On the Eve of IPO, SpaceX CFO Details the 'Integrated Five-in-One Super Closed Loop'

As the IPO window draws increasingly near,$Space Exploration Technologies (SPCX.US)$CFO Brett Johnsen has, for the first time, systematically articulated the internal logic of the company’s five core business segments—a synergistic ecosystem anchored by Starship launches and integrating Starlink, orbital computing, ground-based AI computing capacity, and in-house chip manufacturing.

In an in-depth conversation with investor Gavin Baker, Johnsen disclosed that SpaceX’s AI infrastructure business has already achieved annualized revenue of $3.75 billion, making it the world’s fifth-largest AI computing provider. He described the recently completed Starship E3 maiden flight as a “tremendous success” and projected that once the second-stage vehicle achieves rapid reusability, Starship will reduce the cost per kilogram to orbit by an order of magnitude compared to Falcon’s current baseline.

For investors, Johnsen emphasized that the capital logic and sequencing are equally critical: Starship launches Starlink V3 satellites to unlock growth in connectivity services and generate cash flow; this cash flow funds the substantial capital investments required for orbital computing; ground-based data centers validate the business model upfront; and Terafab underpins the entire system’s scalability from the supply chain perspective.

Johnsen summarized this architecture as “all of these play into each other and work together,” stressing that every business layer is open to external parties—competitors can purchase SpaceX launch services, Starlink is available to everyone, ground-based computing capacity is offered commercially, and Grok is also accessible to external users. This strategy not only enhances vertical integration efficiency but also expands the scale base, which in turn lowers costs across all segments.

Starship: The Launch Foundation for All Businesses

Brett Johnsen positions launch capability as$Space Exploration Technologies (SPCX.US)$the foundation of all operations. "If you want to be a space company, you must first ensure you have a reliable pathway into space," he stated. SpaceX is already the industry’s lowest-cost operator on a per-kilogram-to-orbit basis, and Starship aims to improve upon that benchmark by another factor of ten.

The recently completed E3 maiden flight validated the new Raptor V3 engines, modifications to the vehicle’s base structure, and the second stage’s soft ocean landing capability. Brett Johnsen called it a “demonstration of the entire system’s capabilities,” directly bolstering confidence for upcoming flights. He particularly emphasized that Starship’s significance lies not only in cost reduction but also in payload capacity—its 100-metric-ton low Earth orbit capability enables the commercialization of large-mass payloads, a prerequisite for the orbital computing business.

To accommodate the anticipated high launch cadence, SpaceX is constructing two launch towers in southern Texas, with the first at Cape Canaveral nearing completion and the second expected to begin construction within a year. Brett Johnsen stated that the company is building capacity for thousands of launches annually.

Starlink: The Cash Flow Engine Tapping into a $2 Trillion Market

Starlink currently serves over 10 million users, has launched more than 10,000 satellites, and provides coverage in over 160 countries. Brett Johnsen noted that 10 million users represent just the beginning, with potential to grow to hundreds of millions in the future.

He noted that global internet access and cellular connectivity together constitute an existing market of approximately $1.6 trillion, and Starlink’s competitive advantage lies in its speed, low latency, and coverage—'faster, better, and cheaper has always been the winning combination in technology investments.' Compared with terrestrial solutions, satellite networks have extremely low marginal customer acquisition costs once deployed, eliminating the need to dig trenches or lay cables.

The incremental opportunity comes from direct-to-cellphone services. Brett Johnsen stated that next-generation direct-to-device capabilities will deliver 5G-quality performance within two years, enabling truly seamless global coverage—'whether you’re in the heart of a desert or atop the highest mountain.' He expects this new business line to be highly profitable and to sequentially complement the broadband business over time.

Orbital Computing: A New Market Enabled Only by Starship

Brett Johnsen described orbital computing as 'a market that truly requires Starship to become viable,' citing the dual necessity of heavy payload capacity and extreme cost sensitivity. He anticipates demonstrating orbital computing capabilities as early as next year.

On a technical level, he defined orbital computing satellites as 'server racks in space'—essentially scaled-up versions of Starlink V3 satellites augmented with computing modules, equipped with NVIDIA GPUs and radiative cooling metal plates, leveraging existing inter-satellite links and propulsion systems. 'People think this is a completely new concept, but it’s not—it’s a natural extension of our existing satellite technology,' he said.

The cost dynamics are equally favorable: solar panels in space harvest roughly five times more energy than on Earth, with no atmospheric losses; the vacuum environment enables radiative cooling instead of liquid cooling, and there are no land lease expenses. Moreover, as launch scale increases, the primary satellite cost component—silicon—will continue to decline along Moore’s Law. 'The cost curve for terrestrial data centers is trending upward, so the advantage of orbital computing will only grow larger,' said Brett Johnsen.

Terrestrial AI: Validating the Business Model Early

Prior to the large-scale deployment of orbital computing,$Space Exploration Technologies (SPCX.US)$the company is validating its business model through ground-based data centers. A computing infrastructure hosting agreement with Anthropic has generated annualized revenue of $3.75 billion, according to Brett Johnsen, placing SpaceX among the world’s top five providers of AI infrastructure.

Its AI business also includes the proprietary Grok model, enterprise-facing API services, and the recently acquired and integrated code tool Cursor. Brett Johnsen revealed that Cursor already serves more than half of the Fortune 500 companies, and its internal models have shown significant performance gains after several weeks of reinforcement learning fine-tuning on SpaceX’s AI compute cluster. 'Combining Cursor’s code engine, the Grok large language model, and Grok Build—it’s truly magical,' he said.

He emphasized that one key differentiator of SpaceX’s AI business is real-time data from the X platform. 'We must ensure that the AI models people genuinely rely on are truth-seeking models. The integration of X’s real-time content is a critical differentiating factor for our AI products.'

Terafab: The Final Piece of the Supply Chain Sovereignty Puzzle

Brett Johnsen candidly stated that the 'fundamental driver behind Terafab is concern over the supply chain.' He pointed out that whether it’s NVIDIA chips or TPUs, they all ultimately depend on Taiwan Semiconductor. 'If our goal is to deploy 100 gigawatts of computing power annually, we must ensure a stable chip supply unaffected by tariffs.'

Terafab is$Space Exploration Technologies (SPCX.US)$a joint project with Tesla, and Intel has also joined, contributing decades of industry experience. Brett Johnsen believes that having SpaceX and Tesla as anchor customers—both committed to absorbing all output—can eliminate the most challenging demand-side risk associated with building a new wafer fabrication plant. "What remains is only capital risk," he added.

He also expressed hope for applying first-principles thinking in the semiconductor field: 'Walk into a room with engineers who follow a 50-step process and ask why each step is necessary—then completely rethink how to do it. Elon has already demonstrated the power of this approach across one industry after another.'

Below is the full transcript of the interview:

Gavin Baker: Launch capability underpins everything you do. You’ve been a decade ahead of the world in reusability, and Starship is designed for rapid reuse, enabling larger payloads to be delivered to orbit at significantly lower cost. Could you discuss the importance of launch and where Starship fits into this picture?

Brett Johnsen: I believe that if you want to be a space company, you must first guarantee reliable access to space. That’s why we built our launch capability on a solid foundation from day one. Elon has been singularly focused on reducing the cost of accessing space to an unprecedented level—we currently offer the lowest cost per kilogram to orbit in the history of the industry, and with Starship’s rapid reusability, we aim to achieve another tenfold cost reduction.

Launch is absolutely central to everything we do—it’s the enabler for all our other businesses, whether it’s Starlink, the upcoming direct-to-cellphone service, or now AI computing. When discussing SpaceX, you must start with launch capability. Starship represents the next level because we’re tackling the holy grail of rocketry: rapid reusability. Recovering the first stage was already revolutionary—we accomplished that a decade ago with Falcon, which transformed the industry. Now, however, we’re flying the largest rocket ever built, with the goal of operating it like an aircraft. This is an entirely different paradigm, but I believe it’s precisely what will catalyze the entire space industry and finally deliver the space age we dreamed of as children by the 2030s.

Gavin Baker: Where does the Starship program stand today? The last launch used the E3 version of both the ship and booster. Frankly, it felt like a high-stakes gamble ahead of the IPO, but from my perspective—and that of the SpaceX team—the launch was highly successful. What did we learn? And how close are we to achieving rapid reusability?

Brett Johnsen: Last week’s inaugural flight of E3 was a tremendous success for us. We demonstrated the full system’s capabilities: the brand-new Raptor V3 engines, all the modifications at the vehicle’s base, every operational adjustment, and finally the second stage’s soft splashdown. This gives us great confidence—not just for the coming years, but even for the next few flights. It’s incredibly exciting because this is exactly the platform we’ve described as enabling all our other ventures.

A rocket capable of delivering 100 metric tons of payload into low Earth orbit is critical to everything we do next.

Gavin Baker: How will the cost per kilogram for this 100-ton capability compare with Falcon?

Brett Johnsen: Once we’re able to recover the second stage back to the launch pad and begin rapid reuse—likely within the next few years—I believe you’ll see a tenfold reduction in cost per kilogram to orbit compared to today’s Falcon levels. This represents a massive leap forward for multiple segments of our business.

Gavin Baker: I like to think of launch as the foundational enabler for a wide range of applications, much like how Microsoft has an operating system. Launch is the foundation, and the first—and most widely recognized—application is Starlink. Could you give us an intuitive sense of the scale and growth trajectory of this business?

Brett Johnsen: It’s incredibly exciting. Consider this: it’s only been six years since we truly began mass-producing and launching satellites into space. From that point to now—over 10 million customers, more than 10,000 satellites launched, and coverage across more than 160 countries—this demonstrates that we’re providing a capability the world genuinely needs.

Every month, we send out an email highlighting the number of communities around the world where we’ve made a positive impact—whether it’s Indigenous communities in Canada, local neighborhoods in Brazil, or schools in Africa that previously had no internet connectivity at all. Bringing Starlink to the entire globe is profoundly meaningful.

This also extends our mission: while our ultimate goal remains making humanity a multiplanetary species, we now have the ability to connect another 3 billion people on Earth who currently lack internet access, truly bridging the digital divide. I believe 10 million customers is just the beginning; we can scale to hundreds of millions globally, as satellite-based coverage is inherently more efficient than building terrestrial infrastructure across countless disparate locations.

Gavin Baker: As a seasoned telecommunications industry analyst, let me add a few observations. The telecom market is enormous—internet access accounts for roughly $800 billion, and cellular connectivity another $800 billion, totaling approximately $1.6 trillion. What’s remarkable about this industry is that historically, there has never been a truly differentiated or disruptive product, because all operators essentially use the same base stations, the same right-of-way infrastructure, and the same equipment, resulting in highly commoditized offerings—whether fixed-line, long-distance, or cellular.

I’m a serious gamer—I take gaming as seriously as others take golf—which is precisely why you hired me. My grandfather used to say, 'Age and cunning always beat youth and skill.' In gaming terms, that means having the best GPU and the lowest latency. I’ve tested Starlink everywhere I’ve been, and it’s consistently the fastest with the lowest latency. Could you discuss how this product differentiation helps you scale from 10 million users to hundreds of millions?

Brett Johnsen: Beyond broadband, one particularly promising area is our upcoming direct-to-device service. Our next-generation direct-to-cell capability will deliver 5G-quality connectivity within two years. What we’re about to introduce is truly unique: whether you’re deep in a desert or atop the highest mountain, your phone will have signal coverage—no dead zones, with genuine global roaming. People are willing to pay a premium for that, and even more so if the price is comparable.

Regarding broadband, low latency is indeed critical, as you mentioned. High-speed, low-latency connectivity covering virtually anywhere is simply unattainable with terrestrial solutions. I believe it’s becoming increasingly difficult to justify large-scale terrestrial deployments, given that Starlink already exists.

You’re now also starting to see applications in aviation—both United Airlines and American Airlines have announced integration with Starlink. Once passengers experience low-latency, high-speed internet on airplanes, they’ll naturally expect the same quality at home or in the office.

Gavin Baker: Every time a natural disaster strikes, SpaceX rapidly deploys Starlink terminals and activates them free of charge—an intervention that has undoubtedly saved lives.

Brett Johnsen: Yes, absolutely.

Gavin Baker: Beyond Starlink and orbital computations, what other Starship-enabled applications or businesses excite you the most?

Brett Johnsen: Both connectivity-related businesses will continue to grow. When people talk about space, they often focus on entirely new markets that will emerge in the future—whether point-to-point Earth transport (e.g., reaching Singapore in 30 minutes) or a lunar economy. I believe these will indeed materialize. However, what’s underestimated is that within connectivity alone, there’s already an existing terrestrial market approaching $2 trillion in scale—and you can deliver a superior product in this existing market even before considering any interplanetary scenarios.

And it’s not just about a better product—it’s also about cost advantages. You don’t need to dig trenches or lay fiber, nor do you require massive upfront infrastructure investments. Once the satellite network is deployed, customer acquisition costs are largely limited to shipping a user terminal. Faster, better, and cheaper—that combination has consistently been the winning formula throughout my years of technology investing.

The next major new market, in my view, is AI computing. This is the market that truly requires Starship to become viable, because it involves high-mass payloads and is extremely cost-sensitive.

Gavin Baker: Right—Starship isn’t just about cost—

Brett Johnsen: Exactly—it’s the combination of high-mass payloads and low cost; both are essential. And while you can already see the other two businesses taking off, AI computing will represent a significant incremental opportunity.

If you consider the impact of Starship on our broadband satellite constellation, the V3 satellites we are about to launch via Starship will deliver per-launch capacity equivalent to 20 times that of today’s Falcon launches. This represents a massive enabler for both broadband services and the upcoming direct-to-cell 5G offerings.

Gavin Baker: I’d like to help everyone truly understand what orbital computing is, because after many interactions on X, I’ve realized that many people picture a Pentagon-sized building floating in space—but that’s not it at all. Orbital computing is essentially racks in space. Could you describe what these satellites actually look like?

Brett Johnsen: I really like the phrase 'racks in space'—I might borrow that. In fact, the first time someone internally mentioned a 'space data center,' I was just as confused: 'Wait, how do all these components connect?' They told me, 'Brett Johnsen, it’s essentially just another constellation.' And that’s exactly it.

Virtual networking has been a core technology in networking for many years, and the same concept applies here—satellites are virtually interconnected, with each satellite functioning like a rack. These satellites essentially resemble scaled-up versions of the Starlink V3 satellites we’re about to launch, featuring larger solar arrays and a computing module stacked on top—starting with NVIDIA GPUs, plus a large metal radiator plate for thermal radiation.

From a communications architecture standpoint, it’s fundamentally the same as the Starlink system we already operate—the only difference is that the communications payload is replaced by a computing payload. People may think this is an entirely new concept, but it isn’t—it’s a natural extension of the satellite technology we’re already using today, leveraging our existing inter-satellite links, propulsion systems, and all other foundational elements. When people see side-by-side images of the two types of satellites, they often have that 'aha!' moment—and these teams can genuinely move very quickly on this.

Gavin Baker: Let me help everyone visualize this more concretely: an NVIDIA NVL72 rack—roughly the width of several pizza boxes stacked together—forms a unit about 2.4 meters tall, 1 meter wide, and 1 to 1.5 meters deep, housing 72 GPUs. This rack sits at the center of the satellite, with solar arrays extending roughly 50 meters on either side. The satellite operates in a sun-synchronous orbit, and the radiator extends rearward, remaining in permanent shadow to stay cool. Based on first principles, what advantages does this design offer in terms of energy, thermal management, cost, and ground latency?

Brett Johnsen: I agree with those four dimensions, but I’d like to start from an angle that’s often overlooked: regulation. People are already pushing back against 'Not In My Backyard' data centers—a trend worth noting. What we can offer is a clean-energy solution powered entirely by solar energy, which benefits everyone.

Solar panels in space harvest roughly five times—or even more—energy than those on Earth, as sunlight doesn’t have to pass through the atmosphere. Moreover, in a sun-synchronous orbit, the panels receive continuous sunlight 24/7. And because there’s no environmental degradation in the vacuum of space, the panels don’t require protective glass layers, significantly lowering manufacturing costs. This represents a clear first-principles advantage on the energy front.

On thermal management—which is one of the most challenging issues for terrestrial data centers; if you walk into one of our facilities, you’ll see extensive liquid cooling piping—in space, it simplifies to passive radiative cooling, a natural extension of the approach we already use on Starlink.

Additionally, there are no land lease costs, so your expenses essentially boil down to the satellite itself plus launch costs. From a technical perspective, having previously worked in semiconductors, I’m accustomed to seeing cost curves decline with volume, driven by Moore’s Law. The primary cost of satellites is silicon, and as we scale up factory output and advance across process nodes, costs will continue to fall. In contrast, terrestrial solutions face rising cost trajectories—cooling is getting more expensive, electricity prices aren’t decreasing, and land and regulatory costs keep climbing. Therefore, the cost advantage of orbital computing will only widen over time.

Gavin Baker: Computing is one of the largest markets in the world today. What is your view on the scale of this market and the timing of SpaceX’s entry into the orbital computing market?

Brett Johnsen: Every time I tell my engineers that this is a natural progression—that we’ve already done most of the work—they respond, 'You have to make it clear to everyone that there’s still an enormous amount of work ahead.' I understand that. Launching gigawatts of computing capacity into space annually is a massive challenge, particularly when it comes to scaling. But we have indeed demonstrated our ability to scale—both in launch operations and in autonomously manufacturing thousands of satellites per year.

We could demonstrate orbital computing capabilities as early as next year. Look at the current list of supporters for orbital computing initiatives—it’s essentially a who’s who of the AI tech industry, and Elon Musk is certainly among them. They consider this vision distant primarily because there hasn’t been a suitable launch platform—none of this is feasible without the rapid reusability enabled by Starship.

One day, I’ll ask a question using Grok—SpaceX’s general-purpose AI—and the reasoning will be performed on an orbital computing satellite, then transmitted directly to my phone via Starlink. That will be a truly astonishing moment.

Gavin Baker: To help people visualize the scale: there are currently only a handful of gigawatt-scale data centers on Earth. A single Blackwell rack consumes as much power as 100 average U.S. households, and a gigawatt-scale data center connects hundreds of such racks together. What does it mean to launch gigawatts of computing capacity into space every year?

Brett Johnsen: Moreover, Elon isn’t satisfied with just gigawatts—and neither am I. That’s precisely why we’re investing so much effort into achieving rapid reusability with Starship. With our current first-generation satellites and the recently flown V3 version of Starship, launching one gigawatt of computing capacity requires approximately 200 launches. And that’s just with the first generation of both satellites and rockets. That’s why we’re now building capacity for thousands of launches per year—you can see the two launch towers and pads under construction in southern Texas, the first one at Cape Canaveral is nearly complete, and the second will begin within a year. Just these first four launch towers provide an initial pathway, and we’re already discussing additional launch sites.

Gavin Baker: Before we discuss the AI business, let me ask about one detail: Jensen Huang described bringing Colossus One—the world’s largest Hopper compute cluster at the time—online in 122 days as a 'superhuman achievement,' adding that only Elon could pull it off. What’s it like working under Elon?

Brett Johnsen: I’ve worked with him for 15 years, and every experience has been extraordinary—that’s one of the reasons I’m still here. He fosters a culture where, as we’ve discussed, you set goals that initially seem wildly ambitious, but step by step, you realize you’re moving toward something entirely achievable.

Take going to Mars as an example: when I joined in 2011, mentioning Mars would earn you skeptical stares. Now, when we talk about it, people ask, 'Which year?'—it no longer even sounds like an audacious goal. One thing Elon has done exceptionally well is build a complete business model around every piece of IP required to reach the ultimate objective: to get to orbit, you need reusable rockets; to achieve heavy lift capacity, you develop Falcon Heavy and Starship; to enable crewed missions, you create Dragon and the crewed version of Starship; for space-based communications, you launch Starlink… Each step has its corresponding business model. Once rapid reusability is achieved, you’ll have an entire fleet of vehicles ready to launch whenever the Mars window opens, returning afterward to regular operations.

With launch windows occurring every two years, each mission no longer requires frantic, dedicated investment. Then comes the lunar economy, learning how to live in space, and building capabilities for the eventual journey to Mars. This step-by-step approach has eliminated all initial concerns about questions like 'How do you finance a multiplanetary species?'—now, nobody worries about that anymore, and we can focus entirely on the mission itself.

And the same dynamic is now unfolding in the field of AI—we will leverage this capability to enhance our launch capacity, extending our mission to take human consciousness beyond Earth and preserve it in space, all while representing an exceptional business model.

Gavin Baker: My understanding is that Elon personally engages in engineering work along the most critical path, directly collaborating with engineers to tackle key challenges. The Raptor engine was once the most critical bottleneck, and it’s said there were fixed late-night meetings on Sundays or Mondays that only engineers working on Raptor could attend—even if they were just 24 years old—sitting together in a small room to solve problems.

Brett Johnsen: You’re absolutely right. What amazes me is how deeply he dives into the details, working side by side with engineers to resolve these pivotal issues. Knowing that your leader is fighting alongside you—and likely working harder than anyone else—is incredibly motivating. I truly don’t know how he’s maintained that level of energy for decades. I’ve personally witnessed many of these technical problem-solving sessions myself; the Raptor engine was one example, but there have been numerous others along the way. His direct engagement with technical leaders is genuinely inspiring.

Now we’ve seen the iterative evolution from Raptor 1 to Raptor 3, and the latest Starship flight demonstrated outstanding Raptor performance.

Gavin Baker: SpaceX’s original mission was to make humanity a multiplanetary species. Now, it has added new missions: acquiring xAI, whose purpose is to pursue truth maximization, and X, which champions free speech. These missions converge into a grander narrative—carrying the light of consciousness to the stars. Could you talk about the AI business?

Brett Johnsen: Entering the AI business offers several exciting dimensions. While there are certainly financial opportunities and commercial imperatives driving us toward thousands of launches per year, I believe the more critical point is this: we must ensure that the AI models people truly rely on are models committed to pursuing truth. Integrating X’s real-time content into our solutions is, in my view, essential and will ultimately become a major differentiator for our AI products.

Gavin Baker: Let’s discuss specific offerings. We have Grok, an enterprise API, now Grok Build, and ground-based computing capacity—could you walk us through each?

Brett Johnsen: We operate a highly diversified AI business, consistent with our approach across other domains—our space business includes commercial launches, Space Force missions, and NASA contracts; our connectivity business spans broadband, direct-to-cellphone services, and enterprise and government segments within broadband. Our AI business follows the same diversified strategy.

Of course, providing hosted computing for external parties is part of our future orbital computing ambitions, but the best way to validate this today is by starting on the ground—we’ve just announced an agreement with Anthropic to provide computing power in our terrestrial data centers. This effort currently generates $3.75 billion in annualized revenue, making us the world’s fifth-largest AI infrastructure company, trailing only the top four.

We are also building our own models for both enterprise and consumer markets. On the consumer side, real-time data from X represents a significant competitive advantage. We’re also optimizing the ad engine technology powering X. In enterprise programming, we recognized our progress wasn’t fast enough, so we acquired Cursor to integrate its industry-leading coding solution, accelerating development and bringing in substantial enterprise datasets.

It’s SpaceX AI now—step into the Palo Alto office, and you can feel this convergence. The speed at which we’re integrating the DNA of both companies is exhilarating; our story in AI has only just begun.

Gavin Baker: Regarding the Anthropic agreement, one detail stood out to me: Cursor’s own internal model, Composer, achieved a significant performance leap after just a few weeks of reinforcement learning fine-tuning on the SpaceX AI Colossus 2 cluster—it’s already on the Pareto frontier, meaning it delivers state-of-the-art intelligence per unit cost.

Brett Johnsen: Cursor is an excellent team with strong cultural alignment. Once we found that fit, the data they brought was incredibly valuable—they serve more than half of the Fortune 500 and have thousands of enterprise customers. For them, access to compute power was urgent, and you could immediately see how much their tools improved with that added capacity. Now, combining Cursor’s code engine, our own Grok large model, and the soon-to-be-released Grok Build is truly magical.

Gavin Baker: Finally, let’s talk about Terafab.

Brett Johnsen: Terafab is fascinating. As someone who previously worked in semiconductors, I see this first as a collaboration between SpaceX and Tesla—and now Intel has joined, bringing decades of semiconductor industry experience.

My view is this: if you start a wafer foundry from scratch, it’s extremely difficult to convince other companies to tape out with you—you almost can’t launch a new foundry business unless you have anchor customers like SpaceX and Tesla who commit upfront to taking every wafer you can produce. That eliminates demand risk, leaving only capital risk.

Moreover, it’s very exciting to see some of the most entrepreneurial and innovative minds of our era applying their first-principles approach to the semiconductor industry—walking into facilities with 50-step process flows, asking why each step is necessary, rethinking everything, and rebuilding the whole process from the ground up. Elon has repeatedly demonstrated the power of this approach across industry after industry.

But our fundamental motivation for Terafab stems from supply chain concerns. Whether you’re talking about NVIDIA chips, AI5 chips, or TPUs, you inevitably end up relying on Taiwan Semiconductor. If our goal is to deploy 100 gigawatts of compute capacity annually, we must ensure a stable chip supply that isn’t subject to tariffs.

Gavin Baker: One last question: SpaceX has always maintained exceptional capital efficiency, but now with Terafab, Starship, and orbital computing, you’re literally going to pour massive amounts of capital into both the ground and space. How will this shift unfold, and where will the funding come from?

Brett Johnsen: Capital allocation has always been one of the biggest challenges of this job—and it has been for the past 15 years. Today, the scale of capital is larger than ever, but our track record speaks for itself—we’ve already created enormous value, and I hope people keep that record in mind when considering our future.

We employ a capital allocation approach akin to the 'just-in-time' principle in lean manufacturing—you must plan for different Starship launch towers, air separation units, and factories; you must plan for entirely new satellite manufacturing and solar facilities; and you must also plan for ground-based data centers and orbital computing. All of these are advancing simultaneously, requiring precise quarterly planning without getting ahead of actual demand.

The timeline reveals a very clear logical chain: Starship begins launching Starlink V3 satellites, which unlocks Starlink’s growth—and Starlink itself is a cash-generating business. This growth, in turn, unlocks direct-to-cellphone service, a new business expected to deliver high margins. And all of this peaks precisely when we need to launch a massive number of server racks into orbit.

Gavin Baker: The timing aligns perfectly—

Brett Johnsen: I wish I could say it was my doing, but it’s Elon’s. I’m just executing this person’s breathtaking vision.

Gavin Baker: Elon has said execution is everything—vision comes second—but execution is equally critical. Let me conclude with this: We have Starship launches enabling everything else, ground-based gigawatt-scale data centers, AI models, Starlink, orbital computing, and Terafab—all interlocking and operating in synergy. Could you elaborate on what this represents?

Brett Johnsen: From the outside, people may see a collection of disparate businesses, but that’s far from the truth. At the core, we have a launch platform—first Falcon, soon Falcon plus Starship—that enables every single business you mentioned. Our core launch capability continuously evolves, which is embedded in our DNA as a global aerospace company. Every vertical business line gains a differentiated advantage because our products originate from space, allowing us to deliver superior solutions.

And at every layer, we remain open: competitors can purchase SpaceX launch services, Starlink is available to everyone, ground-based computing capacity is being sold externally, and Grok is accessible to anyone. Thus, despite deep vertical integration in orbital computing, each layer remains independently accessible. Moreover, every individual business stands strong on its own, which further scales operations, drives down costs, and in turn reinforces the advantages of vertical integration.

This truly is an exciting era.

Want to select stocks or get a stock diagnosis? Curious about the opportunities and risks in your portfolio? For all your investment questions,Just ask Futubull AI!

Editor/melody