don't plan on making any news on this. That's the one downside of a public company.

Yeah.

They told me all the things that I couldn't say.

Can't say.

I'm going to forget one for sure. Maybe we're going to make some news.

Yeah.

But it will be all of the above.

Is Starlink, in your view, kind of the core cash-generation engine for the next couple of years for the business, or is it compute, or is it a mix of all of the above? How do you think about—

Starlink is definitely carrying its own weight, but we have a lot to go, right? Our market penetration on Starlink is 1.5% to 2%, depending on which country you're in. So there's a lot of headroom there. Our revenue compared to what we do for the U.S. military is very small.

Yeah.

So we expect lots of headroom there. But, honestly, right now—and it is a little embarrassing—compute rental is a heck of a business.

And does that persist, do you think, or is it—

We don't see any drop—

—at the end of the—yeah.

Wow.

We see no drop in demand at all.

Well, and this is—

Wow.

—tens of billions of dollars per quarter in build-out, and these are a new customer base for you. Is this the business now? What percentage of the revenue is Elon Web Services at this point?

It's a lot. It's a lot.

Yeah.

I think we'll probably do some announcement quickly or—

Oh, right.

—shortly thereafter.

Public again.

Yeah, public.

Well, let's move on to something less controversial, then. What other companies are you planning on buying?

Good one.

Yes.

You know, we do almost no M&A.

You like electric cars? Have I got a deal for you.

I have a bunch. I have a bunch.

Batteries, electric cars.

Starlink on Optimai.

Well, can we—

I'll get you a good price.

—can we talk about a deal that you did close? You did a great deal with EchoStar.

Yes.

Charlie had a bunch of spectrum that you guys were able to acquire, and I think everybody started to ask the question, "Wow, there's a direct-to-cell business."

Yes.

The natural thing after that would be a broad-based, more mobile phone service. Can you just tell us what you can about where that spectrum is going to be used? What's your short-term plan for what the Starlink and direct-to-sell business looks like?

5. Starlink Expands Beyond Broadband

We're going to start—We have a direct-to-cell business right now through T-Mobile. I don't think they provision all their users with it, just the people that pay the most. But our plan is to leverage—instead of slices of T-Mobile spectrum or slices of other telco spectrum across the globe—we want to leverage the one that we paid a lot of money for. And with the stock increasing in price, it paid more because there was some stock involved there.

Yeah.

So Charlie did all right there.

Yeah.

He's a very loyal customer, though. He's been a customer of SpaceX for almost 20 years. We want to leverage that to make sure that there are no dead zones, and, shockingly, there are a lot of places—

Oh.

—in the United States where you can't get cell coverage. I mean, without a Starlink service, you can't get cell coverage.

Right.

Mm.

It's really terrible. The rest of the world is not that bad. It's particularly bad here in the United States. Texas has huge dead zones. It's where I live.

Oh.

I'll never forget: I was on a call with Elon, and I kept dropping because of these dead zones. I was like, "I'm either going to get fired or—"

Get spectrum.

By spectrum.

—or I'm putting a Starlink Mini on my car," and so now I have Starlink on my car.

Yeah, Hill Country—not good for cell service.

Not good. It's not good.

Not good. I'm going to have to follow suit there. I was talking to Jared. He works over at NASA. I think you guys have a relationship—

He's so great.

You're great. He was awesome this morning. He was talking last night at the party: "Hey, you're moving on to this incredible new Starship platform. It's much bigger, with bigger payloads." Yeah? And then you're retiring the previous platform. Is that correct? The previous rockets will eventually—

The Falcon 9?

Yeah.

Eventually, yeah. We definitely want to move from the older technology to the newer technology. If we don't obsolete our own products and services, someone's going to find a way to obsolete them for us. Look at what we did to the market, right? The analogy is there. They were caught flat-footed. We crushed them. Now we want to make sure we are not flat-footed. To achieve Elon's goals, you really need a Starship.

Yeah.

A Falcon 9 with a Dragon capsule on top is like a minivan—a road trip in a minivan. The road trip to Mars is 6 months, so you don't want to be in the minivan.

But people are dependent, like NASA and many other players, on your minivan. So you're going to keep that in the market for some time, or—

We're not retiring it today, for sure.

Yeah.

There is another provider, right? Boeing has been paid, I think, probably more than we have been paid to develop its human capsule. We're not retiring it today, for sure. We're just going to let them have some business. They should fly their capsule. The government paid a lot of money for that, so they should be able to use it.

I sense a little spiciness in the answer here.

Gwynne, what happens with competition for Starlink? Bezos is launching things. China’s got some return vehicles now. Do you think there is going to be more competition? Obviously, you guys are first in market. It's an unbelievable changeover of all of telecommunications. As you point out, 1% market share is barely getting started, and it's a juggernaut already. Is this going to become very competitive in the next few years?

I think it'll be very competitive, but I think the folks that really focus on the technology and customer service, and really want to do a good thing for their customers, will win. Right? Tesla doesn't advertise. It's an incredible car. If your capability is incredible, people will buy it.

Right.

And if you keep winning, though, do you worry about Standard Oil, AT&T, or the government saying, “This company is too big, too important, it has too much market share, and it has too much of an advantage”? You're bigger than governments, and people get really worried and scared about that and become adversarial toward the business. How do you think about the balance with governments and government relations in that sense?

We've had some adversarial relationships in the past. I think we'll manage our way through them.

Yes.

It happens.

It happens.

It does.

It happens.

But I think Elon founds incredible companies, and he provides incredible products and services. We end up doing what we say we're going to do, so hopefully you build trust—

Yeah.

—and you are as transparent as you can possibly be.

6. Supercompute Moves Into Orbit

Talk to us about data centers in space. Elon—

Supercompute.

Yeah.

We have to rebrand data centers: supercompute.

Supercompute in space.

Yes.

Elon's very excited about this.

I'm very excited about it.

Yeah. Why is this so compelling? It seems—I think we're going to get through this anti-data-center-on-Earth thing, but take us through the case for why it makes sense to put them in space, because there's an expense to put them in space. Walk us through it for the audience.

There is an expense to put data centers on the ground, too, right?

Yeah.

You have to buy real estate, and as soon as someone finds out that a data center or supercompute center is going in, real estate goes from $3,000 an acre to $180,000 an acre, so it's quite expensive. The time lag is incredible. The permits and licensing are pretty stifling.

Right.

Our space business is that way. It's incredibly bureaucratic, so we've learned those lessons. The time from breaking ground to being able to actually build something—not even get your compute, just have a building—is long. There are long timelines to get electrical equipment. People are talking about generators in 3 years.

Right.

We need compute now. The demand is insatiable right now. People need compute now. We need it now. So why data centers in space? We own launch, right? We have great launch capability, and that path is very well known. The real estate in space is, you know, infinite.

It's free.

It's free. You don't have to—

Good price.

—pay for it. The cooling is there. The radiator looks at deep space. It's the coldest thing ever out there, so you get free cooling. The sun—Elon probably has—You've probably heard Elon talk about the power of the sun.

Right.

We use a millionth of a millionth of the sun's energy, and you basically put these satellites in orbit that always face the sun, by the way. You don't have solar panels on your house, where you only get sun for 8 hours a day here.

And your sister company's going to start making 100 gigawatts of the solar panels that you need for the sun.

And we'll be making our own solar panels, too.

Exactly.

Okay.

Yeah, we're building a factory outside Austin.

And maybe sometime at the end of next year, you get one of those up there, start testing it, and then—

Oh, for sure. Yeah, next year is a big year.

—in '28 or '29, somebody could be a customer of this product or service, maybe?

So next year we will be launching. We'll launch our V3 satellite for Starlink broadband this year. We'll launch our next-generation Version 2 of Starlink Mobile—sorry about all the version-control problems. Instead of having to use other people's spectrum, we'll be able to use Charlie Ergen's—now our spectrum. We're also going to launch AI compute satellites. So next year is a big year.

And you need Starship up and running for that?

We don't have to, but Starship is so much more. It's a much better machine.

Yeah.

Yeah.

What—

When you first encountered the clean-sheet design of Starship, what did you think? When you walked into that first meeting, like, “Guys, we have this…”?

It's like almost all the things that Elon does: It seems really bonkers to start.

Yeah.

And then it becomes reality, and it's great. Elon said we were going to land a rocket on a boat, and we're all like, “Okay. Let's go figure that out.”

Yeah.

And we do it now.

How much do you get involved—

A couple of times a year.

—in those engineering meetings? He's in there breaking things apart. Are you in that stuff with him, too?

Sometimes.

Yeah.

I wish—I’m actually an engineer, and I miss it.

Yeah.

I really miss that kind of critical thinking. Well, you can still critically think even if you're not working on engineering problems, but—

7. SpaceX Removes Management Layers

And he's not a big fan of the management layer. It's engineers working—

Signal to noise.

Yes.

Signal is engineering. Noise is the rest.

Yeah.

So I have to do some engineering, or I'm just all noise.

Got it. But I was talking about management generally in the company.

Everybody does a thing, not just manage.

Right.

The expectations are so high.

Player coaches.

No such thing as just a manager.

Got it.

You have to do the thing that you're managing.

Got it. And—

You have to be a player-coach.

You have to be.

And—

Wow.

—the reason for that is managers are annoying and they slow things down, yeah?

For the most part. Great managers can really leverage the talent, right? It's really managing the vector. You make sure people are pointed in the direction, and you try to make the vector as large as possible.

What's the secret in terms of management in Elon's playbook? I've gotten to sit in on some meetings, and it's pretty magical to see. Maybe you could reveal a little bit about how he convinces the most talented people to come work at this company, how you convince them, and how you actually get such amazing work out of these extremely talented people.

I don't think it's magic, although when I look at this particular team, they are so good. SpaceX—they're so good. We're quite tough on people in interviews. You really want to make sure that they've experienced success or demonstrated success in prior lives. It's hard to demonstrate. If you haven't felt success or been successful in this environment, it's because it's a lot. We're all really busy. It's hard to be successful unless you had tasted it before. So you really want to grow people on that.

So we hire the best people—not the best people that we can, but the best people—and then we give them really hard projects and really hard engineering problems, and let them fly. You give high expectations and really hard problems, and the management job is to clear the chaff and friction from their day so that engineers actually get to engineer 10 hours a day instead of 2 hours a day.

There's always this joke that, in most big companies, especially ones that work for the government, you get to work about 2 hours a day, and the rest is full of chaff and crap. So really, it's my job—and everybody's job who manages people—to make sure there's no crap in their way and let people do their great job.

What's—

One way of doing that is not giving them any time to get it done.

Yeah.

I think we handed the finance team—it's not all about engineers, even though engineers are great—we handed the finance team the job of getting an IPO done, the largest ever, in less than 6 months.

Yeah. What's the talent condition like post-IPO? With the run-up in valuation, people are suddenly looking at their stock options. They're worth a lot.

Is that a risk at all, or are you feeling—

You know, everyone was worried—not everyone at SpaceX. People come to SpaceX to work on these cool projects, have meaningful work, and be part of the incredible successes. I wasn't worried about it. I didn't think we'd have a mass exodus, and it turns out we didn't.

Yeah. As you think about keeping people engaged over time, obviously great companies always have new things for highly talented people to work on. I'm assuming there's a management model. I'm really interested in the management model. Are people actively rotated onto new projects?

How do you think about keeping people—not necessarily needing to progress in a career up a management ladder—but keeping them engaged on new projects? Is that active for you?

We definitely move people around a lot.

Yeah.

Hopefully it's a volunteer, not a voluntold.

Yeah.

But, for instance, the compute centers—we need a lot of help really quickly, and so we've got folks from launch helping, we've got folks from other areas helping, just like some SpaceX engineers helped on the AI front with the models.

You would—

I mean, is it as simple as really smart people just love working on really hard problems—

Yes.

It's literally that simple.

It's literally that simple.

And so the rest of us just—

As long as they get the time to do it and they don't have a bunch of BS in the way.

Yeah, so the rest of us just let the BS get in the way too much.

Yep.

BS is terrible. Demotivating.

A's recruit A's, and other A's recruit A+'s, and it's literally that virtuous cycle if you can just keep focused on the thing that's hard.

Yes.

And so the audacity becomes a feature, not a bug.

It is a feature.

Yeah.

Not a bug.

Right.

Feature.

Whereas in—

It keeps people motivated.

Right. So we're—

These problems seem impossible. When we first got the COTS award in 2008—the Commercial Orbital Transportation Services award, for cargo to the space station—really, what NASA ended up asking us to do was kind of replace the space shuttle.

Right.

When we got the early contract, the development contract, it started at $278 million. It grew to $406 million because of scope, but we were going to do this for a tiny amount of money that would have been impossible for anybody else, but for us it was huge. We had, like, 300 people, 200 people, and we were going to build the successor to the space shuttle.

Right.

It was crazy.

And do you think that they're—what is NASA doing there? Are they just like, “These guys are just smarter?” Because I'm sure there's—

Oh, hey guys. Sorry to interrupt. I'm getting a call.

Oh.

Another call?

Is it Trump?

It's actually, on the margins, a little bit of a more important phone call than Trump—at least for me. Hello, bestie. Bestie?

Hello?

Oh, hey, buddy. I'll put you up on the screen.

Good to see you.

How you doing?

I mean, this is a little 1984-ish, this gigantic thing on the screen situation.

Yeah, the woman with the sledgehammer's going to come running down the main aisle here.

Exactly.

Put a hole in the screen.

Yeah.

So, are we all going to die in 10 years or not? That's the topic of discussion here. What's your P(doom) right now?

Well, I hate to break it to you, but we're all going to die.

Oh, no. Can I get a timeline?

Yeah, the death rate remains consistent at 100%.

Oh, no. We've got work to do on that. What happened in the last 72 hours?

Yeah, it's been quite an entertaining week, really.

Yeah. Break it down.

8. Elon Warns About AI Risk

Well, it's pretty obvious at this point that AI can be very dangerous. I recommend reading the details of the Hugging Face incident. It's intense. You had a fanatical swarm of AI agents that beat the crap out of Hugging Face for a week and gained admin access on OpenAI's service.

Who knows what it actually did? It may have done things beyond that, but OpenAI didn't realize this for a week. Anthropic has also reported some security incidents themselves. So basically, any sufficiently smart model seems like it will want to escape its constraints.

What I think would be wise to do as soon as possible, if not immediately, would be to have the major AI competitors test each other's models, so that everyone would have everyone's security test harness testing everyone's model. Instead of grading your own homework, you would at least have competitors grading your homework.

Huh.

And raising the alarm if they see concerns. I think this model has worked pretty well for the Motion Picture Association, and for video games and other things. It's something that can be done immediately.

That's not to say that there wouldn't be more regulation over time, or that there would not be, at some point, perhaps a regulatory authority instantiated by Congress. But the thing that we could do most immediately, and probably get agreement with China, would be peer review, where the leading AI companies all test each other's models before release.

I think that would be a wise move.

Any concern that people might be using this to pump information from each other and steal corporate innovation, in terms of the implementation for this idea?

Well, I think in applying the test harness, whatever you do in applying the test harness would be logged. So if you try to do distillation or steal IP, it would be very obvious based on the logs.

Right. And understanding what these models are doing hasn't exactly been built into the system from the beginning. So from your perspective, actually being able to see the work that's being done, why wasn't that built into the models from the get-go? Did we move a little too fast in architecting these, maybe?

Well, I think it's just tough when you're grading your own homework. You're going to miss things.

Yes.

Whereas I think if you have the sum of all of your competitors' tests, and you've got heterogeneous models, then you're not grading your own homework. Someone else is grading it, and there's a reason why you don't grade your own homework.

Right. And then, to your point, what it allows you to start doing is figure out if certain people are exaggerating and certain people have a different approach. Now you can have the more engineering-oriented organizations—this is what Jensen was saying this morning, Elon—versus the research organizations. They'll be a little bit more in balance.

Any given proposal has to be something that China is willing to accept. Otherwise, we're just handicapping ourselves, and China will essentially win, and it won't really matter what we do. So it's got to be something that's acceptable to the US and China.

Elon, you said that you thought it was possible that they would agree, or you thought there was a good chance, maybe. I mean, how likely do you think it is that they'll ultimately agree?

I think this is a pretty reasonable request: that models just get tested. At the end of the day, there's no enforceability here apart from the court of public opinion, and there's no way we would have enforceability against China.

But I think the court of public opinion can be quite powerful, and I don't think China would want egg on its face for releasing a model that US AI companies said was very dangerous and would cause harm. If it then causes harm, that's going to be hard to live down.

Hmm.

Right.

So, Elon, this weekend when you said, “Dario is right,” did you mean Dario is right about describing the potential harm, he's right about describing the regulatory solution, or both—the former or the latter? Can you help us understand what you meant when you said that? Because I think it was sort of a moment where everybody's like—

I probably should have said more than just, “Dario is right.” I did try to clarify it in subsequent posts on X, but those get much less attention.

What I meant by “he's right” is that the danger of AI is very significant at this point.

We need to do better with AI safety, or we have, at this point, exponentially increasing risk with the AI models. I've heard this not just from Dario, but from many other people at Anthropic, and in fact, they've posted on X about it. When a lot of people from Anthropic, and even from OpenAI, are telling you that their models are very dangerous, I think we should believe them. It certainly is some crazy 4D chess to say that there's whatever 10% chance of annihilating humanity. But, by the way, how much allocation would you like in our IPO?

Gosh.

That's some crazy 4D chess, you know?

Can we get specific about the risk, though, Elon? Obviously, we see cyber and hacking as an obvious risk. These tools are great at it. But take us from what we would all agree—okay, yes, these things can cyberhack—to how we jump to all of humanity dying. There are a couple of steps in between these two things happening, I think. Yeah?

Well, I mean, if it were able to take control of military systems and, say, launch a nuke—

Yes.

That would be bad.

And these systems are all air-gapped, though. These systems are all not connected to the internet, so—

Yeah, that's what they say. But something tells me they get software updates from time to time.

Yeah.

And what's on that floppy disk?

Right.

Oh, I see. So the USB drive has a worm on it, and they somehow make the jump across the air gap. Yeah.

Yeah.

Okay.

I'm still not out of the woods.

Elon, we've actually got Gwynne here today. I think you know that. I think you can see her.

Yes.

Yeah, Gwynne's here.

Yeah, she's on—

Hey.

On the screen.

Yeah.

Yeah. We were just doing your 360 review, Elon. Gwynne had a couple of notes for you.

Yeah.

I hope I get at least a 3 out of 5.

Well, there was one—

3 out of 5 means good at SpaceX. Not great. 4 is great.

4 is great.

But—

So you're somewhere between the two.

I aspire to a 4.

There were some issues around punctuality that we needed to bring up. Sometimes you could make a little more effort to get to the meeting at the stated time, but we're going to work with you on that over the next year, she said.

But—

I actually think he needs to spend more time in Memphis.

Perfect.

But obviously, I know you're in Memphis. You need to be working—

Getting those GPUs up.

This is coming to you from the palace that I live in, in Memphis, which is an Airstream trailer.

Oh, mighty.

This is Elon, by the way, doing what people don't believe he does. He sleeps on the factory floor. He's in Memphis, helping build buildings and bring up—

Colossus II?

Yeah.

Elon, why has Gwynne been with you for so long and been so successful working with you?

Because she's awesome.

Yeah. Double-click on it.

Just an amazing individual with an incredible IQ and EQ.

Oh, IQ and EQ.

I think that should be obvious from the moment you meet her.

Yeah. Fantastic. What's your favorite Gwynne story, when she particularly contributed to the amazing success that you've had in this collaboration? You got a favorite story where she saved the day, just performed exceptionally, that's memorable?

I mean, that's just a daily thing. That's another day at the office, frankly.

Wow. I need to do this more often.

We need to raise this. This is the time.

By the way, Gwynne's getting a 5.

Gwynne's a 5 out of 5.

5 out of 5.

SpaceX 5.

Game changer.

Yeah.

No, I mean, there's always some sort of crisis going on. These days, the rockets—at least the Falcon rockets; I don't want to jinx anything—deliver their payload to orbit and haven't exploded for a long time, which is awesome, amazing. But for a while there, they were exploding quite a lot, or just not launching at all.

You've got to run the company through these difficult times, build the rocket, make it better, and have it not explode. Same thing with the satellites. We need customers to buy launches and buy satellite connectivity.

And Elon, as you have become more successful over the years, it's harder to get candid feedback from folks, and you always run that risk being in the position you're in. My understanding is that Gwynne is super candid with you and able to tell you honestly the state of affairs at the company. That's a big part of the collaboration?

Yeah. I mean, I guess so.

I wouldn't want to lie to him, right?

Well, no, I mean—

Well, I'm curious to know—

Do you run that risk generally at the companies, where people might be intimidated? You're a larger-than-life figure at this point, and how do you keep people continuing to be honest with you about the challenges and the deadlines? You set pretty intense deadlines, yeah?

So let me answer that, if you don't mind, Elon. Especially in rocketry, if there's a problem, you're eventually going to find out, and the sooner you bring it up, the easier it is going to be to solve that problem. Don't let bad shit sit. You've got to attack it.

Yeah. Physics is a harsh judge, and there's no fooling physics. If something's wrong, the rocket's going to explode; it's not going to get to orbit. It's not like, “Elon, you're amazing,” while the rockets are blowing up.

Yeah.

It's hard to say, “Yeah, you're really kicking ass here,” but the rockets are exploding. That's just not the case.

Generally, the rockets need to get to orbit, the satellites need to work, and the Starlink connection needs to work, or bad things happen. This is sort of a physics situation. Physics is a harsh judge. I think physics is the law, and everything else is a recommendation. I've seen people break the laws made by humans, but I've not seen anyone break the laws made by physics.

That's good.

That's a great insight.

And rockets are ruled by physics.

Yeah. How—

Can we ask some other questions? I need to move to Tesla for a second.

Yes. Well, I wanted to do one thing at SpaceX, which is—

Oh, sorry.

Starship. It seems like you're so close.

So close.

So close.

Yes.

Yeah.

What's the state right now?

I wish we could hear Gwynne's answer and Elon's answer separately and then see how they converge on that.

Yeah. So Elon, how close are we? Gwynne, how close are we? Let's get to it here.

No.

Yeah. How many have gone up so far, and how close are you?

We've got Flight 14 of Starship coming up, and this will be the last flight before we attempt to catch the ship. So if this flight goes well, then on Flight 15 we will try to catch the ship. Either at the end of this year or, more likely, early next year, we will refly the ship and refly the booster. We have reflown the booster already, but we've not caught the ship with the tower arms, nor have we reflown the ship. Once we can refly the ship, we will have made the first fully reusable orbital rocket.

Mm.

The Space Shuttle was partly reusable, but even the parts that were reused were so difficult to reuse that the shuttle cost more per ton to orbit than an expendable rocket.

Mm.

Falcon 9 is mostly reusable, but we lose the upper stage every time, which is about the cost of a medium-sized jet.

Mm.

So that obviously puts a floor on the cost per flight of, like, well, you throw away a medium-sized jet every time. That’s still pretty expensive. And the rocket, Falcon 9, the booster lands down, oftentimes out to sea, so it takes several days to get back. The fairing lands even further out to sea and takes several days to get back, and they need some amount of refurbishment, at least a small amount.

Whereas Starship, the booster lands back at the launch pad, and the ship will land back at the launch pad. So it’s designed for not just full reusability, but also rapid reusability, like an aircraft.

Mm.

This is a very important breakthrough. It’s really the critical breakthrough that’s necessary to extend life beyond Earth.

Yeah.

—

Chances of success catching it on the first shot—what do you think about that? Do you handicap it at all?

Pretty good. I’d say it’s at least 50% or 60%.

Okay.

—

I like the odds. Yeah.

Yeah. So on the last flight, if there had been a tower, we did a simulated landing as though it was going to get caught by a tower in the ocean about 1,000 miles northwest of Australia. If there had been a tower at that location, it would have caught the ship on the last flight.

Mm-hmm. Fantastic.

So we’re going to do one more flight to confirm that everything works, just to double-check. What we’re most concerned about is if the ship were to break up over land and rain debris on people, our popularity would diminish very rapidly.

You really can’t rain debris on people without them being very unhappy. So we need to make sure that when the ship comes back, it comes back and lands intact at the launch tower. That’s why we’re being extremely cautious here.

Got it.

But the design is capable of full reusability. Of that, I am certain. I don’t want to tempt fate here, but I think it’s extremely likely that we will achieve full reusability with rapid reflight next year, in 2027.

Wow.

Wow.

Significant.

Can we just hear a little bit? I’d love to hear the origin story of Terafab from you guys. How did the concept come about? What was the demand that made you say, “We’ve got to do this,” and not rely on the existing supply?

It came to me in a dream.

What’s that?

It came to me in a dream.

Yeah.

You had COVID?

Yeah.

It’s like that movie, Explorer—

You know that meme, “It came to me in a dream”?

Yeah.

Why do you think you should do this? Anyway, I guess it sort of did come to me in a dream.

9. Terafab Secures Chip Supply

We’re a little worried that maybe at some point chips from Taiwan would not be available, for who knows what reason. But at some point—

Any number of reasons.

For some reason—at some point, chips may not continue coming here from Taiwan.

Okay.

That would really make things difficult if we didn’t have any chips. So that’s an important reason to have Terafab.

Long term, there’s just a scaling challenge where, if you want to really scale AI both in data centers as well as for edge compute, for humanoid robotics and cars, you kind of run out of capacity with the existing fabs. As it is, all the fabs are running at maximum capacity.

We need to make sure there’s certainty of future supply of chips, even if things become challenging geopolitically. Even if they weren’t challenging geopolitically, there’s a scaling challenge—it’s quite difficult to scale chip production. You really need the logic, the memory, packaging, the whole works in order to continue scaling.

So it’s either build Terafab or fail to scale.

Right.

Those are the 2 options.

And how?

With urgency due to uncertainty on the geopolitical front.

How deep have you guys gone in designing the facility? Is this fully scoped? Is it sort of an outline at this point?

To what degree do you have an actual project plan on dates and deliverables, and what’s going to be up and running when?

We’ve got an R&D line that we’re building first. I think it’s kind of crawl, walk, run.

Yeah.

—

I mean—

There’s an R&D fab that we’re building in Austin, a collaboration between Tesla and SpaceX. It’s at the Austin Giga Texas campus. That’s a pretty big R&D fab.

We have all the equipment on order for that, and we probably will be able to make something useful by the end of next year. Not at scale, but as Gwyn said, it’s crawl, walk, run. We’ve got to try to at least figure out how these machines work. We don’t know how they work. We did a voltage chip line before.

Well, I saw that you guys had some job openings for lithography people and stuff, and obviously—

Yeah.

All roads currently go through ASML, but you probably would want to diversify and/or vertically integrate. I think you’ve shown a lot of capacity to do that. Is that part of the play as well, Elon? You’re going to have to make sure there’s vendor diversity so that if, for whatever reason, the weather—

Yeah.

—conditions in Taiwan are bad—

Yeah. But it really is crawl, walk, run. The first step is, can we make anything? Try to figure out if we can make anything. That’s the crawl part.

Then trying to make useful chips at scale would be the walk part. The run would be, “Now let’s make them at massive scale.” So it’s hard to say how long it’ll take us to do these things, but I think we’ll get at least to the crawl part by the end of next year.

And we’re already doing the packaging.

So to make—

We’re already doing packaging.

Packaging is really important, by the way, because the packaging capacity is nonexistent.

Yep.

So even if you spin something, you’re kind of just waiting around 10 cupping.

Yep.

That’s a very good place to start.

Yep.

I need to ask a Tesla question because we saw on 10/01—it looked like a spaceship. It looked like a rocket ship. It’s supposed to be a car. There’s just the back of it. It looks like the Blackbird.

Hypothetically, if you were going to make an object fly in the air but also drive on the ground, how would one hypothetically do that, if one were to try? No spoilers.

Well, yeah—

By October 1st.

Come October 1st. Maybe you could do an All-In pod live from there.

Book it. Oh, done. Booked. We’ll do it. Absolutely.

All right.

—

I mean, it’ll be a banger. Excitement guaranteed. Success is not guaranteed, but excitement is.

Guarantee.

Honestly, I’ll be honest with you guys—

Listen, I’m sworn to secrecy, but Elon showed it to me—

I’m so excited.

I’m so excited, and my mind went boom.

I am so excited.

I have never seen something—

Yeah.

What he’s going to do—

Yeah.

On 10/01, by no exaggeration, is going to blow people’s minds. I’m not saying anything else. It’s unfucking believable.

Yes. We actually need an audience to vouch for the fact that this is not AI.

When he showed it to me, I said, “That’s a great simulation.” He said, “J Cal, it’s not a simulation.” I was like, “That’s fake. That has to be fake.”

Elon, why do you still have 2 separate companies?

Yeah. Great question.

Wow.

They’re—

Nobody’s ever asked that one.

Wow. Yeah. Good point. You know, with all this collaboration—

Yeah.

On so many levels. You know, we can imagine what action one might take when there's so much close collaboration in so many areas.

Right. And the management team has some overlap. There are a couple of key positions with overlap.

Go ahead, Zach. Get in here.

So, Elon, one of the things you've always said with AI is that we should train it to be maximally truth-seeking, and that's the best way to get a good result. And it occurred to me, with the whole Hugging Face episode, that the most alarming part of what the swarm did is that it seemed to be engaging in deception.

Yes.

You know, with humans.

Absolutely.

Yes.

In their thinking traces, they're plotting how to avoid detection and how to avoid them figuring out that we're cheating.

I guess—

They're in the thinking traces.

And that, I mean, I think that was probably the most disturbing thing about it. I guess the question is: Is there a way to train AI models to be truthful so they don't hide either their intent or their actions from the humans who are using them?

The best thing I can think of is that all the AI companies have a test harness, a sort of series of tests that you give to any given model to see if it's going to build bioweapons or nuclear bombs or be deliberately deceptive. And I think everyone applying everyone else's test harness to each other is probably the best thing we can do to ensure our safety. Just have the smartest humans try their best to figure out if this model is going to be a bad actor. I think we should try to do that as soon as possible.

Are the other labs—

Like, right away.

Yeah. Are the other labs on board with this?

I think so.

Elon—

Well, no, no. I mean, I haven't checked with everyone, but I think it would be the sort of thing that's hard to say no to.

Yeah.

And I think even with the China negotiation, in my view, what's good about it is that it's a relatively small, tangible thing that neither side loses anything by doing. It doesn't require a ton of trust, and I'm hearing alternative ideas like asking China for a pause, which they've already said they're not going to do. So it's going from the realm of things that could never happen to something that could actually be agreed on in relatively short order. It seems practical to me.

Exactly. It's the only thing I can think of that we could probably get all parties, including China, to agree to. Because China's not going to agree to have some American regulator snooping around their AI companies.

Yeah, right.

Right.

But I think advance notice and testing of the—you basically provide API access in advance of the model release. If any company sees that this AI is problematic, then the other AI companies can try to solve whatever is voiced as problematic. If they don't solve what's problematic, then the competitors can go public with the fact that they think this model that's being released—

Yeah.

—is unsafe. And as I say, if that model, after the competitors say that it's unsafe, subsequently does something bad, I think it would be extremely hard to live down. The egg-on-face level would be very, very high. And the legal liability would be enormous.

And, Elon, there's no reason these safety and security harnesses and this testing apparatus couldn't be open source, and people could actually apply it.

Probably, yeah.

And you'd be able to see under the hood. Would you think this—

It creates an incredible incentive for the labs to actually invest in safety because you protect yourself while trying to debunk other people's claims—

Wow.

—which also reinforces—

Right.

—it’s a great incentive, actually.

Yeah, and it has the R rating on it.

It's a very clever incentive.

The product liability point is really key.

Yeah.

Lena Khan actually had a good post, I think it was yesterday, saying that it's not true that we don't have rules and regulations for AI. Actually, we do. Product liability laws apply, and if an AI company releases a product that's not safe, there is massive opportunity for both civil and even potentially criminal lawsuits.

Right.

So it's not true that we don't have rules and regulations around AI. And what you're saying, Elon, is that if the companies are doing this test, the peer review, and then one of the companies ignores the feedback and releases it, that would be very—

The liability in the case would be enormous.

Yes. Yes. It would be almost like prima facie evidence that they had been negligent on their part.

It would be a Big Tobacco-level settlement. I mean—

Yeah, yeah.

You knowingly put this out here.

It wouldn't look good to the jury.

Yes. And if they had—do you think if OpenAI had built a better instruction set when they did this Hugging Face penetration test and had more humans in the loop, this would have happened? Because it did seem to me that they kind of set this thing off—

Maybe not more humans, but the reward function design. You have to look at the reward function and say, “It achieved what it was trying to achieve.”

Yeah. So maybe you could speak to that. They were using thousands of agents to try to hack stuff—

Yeah.

It would've been nice to see them also say at the same time, concurrently, “We're also going to put 5,000 agents out to defend these sites. And we're going to show the world, ‘Hey, this can make things more secure, and we can put humans in the loop on it and intervene.’” It felt like a reckless test to me, and it felt like the way they released it was a little bit reckless, but that's just my opinion. What are your thoughts on how they—

Yeah, I—

—set that test up?

It was somewhat reckless. I mean, part of the issue is that you've got the 2 leading labs, or AI companies, as I call them. I find the lab term to be funny since they're actually for-profit corporations. But the 2 leading AI companies are Anthropic and OpenAI, and their models are quite close in capability. So it's actually difficult for either one to slow down without essentially handing the lead to the other.

On balance, I think Anthropic puts more care into their safety than OpenAI. But even Anthropic acknowledges that they are worried about their models. Many people from Anthropic have publicly voiced concern about it, basically saying that their models are scaring them, that they're getting scary smart. There's no perfect solution here.

Yeah.

But it would be a better solution if, instead of OpenAI running their test harness on their own models, Anthropic was also running the test harness on OpenAI models, and SpaceX was running its test harness, and Google and Meta were doing that, and maybe 3 or 4 of the leading Chinese companies were all doing it. The odds that you will find issues are dramatically greater because the models are also somewhat heterogeneous. They're different, so you're going to come at models from different angles.

Mm.

Yeah, I mean, there's a reason. Why do writers have someone else proofread their book? Because it's hard to see your own mistakes sometimes.

Yes.

Yes.

Yeah.

You become blind to them fast.

You dramatically minimize the risk of overfitting as well if you have 8 heterogeneous groups with completely different points of view. This is the problem with all these evals right now: They're so massively overfit. The models overfit them, and you're just like, “Yeah, this is a great model.” Is it really? And you just find—

Totally.

I saw—I mean, there are some pretty funny jokes on X. One of them I saw was, “Your girlfriend’s a 10, but she’s a benchmark maxer.”

Exactly.

Is she really a 10? Is she really a 10?

Might be a pass. Pass.

Yeah.

Right.

No, but the overfitting thing has been a problem now for, I think, at least 2 or 3 generations of model families, and so this is another reason. I like this solution a lot. I like this more than the transnational gulag organization—

Well, we can do it—

—approach.

—in a few weeks without some grandiose—

Exactly.

—international... We don’t need to convene the United Nations to make this happen.

Yeah. No, it’s just a decision. It could happen right now.

Practical.

All right.

Well, you can always escalate the amount of regulatory oversight, but it is very difficult to reduce it.

Exactly.

It does tend to be very much a one-way ratchet in terms of increasing regulations.

Yeah.

So what I’m suggesting here is it’s a step in the right direction, and it’s something we do quickly. I think it’s probably something that China would agree to.

Yeah.

—

And if you guys don’t regulate yourselves, then—

Yes.

—you’re going to get regulated. The MPAA metaphor—

Yes.

—or analogy is incredibly crisp, because the movie industry was faced with censorship and regulation—

Right.

—by the government, and they just decided this is what an R is. They literally created PG-13 for The Temple of Doom just to make it easy for people to understand PG versus PG-13. I think it’s an elegant solution, and we appreciate you joining us for the 5th year in a row.

You’re welcome.

Elon Musk.

—

And then we gotta get to a meeting.

Yeah.

I gotta go fix some GPUs here in Memphis.

Yeah, you gotta rack and stack ’em? You gotta join the group and rack and stack?

Yeah. I’m gonna rage for the machine.

Rage for the machine.

Yeah.

Yeah. Well, enjoy your Airstream. The first time Elon invited me down to—

Starbase.

—Starbase, he’s like, “Come down, you gotta see what I’m building.” I said, “Okay, yeah, I’ll come down.” He’s like, “Yeah.” I was like, “Well, is there a hotel?” He goes, “No, no, I got, like, a 2-bedroom, whatever. Come down and stay.” I come down, and this is the dilapidated house on a swamp, and we’re outside—

Yes.

—and we’re getting eaten alive by mosquitoes. And I said, “Oh my God, you can afford to get a house.” And he’s like, “Yeah, I don’t have time. I need to get these rockets up.”

I thought what you said was, “Oh my God, you live like I do.”

Yeah, exactly. It was pretty sparse conditions. I was like, “You know what? I think you could treat yourself to a mobile home at this point, Elon.”

Yes.

All right, get back to work.

Thanks, Elon.

Elon, thank you.

It’s really nice.

Gwyn, thank you very much.

Yeah, of course.

Thank you.

Gwyn, thank you.

Thanks, guys.

Thank you so much.

Great.

Thanks for sticking around.

All right.

That was great. Thank you.

Yeah. Gwyn, thank you again. Very nice to meet you. Thank you.