Gavin Baker 在 Mission Control 专访 SpaceX CFO Bret Johnsen

Launch is foundational to everything you do.

Absolutely. I think launch is the starting point. If you're going to be a space company, I tell people it's hard to be a space company and not have assured access to space. So we started out making sure that we had launch nailed down, and Elon really focused on driving down the cost of access to space at a far cheaper price than anyone had ever even thought of before.

We're now at the lowest cost per kilogram to space ever in the industry, and we're looking for Starship to do another 10x improvement as we get to rapid reusability with Starship. So it is definitely at the core of what we do, and it's the enablement for all of the other businesses, whether it's Starlink, direct-to-cell very soon, or AI compute.

I think you absolutely have to start, when you talk about SpaceX, by talking about our launch capabilities. Starship will be, to your point, next level, because what we're doing now is taking on what I would say is the holy grail of rocketry, which is rapid reusability. It's one thing to bring back the first stage, which is amazing and transforming the industry, but we did that 10 years ago with Falcon. What we're doing now is flying the largest rocket ever built, with the goal of getting it to aircraft-like operations. That is a completely different dynamic, but I think that's what it will take to be the catalyst for the whole space industry and turn the 2030s into something that we had expected when we were kids related to the space industry.

1. Starship Targets Rapid Reuse

Where are we in the Starship program? We just had a launch. It was a V3 of the ship and the booster. I will admit, it felt like a high-risk event heading into an IPO. We were both there, and it felt like a high-risk event, but from my perspective, and as far as I can tell from the SpaceX team, it seemed very successful. So what did we learn? Where are we with Starship, and where are we on that path to rapid reusability?

I think our first launch of V3 that you're referring to from just last week was a huge success for us. The fact that we were able to demonstrate the full system capability, the new Raptor 3 engines, all of the changes that you saw on the bottom of the vehicle, and all the operational changes that we made as well—and then you saw that soft splashdown at the end of the second stage—I think gives us a lot of conviction about where we're going, not just in years to come, but really even in the next couple of flights.

That's really exciting because, as we were talking about, that platform is a catalyst for the rest of the business. Having a rocket that can take 100 metric tons to low-Earth orbit is going to be hugely important for everything we're about to go do.

And that's 100 metric tons at what cost relative to Falcon?

Well, I think as soon as we can bring back that second stage to the tower and start rapidly reusing it—in the next couple of years, we'll just say—I think you're going to experience a 10x reduction in cost per kilogram to space from where we're at with Falcon today. I think that's a huge springboard for multiple pieces of our business.

2. Starlink Scales Global Connectivity

I think of launch as enabling a variety of applications for SpaceX as a company. Microsoft had the operating system that enabled a lot of applications. Launch is foundational, and the first, biggest, and most well-known application is Starlink and your connectivity business. Can you just give us a sense of the scale and growth of that business?

Yeah, it's very exciting because we actually haven't been in the connectivity business for that long when you think about it. We really had our first production satellites in space six years ago, and we've ramped that business to where we're at today, with over 10 million customers, over 10,000 satellites flown to space, and service in more than 160 countries around the world.

It's clear that we're delivering a capability that the world really wants, and it gives you so much conviction when you know you're delivering a great product. We send out an email each month with how many communities around the world we've had a positive impact in. It's really special, whether it's indigenous tribes in Canada, communities down in Brazil where we're connecting schools, or schools in Africa where they haven't had any connectivity before. Bringing something like Starlink to the world has been pretty amazing.

What it's done is give us an extension of the mission we were talking about. First and foremost, we want to make mankind multiplanetary. But above and beyond that, having the ability to connect the other 3 billion people on the planet and really bridge that digital divide has been really special for folks. I think that really resonates.

Starlink was that first piece of the business, and I think those 10 million customers can become hundreds of millions of customers around the world in time. It's so much more efficient to deliver to so many different locations from space than it is terrestrially, and we're just heading down that path right now.

We're incredibly vertically integrated in that business, so we end up being a little bit of our own blocker now, which is great. That's exactly the way we want it. But Starlink was also, I would say, the catalyst for the Falcon business to really ramp up operationally. If you don't have payloads lining up, you don't really have a reason to go from 10 launches to 22.

We launched 165 times last year with our Falcon vehicles. We bring multiple stacks of Starlink satellites to the launch site so that when there's an opening between launches of third-party satellites, we can take that slot for Starlink. As we go into our next phase with Starship, I think we have another catalyst like that. Above and beyond Starlink, we're now going to have AI compute satellites.

Absolutely, 100%. I'm super excited to talk about AI compute, but I do think Starlink deserves a double-click because I am a former telecom analyst, so it is near and dear to my heart.

Yes.

A few observations come to mind for me. First, people have forgotten how big the telecom markets are. I just use round numbers: $800 billion for internet access and $800 billion for cellular connectivity. So, a $1.6 trillion market, plus or minus—I'm within the nearest $200 billion.

What really strikes me, as a student of the telecom industry, is that there's never actually been a differentiated or disruptive product in telecom before, because everybody has access to broadly the same towers, the same rights-of-way, and the same equipment. The products end up being very similar. This was true for local telephony, it was true for long distance, and it's true for cellular.

I am a very serious video gamer. I'm not very good, but I play video games the way a lot of my peers play golf.

You're my boss.

But I take it very seriously.

Yeah.

My grandfather used to say that age and treachery will always triumph over youth and skill. In video game terms, that means having the best GPU and the best connectivity.

Sure.

At every location I've been to, Starlink is better. It is faster and has lower latency. I think latency is a very important point for the user experience. So can you talk about how this product differentiation is going to help you go from 10 million to 100 million to hundreds of millions of customers?

Yeah, I think it's pretty exciting. We'll talk about broadband in a minute, but especially if you think about direct-to-device, which we're going to ramp, our next generation of direct-to-cell or direct-to-device will be 5G-quality in the next 2 years. We're about to bring something that's pretty unique, to say the least, and something I think anyone would want: to go anywhere with your phone, have global roaming, be out in the middle of the desert or on the highest mountain, and not have dead zones.

These are pretty special things that I think people would want and probably pay extra for, especially if it's the same price. So we are very excited about bringing a very differentiated product to the market. It feels great when you're bringing a product that's great for everybody.

It's also going to be great for disaster recovery and emergencies. The fact that the network doesn't go down means you don't have to worry about a tower being out. Circling back on the broadband side, the fact that you can deliver low latency, to your point, is really critical. Low-latency, high-speed capability pretty much everywhere—other than the countries that I'm not allowed to go into—is something that just won't make sense terrestrially.

And so I think it will become harder and harder to justify many of the deployments from a terrestrial perspective because Starlink now exists. I do think people are really starting to see that, even on aircraft, when you start to see all the announcements from United Airlines, now American Airlines, and many of the others, without listing them over and over again. Getting on an airplane and having that type of low-latency, high-speed experience really opens people's eyes to what it could be at their own homes or offices as well.

Awesome. And I'm glad you brought up the disaster. What SpaceX does anytime there's a natural disaster—

Yeah.

It has undoubtedly saved lives. Surging Starlinks for free and turning them on—this has saved lives.

Yes.

So outside of Starlink and orbital compute, which absolutely merits its own discussion, what are the other applications or businesses enabled by Starship? What are you most excited about?

Well, I do think on the connectivity side, those 2 businesses will ramp nicely. What's also great about those is that when people think about space, they think about the new markets that we will create in the years to come, which I agree with. Whether it's point-to-point transportation on Earth—flying to Singapore in 30 minutes, kind of thing—or whether it's the lunar economy, I think all of those will happen.

But what I think people discount is that there are massive, to your point, almost $2 trillion, existing markets just on connectivity. They're existing markets today, really, and so the fact that you're able to deliver a better product in an existing Earth market before even talking about space is probably what gets missed some of the time.

And it's not just a better product. You do have a cost advantage.

Yes.

You're not digging ditches, right? You're not spending all that huge amount of upfront cost. You really send the terminal out once you've established your network of satellites. You send the terminal out, and that's basically most of your customer acquisition cost.

Yeah.

Better, faster, cheaper has been a winning experience. Good combo in my time as a tech investor. But any 1 or 2 of those markets—whether it's freight, whether it's point-to-point travel—

3. Orbital Compute Takes Shape

I think the next big one, though, is going to be AI compute.

Okay.

We should definitely spend some more time on that. But I think that is a market that really needs Starship to really happen.

Oh, yeah.

There are large payloads, and you really are focused on cost. I think that's the one that's kind of near and dear for us because you already see those next 2 happening, right? This will just be a significant enhancement.

If I think about what Starship's going to do for our broadband satellites—our V3 satellites that we're about to fly in the coming months here on Starship—every Starship launch of those satellites brings 20 times the capability of what we're flying with a Falcon launch today. So it is a huge enablement for our broadband business, and soon to bring 5G capability on direct-to-cell.

So Starship's rapid reusability really enables this much larger Starlink constellation that takes us from 10 million to the hundreds of millions. It actually lowers the cost of Starlink further by lowering the launch cost. It will enable direct-to-cell, which is awesome, and it will enable orbital compute. I would love to just have people hear “data centers in space.”

After lots of interactions on X, I've realized that a lot of people are picturing a Pentagon-sized building floating around in space, and that is not what it is. Orbital compute is racks in space. Let's just start there. I'm sure there are going to be images available to people, but describe what one of these satellites is going to look like so people can conceptualize it.

Yeah, I love “racks in space.” I might steal that from you. I think that's the right way to think about this. I actually had the same issue when someone said “data centers in space” internally the first time. I thought, “Oh, wait. How are we going to connect all these pieces together?” And they're like, “Johnsen, this is literally like another constellation.” I thought, “Oh, okay.”

But then you realize virtual networking has obviously been a key component of networking for a number of years now. This is that same concept of virtually networking rack by rack, to your point, with satellites that look basically like larger versions of the Starlink V3 satellite we're about to fly for broadband. There's a lot more solar, and now compute over the top. We'll just start with NVIDIA GPUs on there and a large sheet of metal from a radiative-cooling-capability perspective.

Otherwise, it looks largely the same as what we're flying from a comms perspective, although we've obviously pulled the comms payloads off. I feel like people think that this is a completely new concept for us, and it's not. It's a logical extension of the satellite technology that we're already using today, with all the connectivity down to Earth, the intersatellite links connecting the different satellites together, and the prop systems.

There's so much of what we're already doing with Starlink today that we immediately get to benefit from on these. I think when people see the picture of one satellite versus another satellite, that aha moment kicks in: “Oh, well, these guys are going to be able to do this very quickly.”

Absolutely. Just maybe help anyone listening conceptualize a rack—an NVIDIA rack, the NVL72 rack.

Yeah.

A server is, I don't know, a couple of pizza boxes. You stack those servers together and make a rack 8 feet high, 3 feet wide, maybe 4 or 5 feet deep. That rack is going to be at the center of the satellite. It may not have those exact dimensions. It may have more or less than 72 GPUs, depending on where the engineers land, but we're going to have that rack.

We've got solar wings. I tried to eyeball it. It looked like they were 200 or 300 feet long, maybe 150 feet long out to each side. It's in a sun-synchronous orbit, and the radiator extends behind it so it's always in the shade and cold. Let's talk about what that design enables—what advantages it enables from a first-principles perspective for orbital compute around power, cooling, cost, and latency back to Earth.

Yeah, it's funny. I agree on all 4, and the one I think we should maybe start with is regulatory, right? The fact that people are already having concerns about “I don't want this data center in my backyard” is clearly a trend—a little concerning trend. The fact that we can bring a clean-energy, good-for-everybody-type solution by just sending these up and having them powered entirely off the power of the sun is pretty amazing.

The fact that the solar cells are in space and get roughly 5 times, if not more, energy per cell than they get terrestrially because they don't have to deal with going through the Earth's atmosphere, and they're in sun-synchronous orbit where the sun is on those cells 24 hours a day. And the fact that you can make them cheaper because you don't need the protection—the glass protection—on these cells because there are no environmental issues when you're up in a vacuum. You start to see these first principles kicking in significantly on the power side.

From a cooling perspective, which is one of the more challenging things already to deal with on the terrestrial side, if you walk into one of our data centers, all that liquid cooling, all the plumbing that we've had to do, and all the engineering around that to really figure out innovatively how to do it becomes very straightforward radiative cooling—simply kind of extending the radiative-cooling solutions that we're doing with Starlink and now doing it with AI compute.

Then, obviously, there's no land lease to deal with. Really, your cost is your satellite and your launch. I look at it as a tech person because I was in semiconductors before space. You kind of have those cost curves traditionally that you would have, right? As you ramp up in volume and time, your costs go down, and you're benefiting from Moore's law—or maybe soon a different type of Moore's law.

If you look at the satellite, most of the cost is silicon, and so we're ramping up factories and benefiting from silicon cost reductions, process node to process node. Our costs are going to go down over the few years. If you look at the terrestrial solutions, the curve is going the other direction, right? Everything's getting more expensive. The way you're doing the cooling, power bills are not going down, and land and regulatory issues are getting more and more challenging.

I think this all lends to something that's better for the population and a trajectory that's going to be significantly better from a cost perspective as well.

Awesome. Compute is one of the largest markets in the world today. How do you think about the size of that market when you enter it? We'll talk about terrestrial compute, but when will you begin to have orbital compute working? Maybe a range—this is really hard stuff.

Yes, it is. In fact, my engineers have heard me explain why it's a logical transition, and we've done most of the pieces of this already. They say, “You've got to make sure people understand there's a lot of work still to do.” I said, “I understand that.” Some of that is actually scaling this—scaling to the numbers we're talking about, such that you can put up gigawatts a year into space—which is a very hard challenge.

We've just been able to demonstrate scale, whether it's launch scale or whether it's flying and building thousands of satellites a year ourselves. That is definitely going to be one of the challenges here that we deal with. But certainly, we're going to be able to demonstrate capabilities as soon as next year. I think it is near.

And I think, if you list the people who are supporting orbital compute solutions at this point, it's basically a who's who of the tech and AI industry. I won't list other people's names, but certainly Elon Musk is at the top of that list, right? But I think they think it's a lot further away because you can't do it without launch, right? Without the rapid, reusable launch that Starship is going to provide.

I will ask Grok a question. Grok is xAI's general-purpose AI. I'll ask Grok: The inference will happen on an orbital compute satellite and come down via Starlink Direct to Cell to my cell phone—

And it will be amazing.

What a moment it will be whenever it happens.

Yeah.

I will personally find that very exciting.

I completely agree with you. Yes.

Can we conceptualize the scale of this? I think when people hear megawatts and gigawatts, there are maybe a few gigawatt-scale data centers now, but there are only a few here on planet Earth today.

I know you operate one of them. We've been to it together. Give us a sense of the scale of a gigawatt data center and how much power that is relative to—I think a single Blackwell rack consumes the power of 100 American homes, and we're putting in a gigawatt-scale data center with hundreds of these racks together and stitching them together.

What does it mean to have gigawatts of orbital compute per year? I don't think your boss is going to be satisfied with gigawatts.

He is not, nor am I. I think that's why you see us spending so much time getting Starship to rapid reuse. Right now, with the existing first iteration of our satellite and the V3 version of Starship that we just flew, it's roughly 200 launches for every gigawatt that we put up. I would just emphasize that that's the first generation of the satellite and the rocket, so we're just starting down this path.

But even at that point, we're capacitizing for thousands of launches a year right now. You see the 2 launch towers and pads in South Texas. You see the first one almost done at Cape Canaveral and the second one on its way with Launch Complex 37 within the next year. Those first 4 towers alone give you that initial path. We've got other locations that we're starting to talk about, too.

I think that is critical to start down the path of having that capacitization as the rocket itself starts to ramp.

4. Elon Creates An Engineering Culture

Awesome. Before we talk about your AI business, I would love to talk about Elon. Jensen said that bringing Colossus 1, which was the largest coherent cluster of Hopper GPUs in the world, online in 122 days—and these are Jensen's words—was superhuman, and that he thought only Elon could have done it.

Elon and the talented team that works with him—I would just love to hear from you: What is it like working for Elon? And how does Elon being so involved in the engineering, plus this incredible sense of mission, play into the talent of the company? Why could xAI, now part of SpaceX, do something like that?

I vividly remember that a lot of people thought it was impossible and that it couldn't be done, and they weren't exaggerating. I think it wasn't until Jensen said that that people believed it had been done, because it seemed so impossible. So just talk about working for Elon and the talent.

Yeah. I've got 15 years of working for him. It is always special. It's one of the reasons I'm still here, to be honest with you. He creates a culture where, like we talked about, you set out with these initially audacious goals, and then, step by step, you realize that you're marching toward something that is absolutely achievable.

The initial schedules themselves are sometimes challenging to make, but what's amazing is that he does everything he sets out to do. I think that's the key here. In fact, if I think about going to Mars, for example, when I first got here in 2011, 15 years ago, people would be rolling their eyes when we talked about Mars and being a multiplanetary species. Nowadays, when we say that, literally the response is, “What year?” It no longer even sounds audacious.

I think what Elon has done a masterful job of, along with probably a lot of other things, is setting out these targets and then creating a fantastic business model around each piece of IP that you need for that end goal.

It reminded me of when I was at Broadcom. We would want to do a new SoC, and if we were missing pieces of IP, we would do an acquisition, right? You grab those pieces because you wanted to win the next socket at Apple or somewhere.

Here, if we need the next piece of IP, we're doing it organically for sure. We had to get to orbit, then we needed reusable rockets, then we needed heavy lift with Falcon Heavy and now Starship. We wanted manned carriage, right? So now Dragon and now Starship. Then we needed comms in space.

You go step by step by step. Then you need rapid, reusable launch. Once you have that, you get to the point where, if you're flying thousands of times a year, when a Mars window opens, you have a fleet of vehicles ready to launch for that month or so. Then you're back to your operational cadence again, right? Because it's every 2 years, and you didn't really have to make a crazy investment to make that capability happen anymore.

Now it's, “Hey, let's get the lunar economy going. Let's learn what it is to live in space,” knowing that we're going to want those capabilities for the moon and then Mars as well. I think that kind of step at a time has removed any concerns that people initially had: How are you going to raise all this money for something, and what's the business model around a multiplanetary species?

Now people don't worry about that, and we all get to stay here focused on the mission. Yet the business model itself is pretty amazing. I think you start to see that same dynamic happening with AI, where we're going to leverage this capability. It's going to ramp up our launch capability along the way.

I think it's going to absolutely give us an extension to our mission related to bringing the human consciousness off-planet, right? Preserving that consciousness in space, and yet it's a great business model for us, step by step.

Just double-clicking on the talent and the engineering environment that Elon creates.

Yeah. That goes into making each one of these pieces of the business—there's IP around it. Each element of Starship, each element of Starlink, is, to me, full of interesting pieces of engineering.

But what I always think about, just in terms of making sure everyone understands the culture of SpaceX, is that my understanding is Elon works on whatever's in the critical path, whatever the hardest problem is, directly with the engineers. He does.

We have a mission to make humanity multiplanetary. We have other AI missions that we'll talk about. This is really exciting. We get the world's best engineers. My observation is that an underappreciated part of what Elon does is that, if I talk to any engineer, one of their favorite times in life was that college engineering course where they were assigned to a team and had to build a remote-control drone or a race car, and they worked all night. It feels to me like that is what Elon creates for these exceptional engineers who are brought in by the mission.

I don't know—you would know better than me—but I believe at one point the Raptor engine was the critical gating factor, and there was a standing meeting every late Sunday or Monday night. It just had to be the engineers who were working on the Raptor engine. Maybe they were just 24 years old, but they were there in a small room working on that problem.

Can you just talk about that environment? I think we've seen that type of dynamic a handful of times, for sure.

And you're exactly right. I think what's amazing to me is that he is in there in the details, working with the engineers on these critical issues. It's also incredibly inspiring to know that your leader is in the trenches with you and working probably harder than any employee that I know.

And I don't know how he keeps that energy level up for the length of time—the decades, literally—that he's been able to do it. It's pretty amazing. But I absolutely would tell you I've witnessed those technical challenges, or technical discussions over the challenges that we face, on a regular basis. The Raptor engine was one of them, for sure, but there have been many others along the way. And he's going toe-to-toe with the technical leaders on those items. It's just incredibly inspiring.

And now we have that photographic evolution from Raptor 1 to Raptor 3. It seems like all the Raptors did great on the latest Starship.

They did. It was pretty amazing.

Yeah. But if the first mission that SpaceX started with was “make humanity multiplanetary,” I think it now has other missions. Because you acquired xAI, whose mission was “maximally truth-seeking AI.” X has a mission dedicated to free speech, and I think these missions resonate with people. Maybe these all come together, and you've said—I’ve heard—“expanding the light cone of consciousness,” bringing the light of consciousness to the stars.

This is kind of the overarching mission now that ties them all together. So I would love to hear from you, Bret, how you think about the AI business, and then we'll riff on some of it.

Yeah. No, listen, I appreciate it. I think that going into the AI business has a couple of different, incredibly rewarding elements. Certainly, there's a financial opportunity, and there's the opportunity to give us the reason to really ramp up to thousands of launches from a financial perspective. But I think it's also critical that we have the ability to make sure that the AI model that people start to really rely upon is a truth-seeking model. And so having the real-time content from X integrated into our solution, I think, is important to us and will, in the end, be a huge differentiator related to our AI solutions.

Awesome. Well, let's talk about the business. The first part of the business that I think is probably most accessible to people is Grok.

We've got Grok. We have an enterprise API business. We now have Grok Build. Yep, which puts a harness around it. We have terrestrial compute here on Earth. Let's walk through each of them.

Well, it's probably not a surprise that we have a very diversified AI business because we've done the same thing with our other businesses. If I think back to our space business, the fact that we launch commercial missions, Space Force missions, and NASA missions really gives us diversification. And then you look at connectivity with broadband and direct-to-cell, and the fact that within broadband—we didn't even talk about enterprise and government—but that's certainly a big piece of our broadband business as well. And so now you talk about AI, and again, a very diversified business.

Certainly, hosting others was a key item that we talked about related to, even in the next couple of years, with orbital compute. But there's nothing better to prove out that business model than demonstrating it right now with the Anthropic deal that we announced. We can do that right now with our terrestrial data centers.

So certainly hosting others, as well as building our own model, both enterprise and consumer. To your point, certainly on the consumer side, the differentiation of our real-time data from X is huge. By the way, even X, I think we can do far better related to the optimization of the ads engine. We're investing in that technology right now, and you saw us not be happy with where we were on the enterprise solution, on coding specifically, so we quickly went out and did the deal with Cursor and brought in one of the industry-leading solutions, even on that front, to move us faster and bring in all of that enterprise data.

I think in all of those areas you're seeing huge strides already. But what I also love is that it's now SpaceX AI. If you go up there at Palo Alto, it is SpaceX, and you feel it already. You're seeing the impact of bringing our DNA and integrating the 2 companies as fast as possible; it's been really special. I think you're just in the early days of seeing what we're going to do in AI.

Yeah, just factually, the rate of product releases has accelerated significantly. I do want to double-click on this Anthropic deal. With the Anthropic deal that has been signed and reported on, you're at a $3.75 billion run rate per quarter. So you are 50% larger than the company that I think of as being just outside the top 4. And so you have a top-5 AI infrastructure business, by my math. That's just my napkin math.

But can you talk about that deal, what that was like, how many more deals like that there could be, and how quickly we could bring on terrestrial compute now that we have confidence there's offtake?

Yeah. I think there's nothing better than just demonstrating the capability even before we get to space, that we feel like we're bringing a solution to the AI industry that's desperately needed. I think Elon was very vocal even a year ago that he felt like the constraint was going to be compute and power, and you're already seeing that.

What we have are these dense training clusters, which makes it really premium compute that we're able to offer folks as well. But I also don't want to give people the sense that that means we're backing off related to our own internal solutions. We're absolutely not.

The fact that we can keep our model and our solutions training and doing inference on the bleeding edge related to GB300s in this particular case, and really monetize the other compute in our terrestrial data centers, is huge. I think you're going to see us be able to do more of that, ideally with more folks, over the next whatever time period we want to give it here, even before we get to space. And that's really been part of the key.

At our core, we are builders, right? We are strongly entrenched in building the infrastructure of the future, with AI being one of those key industries. I think you'll likely see more and more folks want to lean to us. There's just a tremendous amount of technical challenge in putting up these, to your point, gigawatt-scale data centers.

So we should expect to see you guys ramp up the expansion of terrestrial data centers to support third parties who want to rent compute from you and your own internal services.

Well, on Grok, I'd love to talk about that. I'd love to talk about what Cursor could mean for the Grok business. One thing that really made an impression on me is the very significant leap in Cursor's own internal model, Composer, after just a few weeks of mid-training reinforcement learning in the SpaceX AI Colossus 2 cluster. You saw a noticeable jump in performance, and it is on the Pareto frontier for models that measure cost relative to intelligence. This is effectively intelligence per unit of cost.

Just talk about Cursor, talk about Grok, and talk about where that business could go.

Well, it's a really great team, first and foremost. When we were doing the diligence and talking to Michael Truell and the team, I think when you find a team that's a good cultural fit—and once that was identified, looking at the data that they were able to bring to the table for us, that was huge. The fact that they're working with over half the Fortune 500 and have thousands of enterprise accounts was certainly a huge opportunity.

And for them, they were, to your point, starved for compute, and you saw immediately what the benefit was for their tool. And so now having that Cursor coding engine, as well as our own Grok LLM and really the harness with Grok Build that's coming out right now, is pretty magical.

Awesome. I'd love to hear just a few words about Terafab. Let's talk about Terafab and how the company's thinking about that.

Yeah, I think Terafab is a really interesting one. Being a semiconductor guy before, we were talking about—and certainly, it's a partnership first and foremost with SpaceX and Tesla. And so now bringing Intel into the fold is huge, right? Having an industry leader of decades in the semiconductor industry brings that know-how to the table.

I look at it as, if you were starting up a foundry business from scratch, it would be so challenging to convince people to tape out their products to you. And so I don't even know that you could start a new company from scratch unless you had a captive customer, like in this scenario, SpaceX and Tesla saying, “We will take every wafer that you can yield out.” And so then it takes that risk away, and really the risk is a capital risk only.

I also think that there are going to be some very exciting elements of having, in my view, the entrepreneur-innovator of our lifetime go and challenge the requirements in a new industry again, sitting with the process engineers that have 50 steps to their requirements list, and Elon pushing on each one related to why that has to exist that way, bringing everything under one roof and taking a completely different approach.

I think we'll yield a pretty amazing result, and Elon has been able to demonstrate that industry after industry after industry. But what I would say is that we are doing this because we're very concerned about the supply chain constraints that might exist very soon, in the next handful of years, related to being able to fab out anywhere else.

If you look at it, when you start talking about NVIDIA, the AI5 chip, or a TPU, all of a sudden you start talking about TSMC, right? And so it isn't that you're able to diversify out from a supply chain perspective if you go one more layer down. Our concern is really, more than anything else, that the supply chain won't be there for us to ramp to many gigawatts. Ideally, 100 gigawatts a year is our target in the years to come, without being able to have a fab. We need to make sure that we have assured supply of silicon.

I would just say, as an American, it's awesome to have more semiconductor manufacturing in America. It's awesome to have more manufacturing jobs here in America being created by SpaceX, and I'm sure you and everyone are super proud of that. It's great for the country, so thank you.

5. Capital Allocation Funds The Buildout

Maybe the last question: SpaceX has been very capital-efficient over the course of its life, but now we have Terrafab, Starship, and orbital compute. We're going to start putting a lot more capital both into the ground and into space, in the most literal way. Can you talk about that transition, and can you talk about how it's going to be funded?

I think capital allocation has really been one of the bigger challenges of this job for 15 years. What I would just say is that we have a pretty good track record of being able to do this. Certainly, the dollars are now bigger than they have been in the past, but I am also incredibly proud of the value creation along the way that we have already been able to achieve. I think that track record is also something that people should consider when they're thinking about the future for us as well.

We really take an approach to capital allocation similar to what you would do in a just-in-time model for manufacturing. You're looking to capacitize Starship with different towers and air-separation units for the fuel, and additional hangars. You're looking at whole new facilities for satellite builds and solar capability, and now you're looking at terrestrial data centers and orbital compute, all happening at the same time. So you really have to map it out quarter by quarter and just not get ahead of yourselves in terms of when you need that capacity in each of these businesses.

And the timing—one might think there was some sound business logic and some very capable minds at work here, because the timing does seem to work out really well. Starship is starting to fly, and Starlink V3 enables—that unlocks—growth for Starlink. Growth for Starlink means cash flow because it's a cash-generative business. That directly unlocks Direct to Cell, which is a new business that will presumably have high margins, and all of that is coming online and ramping right when orbital compute is coming online, when we need to start putting a lot of those racks into space.

That's right. That's right. I wish I could take credit for that, but that's certainly Elon. I just execute on the man's incredible vision, but I agree with you: the timing is pretty magical.

Well, Elon would say execution is everything, and vision is a distant second to execution. Execution is important, but I also just think, maybe to bring us home, that we have a Starship launch that enables everything. We have terrestrial compute—gigawatt-scale data centers. We have an AI model, we have Starlink, we have orbital compute, and we have Terrafab. All of these play into each other and work together. Can you just talk about what that enables? It's a lot of vertical integration.

It is. When people look at it from the outside, they think about all these disparate businesses, but it's actually not that at all. It is the fact that we have a launch platform that was Falcon and now is about to be Falcon and Starship. That really is the enablement for every business you just listed.

We get to just go get better and better at that core launch capability, that core DNA of being the space business of the world. Each of these vertical offshoots comes into play where you have a differentiation because you're able to deliver a better product because of the fact that it's coming from space. In most cases, it's different versions of satellites, or it is back to our core of being incredibly vertically integrated in each of these core pieces of the business, and really bringing infrastructure to the forefront in each of these.

And at every level, your competitors can buy launch from SpaceX.

Yep.

Starlink is available to everyone.

Yep.

You're selling compute terrestrially to competitors. The model can be used by anyone. Absolutely. So while it's vertically integrated in orbital compute, Elon has said that competitors are welcome. While the model is vertically integrated, it is also open at each level, for sure.

And it can be accessed discretely. Then, of course, if each one of these elements is a standalone business on its own, that just gives you more scale, further lowers costs, and plays into the vertical integration.

That's right.

Yeah.

Exciting times. Exciting times.