SpaceX's Ambition Revealed: AI Computing Power Ventures Directly into Space, Mars Is Just the Starting Point

SpaceX, built on Starship and Starlink, upgrades space from a launch site into an industrial platform for energy, computing power, and the expansion of civilization.
(Recap: Is investing in SpaceX IPO at $135 per share worthwhile? SpaceX IPO date, stock price, and how to buy SPACEX(PRE) and SPCX)
(Background: ProShares will roll out a “2x leveraged long ETF” on the day of SpaceX’s listing—an all-time largest IPO celebration is about to begin)

Table of Contents

Toggle

• Compensation Plans and the Endgame Narrative
• The Space Blueprint of Civilizational Fiction
• Mars and Lunar Industrialization
• Lowering Orbital Entry Costs
• From Warehouses to Global Dominance

This article is compiled from analysis by Marc Andreessen. He discusses SpaceX’s unique competitive strengths and future narrative possibilities.

What makes SpaceX special is that it folds rocket reusability, satellite internet, AI computing power, robotics, semiconductor manufacturing, and lunar industrialization into a single roadmap—forming a cross-industry, cross-cycle infrastructure layout.

The author’s key judgment is that SpaceX’s long-term value depends on whether it can continuously reduce the marginal cost of entering space, and shift space from science and defense scenarios into new industrial spaces for energy, computing power, and manufacturing.

The article opens with Musk’s extreme compensation plan for SpaceX: he will only receive real returns if the company’s valuation reaches $7.5 trillion and a permanent human city of one million people is established on Mars, or if it operates data centers in space consuming 100 terawatts of power. This design itself reveals SpaceX’s endgame narrative: launching satellites more cheaply is only the beginning; the true objective is to push energy, computing power, manufacturing, and human habitat spaces together beyond Earth.

At present, AI infrastructure is running into bottlenecks in power, land, approvals, and supply chains, and the marginal costs of traditional on-the-ground expansion are rising. If computing expansion begins to look for energy sources and deployment space outside Earth, the boundaries among aerospace companies, cloud providers, energy firms, and semiconductor manufacturers will be redrawn.

Viewed within this framework, for SpaceX the key might no longer be how many rockets it launches today, but whether it can upgrade “entering space” into an industrial platform that carries energy, computing power, manufacturing, and the expansion of civilization.

Of course, this narrative depends heavily on Musk’s judgments about technological progress, cost curves, and organizational execution—and it also clearly carries an investor’s perspective. Readers are better off treating it as a projection about future industry structure: its value lies in helping us understand space, AI, and energy—three previously dispersed topics—through the same cost curve, and in pointing to where the next-generation industrial platform may emerge.

Elon Musk’s compensation plan for SpaceX is built around two goals. The first reward unlocks when the company’s valuation reaches $7.5 trillion and a permanent human settlement of at least one million people is established on Mars. The second reward unlocks when SpaceX operates data centers in space—where those data centers consume at least 100 terawatts of power. That scale exceeds the total electricity consumption of all data centers on Earth by 1,000 times. If neither goal is achieved, Musk will receive nothing, aside from the $54,080 annual salary he has taken since 2019.

Board members who signed this compensation plan have, over the past two decades, watched Musk make one “impossible-sounding” prediction after another for SpaceX—only to see them come true one by one. He once said SpaceX would send humans into orbit, and before that, no private company had ever done it; today, SpaceX has made transporting NASA astronauts routine. He said SpaceX would land and reuse orbital-class rockets, while at the time the entire industry treated boosters as expendable; since then, SpaceX has completed hundreds of similar recoveries. He said that when satellite internet was close to becoming the graveyard of near-bankrupt companies, the business could be worth hundreds of billions of dollars; today, Starlink’s revenue has grown from zero to $11.4 billion in just a few short years. These predictions are often aggressive on timelines, but almost never wrong on direction. The original direction—written into the company’s mission back in 2002—was to make humanity a multi-planet species. That is why the board ties his compensation directly to this mission itself.

If this mission sounds like science fiction, it’s perhaps because it really is.

Iain M. Banks spent 25 years writing a civilization system called “The Culture.” By most reasonable standards, it might be the best utopian society humans have ever imagined. There, humans live alongside superintelligent AI called “Minds,” which operate colossal orbital habitats like small worlds. The relationship between humans and AI is neither slavery nor competition—it is partnership. Nobody has to work. Nobody goes hungry. Minds handle the massive computational load needed to run the space cities. Humans handle being human, which turns out to be a full-time job in its own right.

SpaceX’s three autonomous, unmanned landing ships—floating platforms where Falcon 9 boosters land at sea—are named after conscious starships from Banks’s novels: “Of Course I Still Love You,” “Just Read the Instructions,” and “A Shortfall of Gravitas.” In a 2023 interview at the UK AI Safety Summit, Musk was asked what a good AI future would look like. He replied: “Banks’ ‘The Culture’ series is, so far, the best imagination of the future of AI. Nothing comes close, and it makes you feel what a fairly utopian—or gradually utopian—AI future might be like.” He has essentially been telling us what he wants to build all along through the names on the hulls of the landing platform ships.

“Of Course I Still Love You” landed the Falcon 9 first-stage booster on April 8, 2016. This was the first successful unmanned ship landing in history, and in that moment, reusable orbital spaceflight stopped being merely theoretical. The ship’s name comes from a conscious starship in Iain M. Banks’s “The Culture” series of novels. (Image: SpaceX)

“The Culture” is not a frictionless paradise. Banks’s novels are full of war, conspiracies, and moral complexity. It is utopian because this civilization has solved survival prerequisites to a sufficient degree, allowing trillions of humans to freely pursue what Banks calls “the things that truly matter in life—sports, games, love, studying dead languages, barbaric societies, impossible problems, and climbing mountains without any safety net protection.”

Such a future rests on four premises. First, the ability to harness a significant portion of a star’s energy output—several orders of magnitude more than the energy produced by today’s human civilization. Second, large-scale physical intelligence: machines that can build, mine, smelt, and repair anything anywhere and do not require human involvement. Third, cheap digital intelligence that surpasses biological intelligence. Fourth, there must be a way to move mass off Earth at low cost, high frequency, and with reliability—because none of the above can be scaled only by extending within Earth itself.

Most analyses of SpaceX work forward from the present: rockets, satellites, contracts, revenue. But if you want to see what is truly happening, a more useful approach is to start from the destination and work backward.

Compensation Plans and the Endgame Narrative

A Mars city. The operational goal is to establish a self-sufficient city of one million on Mars within the lifetimes of people alive today. The difficulty lies in “self-sufficiency.” This means that if Earth stops sending ships to Mars, the city must survive; it needs to produce everything itself: food, water, air, energy, medicine, machinery—eventually, reproducing more humans as well. According to SpaceX’s own calculations, to send one million people and hundreds of millions of tons of cargo there within a few decades will require thousands of Starship flights, with more than ten launches per day during each transfer window. Due to orbital mechanics constraints between Earth and Mars, these windows are only a few weeks long and open once every 26 months.

SpaceX’s rendering of a Mars city. (Image: SpaceX)

A lunar city. A closer and easier dress rehearsal. In permanently shadowed crater pits near the Moon’s south pole, there is ice, and some ridges receive continuous sunlight, making them naturally suitable for base locations. But what Musk talks about is not just a scientific outpost—he imagines something far bigger. He envisions building factories on the Moon that produce AI satellites and launching them one by one into space using mass drivers. The mass driver is also a concept Musk borrowed from science fiction. Fundamentally, it is an electromagnetic launch system that uses the Moon’s environment—only one-sixth of Earth’s gravity and no atmosphere—to industrially hurl solar-powered satellites into deep space. If these satellites are built locally on the Moon, there is also the materials foundation: lunar regolith contains, by weight, about 20% silicon and 10% aluminum, which are precisely the two major raw materials for solar cells and satellite structures. Musk explains: “If you want to go beyond a scale of one terawatt per year, you have to go to the Moon.”

A rendering of SpaceX launching lunar-made AI satellites (data centers) into orbit using mass drivers at an Alpha lunar base. (Image: SpaceX)

Orbital data centers. Musk is betting that in a few years, space will become the most economically attractive place to deploy AI data centers. The bottleneck for AI is energy. Aside from China, energy supply has hardly grown, while AI compute demand is growing exponentially. Power provided by solar panels in orbit is four to ten times that of equivalent solar panels on Earth, depending on ground location and sunlight conditions, because in space there is no atmosphere, no day-night cycle, no clouds, and no seasonal changes. NASA has calculated this decades ago, and now rockets are finally cheap enough to make it real. Musk expects that within five years, SpaceX’s annual AI compute launched into orbit will exceed the total cumulative installed AI compute on Earth. That is why SpaceX merged with xAI in February. Rockets and intelligence are becoming the same problem.

Starship is the transportation tool that makes everything upstream possible. Starship V3 completed its first flight this year. It is the largest and most powerful rocket humans have built to date—taller than a 40-story building, with thrust more than double the Saturn V that once sent astronauts to the Moon. According to NASA statistics, the cost to reach orbit in the past has been about $18,500 per kilogram. In 2010, the first Falcon 9 reduced this cost by about 85%, down to roughly $2,700 per kilogram. In 2018, Falcon Heavy further reduced it to about $1,400 per kilogram. Starship’s design goal is to become the world’s first fully and rapidly reusable spacecraft and further drive costs down to $100–$500 per kilogram. What once cost billions per launch now costs in the tens of millions.

Starlink is the cash engine that helps pay for everything else. According to SpaceX’s IPO filing, the connectivity division—almost entirely composed of Starlink—will reach $11.4 billion in revenue in 2025, up about 50% year over year, with an adjusted EBITDA profit margin exceeding 60%. As of March 2026, Starlink has 10.3 million subscribed users in 164 countries and operates on more than 9,600 satellites. Starlink started as a side project to fill the company’s own launch capacity, but it is now becoming one of the greatest consumer businesses in history. Back in 2019 during due diligence for SpaceX, many told us that this business’s economic model would never work. The technology needed for user terminal antennas had previously been used only for F-22 fighters and Navy destroyers, and had never been mass-produced for consumers. SpaceX’s earliest terminal manufacturing cost was about $3,000, yet it was sold for $499. But they found ways to cut manufacturing costs and proved the skeptics wrong.

Falcon 9 is the workhorse that buys time for everything else. It is the only orbital booster in the world that has achieved large-scale full reusability. A single booster can typically execute more than twenty missions before retirement. In 2025, SpaceX launched 83% of the total global payload mass into orbit. Despite other players having a half-century head start advantage, SpaceX’s total payload mass delivered into orbit is already greater than the sum of all other countries and companies combined.

That is the whole stack, from top to bottom. After generations, “The Culture” sits at the very top. Falcon 9 and Starlink sit at the very bottom, paying the bills for everything today. Each layer makes the next layer possible.

SpaceX CFO Bret Johnsen explains what it looks like from inside the company:

“Musk created a culture: you set a few goals that initially look almost crazy, and step by step you realize you are moving toward something that is completely achievable… like going to Mars. When I first joined in 2011, whenever people talked about Mars and making humanity a multi-planet species, they would roll their eyes. Now when we say it, the reaction really becomes: ‘Which year?’… I think Elon has done an exceptional job in setting these goals and building an excellent business model around every key technical asset required to achieve the ultimate goal.”

Musk initially did not want to start a rocket company. In 2001, a 30-year-old Musk was thinking about what he would do after PayPal. He had long been interested in space, and when he looked into NASA’s plans to land humans on Mars, he was surprised to find that no such plan existed. So he conceived a plan: send a small greenhouse to Mars and transmit the images back to Earth. His idea was that if a green sprout appeared on a dead red planet, it might rekindle public interest in space—and also reignite the willingness at the political level to fund a real Mars program. He only needed one rocket to send the greenhouse over.

Later that year, he traveled to Moscow in an attempt to buy a refurbished intercontinental ballistic missile. This was his first of two trips to Russia. It is said that the negotiations were full of vodka and grandstanding. “We’d go into a small room, with a whole bottle of vodka in front of each person,” Adeo Ressi—Musk’s best friend during his time at the University of Pennsylvania, who also joined the trip—recalled in a 2012 Esquire interview. The Russians did not take Musk seriously. At one point, a chief designer even spat at Musk and his team as an act of contempt. The second trip was in February; Musk asked how much a missile would cost. They said $8 million each. When Musk countered that he would buy two for $8 million, Musk’s aerospace adviser Jim Cantrell remembers the other side saying something like “No, kid,” implying they had no money at all. Musk decided they were not serious about doing business, so he turned around and left.

Cantrell thought the trip was over. On the return flight, he and Mike Griffin, who later became NASA Administrator and was a consultant on the second trip, clinked glasses and toasted as they celebrated finally leaving Moscow. Musk sat in the front row, hunched over his laptop. Then he turned around. “Hey guys,” he said, “I think we can build this rocket ourselves.” He showed them a spreadsheet listing the raw materials required for the rocket—aluminum, titanium, copper, carbon fiber—and the cost of each material. Raw materials cost only 2% of the quoted price. As Musk later put it, “Obviously, you just need to come up with clever ways to combine these materials into the shape of a rocket.”

Within months, Musk decided to risk $100 million to start a rocket company. That was more than half of what he earned after selling PayPal. He founded SpaceX in a warehouse in El Segundo, California. He sent invitations to a founding team of five people. Three declined, including Cantrell and Griffin. The two who agreed were Tom Mueller and Chris Thompson. Mueller later became Vice President of Propulsion and the company’s employee number one; Thompson became employee number two, responsible for operations and production.

“In 2002, SpaceX was basically just carpet and a Mexican wandering band. That’s it,” Musk later joked. “As you can see, I’m a dancing machine.”

Many years later, Musk referred to the principles behind his spreadsheet diagnostic tool as the “idiot index.” If the ratio between a part’s selling price and its raw material cost is very high, then either you’re an idiot or you’re working with idiots. It sounds like a joke, but it is the foundation of SpaceX’s strategy.

Every component SpaceX procures comes with an idiot index calculation. In the company’s early days there was a legendary story featuring Steve Davis. After graduating from Stanford, he went straight to SpaceX and became the 14th employee. His job was to procure an actuator for the Falcon 1 rocket’s upper stage for steering. When he reported that a traditional aerospace supplier quoted $120,000 for the part, Musk laughed, saying the complexity of that component was no more than a garage door remote. Musk gave Davis a $5,000 budget and told him to build it from scratch. As biographer Ashlee Vance describes, Davis spent nine months iterating on the design and ultimately produced a functional actuator that cost only $3,900. When Davis sent Musk the successful technology, Musk replied with a typical brief email consisting of two letters: “Ok.”

The Space Blueprint of Civilizational Fiction

To push the idiot index down toward its theoretical minimum, you must vertically integrate and control the entire process end-to-end. But vertical integration creates fixed costs, and is only worthwhile at high volumes; and in the rocket industry, high volumes mean breaking the industry’s conventional way of operating.

Traditional launch service providers like ULA and Arianespace treat each mission as a custom project. Customers specify orbit, payload, and integration requirements, and launch service providers design a tailored mission around each satellite. This model assumes only a few launches per year, with extremely high cost per mission—making scaled manufacturing impossible.

SpaceX did the opposite. They published a Falcon user guide that explicitly specifies the rocket’s precise parameters and tells customers to design their satellites according to those specs. At the time, this was seen as very radical, and it caused SpaceX to lose some business early on. But it unlocked the manufacturing flywheel.

Standardization and reusability reinforce each other. Because each Falcon 9 is the same, once a booster is recovered, it can be certified and prepared for flight again as a ready product. The first Falcon 9 booster to fly twice did so in 2017. By 2020, each booster could fly five times. By 2021, ten times. Today, the record holder has completed 35 missions. This reusability changed aerospace economics—and it is hard to see how competitors could catch up. In 2021, Musk estimated that under best-case conditions, the marginal launch cost of sending a 15-ton payload to orbit with Falcon 9, excluding overhead management expenses, would be about $15 million. He said this was “about half to a third of the cost of other options.” Today, SpaceX relies on reusable boosters to launch a rocket every two to three days, while competitors can only launch a handful of custom rockets each year.

But SpaceX’s advantage is not only economies of scale, vertical integration, and better strategy. It also comes from speed and culture.

Traditional aerospace companies eliminate uncertainty through analysis. In NASA’s polite phrasing, Boeing’s crewed commercial aerospace project “adopts mature systems engineering methods, investing up-front in engineering research and analysis before construction and testing, in order to mature the system design.” Measure twice, cut once. SpaceX does the reverse. The company manufactures lots of cheap prototypes, pushes them toward failure, learns from failures, and iterates. Starship’s testing program has produced the most spectacular chain of explosions in aerospace history—but each failure is a data point telling the team where reality diverged from the model.

Anyone who has worked in both worlds can see the contrast. Garrett Reisman, a former NASA astronaut who flew two Space Shuttle missions, left NASA in 2011 and joined SpaceX as a senior engineer. He described the prevailing internal NASA view of SpaceX at the time: “They’re a bunch of cowboys; they’re dangerous; they’ll kill people.” What changed his mind was seeing how SpaceX actually works. “What they build in a month, NASA might take a year. We were stunned.”

The clearest example is the Falcon 1 project. Between 2006 and 2008, SpaceX launched four Falcon 1 rockets from a small atoll in the Pacific called Kwajalein. The first three failed, but each failure was different and instructive. The first was a fuel leak. The second was abnormal propellant sloshing. The third was stage separation collision caused by residual engine thrust. By September 2008, the company had money for only one more launch left. Not only was SpaceX near collapse; Musk’s other company, Tesla, was also weeks away from bankruptcy. He had to decide whether to concentrate his remaining PayPal cash into one company or split it between two.

“That was a very difficult decision. In the end, I decided to split the remaining money I had and try to keep both companies alive. But it could have been a terrible decision—and it turned out that both died,” Musk recalled. “I never thought I’d have a mental breakdown, but I was really close.” He couldn’t choose one over the other because, in his worldview, both missions were crucial: Tesla would accelerate the world’s shift toward sustainable energy, and SpaceX would make humanity a multi-planet species. “All available resources had to go into those companies,” Musk’s then fiancée, Talulah Riley, said in the BBC documentary series “The Elon Musk Show.” “He gave me the option to leave. He said, ‘The hardest part is next—you don’t have to stay and go through it with me.’”

In 2006, Elon Musk examines the wreckage of the first Falcon 1 on Omelek Island. (Image: Hans Koenigsmann)

The fourth launch succeeded. In December of that year—just weeks before SpaceX ran out of funds—NASA awarded it a $1.6 billion cargo contract. When NASA called to inform Musk, he was overwhelmed with relief and blurted out, “I love you guys.”

This pattern, formed from rapid failure and rapid correction, later became the culture for every project at the company. It is also the same pattern that allows SpaceX today to iterate Starship between flights, whereas traditional aerospace projects often take years to redesign a spacecraft after an anomaly in a single flight.

This method is superior because, when facing problems you haven’t fully understood, you cannot arrive at a perfect solution merely by thinking. In reality, the only sufficiently effective validator is reality itself—and the key is to reduce the cost of consulting reality enough so that you can consult it frequently.

The above is SpaceX’s iterative loop told through stories, but it also has a written version. Over the past two decades, Musk encoded SpaceX’s approach into a five-step operating process. The company calls it “the Algorithm.” Tim Berry, who worked at SpaceX for ten years and led the Falcon 9 and Falcon Heavy upper stage production teams, said the method has been “poured into our brains.” In his Musk biography, Walter Isaacson presents the standardized version:

First, question every requirement. Every requirement should include the name of the person who made the request. You must never accept requirements from a department—such as Legal or Safety—without knowing exactly which individual proposed it. You need to identify who the specific person is, and regardless of how smart they are, you should question the requirement. Requirements proposed by smart people are the most dangerous, because people are less likely to question them. Then, make those requirements less stupid.

Second, delete every part or process that can be deleted. After that, you may have to add some back. In fact, if you don’t end up re-adding at least 10% of what you cut, it means you didn’t cut enough.

Mars and Lunar Industrialization

Third, simplify and optimize. This step should happen after step two. A common mistake is to simplify and optimize a part or process that shouldn’t exist in the first place.

Fourth, accelerate the cycle time. Every process can be accelerated. But you should do it only after completing the first three steps. Musk has said that at Tesla, he made a mistake: he spent too much time speeding up processes that later he realized should have been deleted.

Fifth, automate. Automation should be last. Tesla’s mistakes at its Nevada and Fremont factories included trying to automate too early—before questioning requirements, removing parts and processes, and clearing out the loopholes.

Most engineering organizations jump directly to step five. They take a process that never should have existed and automate it. SpaceX, however, executes these steps sequentially, every time, in every part of the company. Once this “algorithm” is applied to a piece of hardware enough times, it begins to look unlike anything else in the industry.

SpaceX’s three generations of Raptor engines, from V1 to V3. (Image: SpaceX)

Raptor 3 is the product of a team iterating on the same engine over ten years. It has 22% higher thrust than Raptor 2, is 40% lighter, and no longer needs a heat shield because the plumbing and wiring that previously hung outside the engine are now integrated into the engine’s metal structure through 3D printing. Musk said: “Simplifying the Raptor engine, building secondary flow paths into it, and adding regenerative cooling for the exposed parts—the amount of work required is incredible. It’s getting close to the known physical limits.”

No known engine project in the history of aerospace has iterated this fast. The main engine of the Space Shuttle has basically been the same design for the last thirty years. The RD-180 that powers Atlas V is a derivative of a design from the 1970s. And in less than ten years, SpaceX has already completed a third full redesign of Raptor, with each generation making major advances over the previous one.

The same philosophy applies to people. By mid-2018, Falcon 9’s reusability had entered a reliable rhythm, and Musk shifted his focus to the satellite internet constellation—an initiative that later funded everything upstream. The Starlink team is based in Redmond, Washington, and many senior engineers came from Microsoft, where the development pace was slower than Musk wanted. In June, he flew to Redmond, fired the senior leadership team, and brought in young star engineers from the rocket division. He gave them a year to launch the first batch of operational satellites. This management style is ruthless. Based on media reports about the layoffs at the time, the division seemed like it was imploding from the inside. But 11 months later, in May 2019, the first batch of satellites launched. Musk removed the bottleneck and moved on to the next problem.

This is how he manages everything. In 2018, when Tesla was in “production hell,” trying to scale Model 3 production while burning cash at a life-or-death pace, Musk really moved into the factory. Many years later, he recalled in an interview: “I lived in the Fremont factory and the Nevada factory for three straight years. I slept on the floor under my desk so the whole team could see me during shifts. That’s important, because if the team thinks their leader is somewhere else having fun—drinking Mai Tais on a tropical island—it will hurt morale. The fact that they saw me sleeping on the floor made a huge difference—they knew I was there. It made them give everything.” Later, he turned this into a company-wide rule: the higher your position, the more visible your presence must be.

To find someone whose operating style is comparable to Musk’s as CEO, you have to return to the era of industrialists in the late 19th and early 20th centuries: Henry Ford, Andrew Carnegie, Thomas Watson, Andrew Mellon, Cornelius Vanderbilt. Musk’s operational style is unique because of his relationship with concrete work. It is said that he shows up at each of his companies every week, identifies the biggest problems, and solves them. Doing this for 52 consecutive weeks means each company roughly solves the 52 most critical problems of that year.

An engineer who joined SpaceX from another aerospace company described it like this: “It’s like getting parachuted into a zone of astonishing competence. Everyone around you is absolutely capable at their job.”

SpaceX looks like a company, but a more useful way to understand it is: it is the central node in a constellation of companies. All of these companies are run by the same person, built toward the same long-term mission, and are nearly impossible to split apart. Over the past two decades, Musk has been assembling a set of companies—each one solving a bottleneck that would otherwise limit other companies. And now, they are starting to compound on top of one another.

This February merger with xAI is a snapshot of what SpaceX is becoming. If compute power ultimately enters orbit—which is Musk’s bet—then SpaceX has the most credible path to deploy it at the scale AI needs. Sending mass into orbit and producing intelligence at massive scale may become the two most decisive capabilities in the coming decades—and now they reinforce each other under one roof.

xAI brings Grok, a cutting-edge model. Because it can tap into X’s real-time information streams, it has a unique position in real-time information. xAI also brings the engineers who built supercomputers Colossus 1 and Colossus 2 at speeds that many people consider impossible.

Colossus 1. (Image: xAI)

The construction of Colossus is worth pausing to examine. xAI took over an old factory in Memphis and trained 100,000 GPUs in 122 days. Once the racks started moving in, they had the cluster running in just 19 days. Nvidia CEO Jensen Huang praised Musk: “Starting from a concept, building a large factory with liquid cooling, powering it up, getting permits, and completing it in that time—this is superhuman. As far as I know, only one person on Earth can do this. What they accomplished is unique. No one has done this before. A cluster of 100,000 GPUs was, easily, the fastest supercomputer on Earth at the time. Usually, a supercomputer like that needs three years of planning, then delivery of the equipment, and another year to get everything running.”

A project that would normally take at least four years for other companies—Musk and the xAI team did it in four months.

Lowering Orbital Entry Costs

In May, Anthropic agreed to pay SpaceX $1.25 billion per month for all the compute power of Colossus 1. A few weeks later, in a revised version of its IPO filing, SpaceX disclosed that Google would pay $920 million per month to gain access to 110,000 GPUs—about half of the compute power Anthropic obtains. Combined, these two deals generate annual revenue of about $26 billion, coming only from two customers. And this business did not exist until SpaceX absorbed xAI earlier this year. Chips, power, and land are scarce. SpaceX is becoming one of the few companies with enough AI infrastructure to rent compute power to others while also pursuing its ambition to build leading frontier models itself.

What xAI brings from SpaceX is a longer-lasting solution to the power constraint. Musk believes power will be the bottleneck for AI in the next few years. To produce enough power to meet the intelligence demand he expects, you need grid expansion, new power plants, and multi-year approvals that industry can’t afford to wait for. In his view, orbital solar is the way forward because it is practically infinite. And SpaceX is the only company with a launch vehicle capable of scaling compute into space. Whether he is right is one of the most important open questions in technology. But SpaceX’s IPO filing shows the company is taking this bet extremely seriously: it expects AI to become the largest market for the company to date in the future. Compared with these ambitions, the space business that originally built the company looks almost like a rounding error.

Tesla is another key puzzle piece in this constellation, and the integration between the two unfolds deeply in a different way. Tesla and SpaceX share the same founder, the same talent pool, the same operating culture, and a set of increasingly overlapping technical roadmaps.

Tesla supplies three things to the SpaceX-xAI side of the constellation. First is chips: AI5, AI6, and Dojo3—all designed in-house by Tesla. Musk has already made it clear that these chips are not just for cars; they are components of a larger compute stack for the broader constellation. AI5 is for autonomous driving inference; AI6 targets Optimus and AI data centers; Dojo3, designed to work with the planned AI7, is built for orbital compute. Second is robotics. Tesla’s bet is that Optimus will become the physical AI layer for factories, warehouses, and homes—enabling these scenarios to run without human labor and ultimately serving the lunar and Martian cities Musk envisions. Third is solar energy. Musk has said that Tesla and SpaceX are each building about 100 GW of annual solar cell production capacity to support AI buildout on Earth and in orbit.

Then there is TeraFab. In April, Tesla disclosed that it has begun ordering equipment for a research semiconductor fabrication plant located in the Giga Texas campus. In Tesla’s Q1 2026 earnings call, Musk told investors: “We expect this to be a project of about $3 billion, producing a few thousand wafers per month.” SpaceX, meanwhile, is separately funding the construction of a much larger facility, because no existing fab can scale up at the speed Musk envisions. Once mature, the facility’s designed capacity would be about 1 million wafers per month. And Musk’s envisioned scale is measured in gigawatts. “This isn’t a commitment to what we will do,” Musk said last week. “It’s what we are going to try to do and we believe it has a high probability of being done: by the end of next year, to reach about 1 gigawatt annualized speed for space AI compute. Then, ideally, increase it by an order of magnitude each year. That means two and a half years from now, 10 gigawatts annualized speed. Three and a half years, maybe 100 gigawatts. Then, depending on chip manufacturing progress worldwide and TeraFab’s progress, further scale to 1 terawatt per year, which is 1,000 gigawatts. That’s double the electricity consumption of the United States.”

SpaceX’s TeraFab design goal is to reach 1 terawatt of annual output—roughly twice the current electricity consumption of the entire United States. (Image: terafab.ai)

Comparing Musk to the gilded age magnates does hit some real notes, but it also highlights the differences. Carnegie built the steel empire; Vanderbilt built the railroad empire. Each led a key sector of the industrial foundation of their era. Musk, however, is trying to push forward multiple fields at once—space, energy, artificial intelligence, robotics, tunnels, brain-machine interfaces, autonomous cars—and bend them all toward a single goal that most people consider fantastical. Whether it all ultimately works remains unknown. Many pieces may fail. But this attempt itself has no historical precedent, and may become a preparation ground for a different century.

Before the Space Shuttle retired in 2011, it cost about $54,500 to send one kilogram of cargo into orbit. Once Starship matures, Musk expects this figure to drop to $100 per kilogram. When the cost of entering space falls by more than 500 times, every industry that can, in theory, exist in space will begin to be economically viable. There are many such industries.

The design goal for SpaceX’s Starship and Super Heavy is to return to the launch site after flight and be caught by the launch tower, enabling rapid turnaround and the ability to launch again without needing refurbishment. (Image: SpaceX)

The closest historical analogy might be the transcontinental railroad. Before 1869, traveling from New York to San Francisco took six months by horse-drawn carriage, costing roughly a year’s wages and carrying very real risks of death. After 1869, the journey took only a week. The railroad itself was an engineering marvel, but the real story is what it opened up: meatpacking giants like Sears Roebuck, Swift, and Armour, Standard Oil, and ultimately U.S. Steel, which integrated the industrial empires born during the railroad boom.

If Falcon 9 is like the transcontinental railroad of the space age, then Starship might be the plane upgrade. Railroads opened up an entire continent. The jet age opened up the whole planet. Starship will open up the solar system.

Ever since humans have looked up at the Moon, it has held scientific significance. Now it is starting to hold economic significance, because it is a whole world made of industrial raw materials.

Let’s start with how to move things off the Moon. As described earlier, the Moon has only one-sixth of Earth’s gravity and no atmosphere, which makes mass drivers—rather than rockets—a natural way to transport cargo away from the lunar surface. This will fundamentally change transportation economics. Once in orbit, the marginal cost of delivering finished goods will be mainly determined by power rather than fuel; lunar power is sunlight. A package thrown from the Moon’s surface with a heat shield, reentering Earth’s atmosphere, deploying a parachute, and landing at a recovery site could turn into a routine cargo process. When throughput becomes high enough, marginal costs begin to resemble those of shipping rather than aerospace flights.

Next comes what can be manufactured there. The same lunar regolith that provides the silicon and aluminum needed for solar cells and satellites is also the raw material base for the entire industrial foundation. In the 2030s and 2040s, the picture might look like this: autonomous mining vehicles process lunar regolith day and night; smelting plants produce aluminum and silicon; factories assemble satellites, solar panels, and the chips that drive them. Most industries on Earth have a lunar version waiting to be built—and no single company like SpaceX can build all of these on its own. Those who build “lunar Alcoa,” “lunar Caterpillar,” and “lunar Union Pacific” will become the giants of the 21st century.

Starship HLS is a lunar lander designed by SpaceX for NASA’s Artemis program, aiming to return humans to the Moon’s surface for the first time in more than 50 years and deliver permanent presence delivery base modules near the Moon’s south pole. (Image: SpaceX)

By the 2030s, AI bottlenecks will likely shift from chips to power. The obvious response is to build more solar in Texas or Nevada, but that will hit walls faster than people expect. One terawatt of continuous solar power requires about 1% of the land area of the United States, and approval processes for new utility interconnections can take a year or longer. For xAI to build Colossus in Memphis requires deploying an entire fleet of temporary gas turbines, fighting state government permit and approval processes, and then setting up an independent power hub across the Mississippi before it can go online with 1 gigawatt of power. Scaling that up to hundreds of gigawatts for AI construction is simply not feasible. Even the internal guide vanes and blades inside gas turbines used to provide backup power for solar—orders for those are already queued beyond 2030.

Baker Hughes Frame 5/2C gas turbine generator. Cast guide vanes and blades inside turbines like these are produced by a small number of specialized foundries, and orders for them are already queued beyond 2030. A super-large, cloud-scale data center alone might require dozens of such units. (Image: Baker Hughes)

The solution is to move compute to where sunlight already exists. When Starship achieves daily flights and orbital deployment becomes routine, this will become easier. As the cost curves for rockets, solar panels, and chips continue to decline, economics will improve further. SpaceX CFO Bret Johnsen explains: “We are ramping up factory capacity, and we benefit from falling silicon costs, so our costs will come down over the next few years. If you look at ground solutions, the curve is moving in the opposite direction. Everything is getting more expensive: cooling approaches, electricity bills, land, and regulation.”

A common objection comes from those who hear “space data centers” and imagine a massive building the size of Colossus being launched into orbit—but that is not the case. “It’s roughly the size of a Blackwell rack frame, with solar wings extending 500 feet on each side. You put it into a sun-synchronous orbit, so the solar panels are always in sunlight,” said early SpaceX investor Gavin Baker. “Over the years, I’ve spent a lot of time at Starbase and talked with many SpaceX engineers. I do believe it is the most talented engineering group on Earth, and they are extremely confident they’ve solved this problem.”

AI Sat Mini is built to utilize solar power. (Image: terafab.ai)

In fact, Musk believes AI Sat Mini will be easier to build than Starlink satellites. “You still need some laser links, but you don’t need the extremely complex antennas on Starlink satellites,” Musk explained. “Compared to that, AI satellites are easier to design… AI satellites don’t require magic. A lot of the technology has already been built for Starlink V3. Compared to what we’re already doing, we don’t think this is a particularly hard problem.”

From Warehouses to Global Dominance

He estimates that within five years, SpaceX’s annual launches of AI compute will surpass the total installed compute on Earth. Roughly, that’s 10,000 Starship launches per year—more than one launch per hour, around the clock. By the late 2030s, with lunar mass drivers operational, the terawatt threshold will come into view: equivalent to 1,000 times the compute deployed on Earth in 2030, launching satellites into deep space every few minutes.

Mars trajectories were supposed to begin from this year. Musk had announced in September 2024 that SpaceX would launch five uncrewed Starships to Mars during the transfer window in November 2026, carrying Optimus robots to test landing systems, find ice, and begin building the infrastructure for future crewed missions. In May 2025, he said the odds of meeting that timeline were “50/50,” but earlier this year, things changed.

In an X post on February 8, Musk announced that SpaceX would delay the Mars schedule and shift the short-term focus to building a self-sufficient city on the Moon. The reason: Mars launch windows open only once every 26 months and require six months of flight time; by contrast, the Moon has an accessible window every ten days with only a two-day flight. “This means that compared with a Mars city, we can iterate and complete a lunar city much faster,” he wrote. “That said, SpaceX will still work to build a Mars city and start doing so in about five to seven years, but the top priority is to ensure civilization’s future—and the Moon is faster.”

On the surface, this looks like a pivot. But in reality, it is the moment when the path to a million-person Mars city becomes clearer.

The orbital data center thesis gradually took shape from late 2025 into early 2026, giving the Moon a new role. To reach terawatt-scale orbital compute, you need lunar mining, smelting and manufacturing solar panels, radiators, and satellite structures, and then launch them into orbit via mass drivers powered from the lunar surface. An industrial foundation at this scale requires a permanent population—and permanent populations require a city. This city can be funded entirely by the orbital compute industry while also serving as a dress rehearsal for Mars. Every problem SpaceX must solve to build a self-sufficient Mars city—radiation shielding, life support, in-situ resource utilization, extraterrestrial permanent-population governance, supply chains across gravity wells—are also the problems that must be solved first when building a lunar city. Building a lunar city will let SpaceX learn how to build a Mars city with a much faster iteration loop.

According to Musk’s timeline, the earliest target for a first unmanned lunar landing demonstration is in 2027, and the lunar city will appear within less than ten years afterward. Mass drivers, lunar industrial construction, and lunar manufacturing aimed at orbital compute infrastructure will be advanced in parallel. Then, and only then, comes Mars.

But the hardest part will not be transporting people. The hardest part is building infrastructure on the Mars side that can accommodate them. The lunar dress rehearsal will help. Optimus will help as well. Musk has repeatedly emphasized in his Mars talk at Starbase in May 2025 that early unmanned Starship will carry Optimus robots. They will scout resources and begin building the infrastructure for humans to arrive. The company is building a production line in Fremont that can produce 1 million units per year, and also building a 10 million units per year line at Giga Texas. These robots are still in early production stages and have not yet completed truly meaningful practical work inside Tesla factories. But the production capacity launching in the next two to three years will be crucial for guiding the initial construction of the Mars base.

In SpaceX’s rendering, Optimus robots work on Mars, reenacting the classic photo “Lunch atop a Skyscraper” taken in 1932 during construction at Rockefeller Center in Manhattan. (Image: SpaceX)

The mission statement SpaceX adopted after absorbing xAI in February this year is: scale up, build an intentional conscious sun to understand the universe, and extend the light of consciousness to the stars.

Depending on how you interpret the line, it is either the most absurd expression ever placed on a mission page by a serious company, or the most honest expression. We believe it is the latter.

If you only look at the org chart superficially, SpaceX is a launch services provider, with an internet subsidiary and a newly acquired AI lab. If you take its technical roadmap seriously, it is the only company on Earth assembling the full prerequisite stack needed for the scarce post–turning point transition. If you take its mission statement seriously, it is one of the most capable founders of the modern era, attempting to push humanity through that bottleneck: on the other end of the bottleneck, either we become an interstellar species and share the universe with the intelligent machines we create; or we ultimately become a footnote on a rocky planet that failed to make the leap.

By the time the first child born on Mars asks their parents why their family is there, Starship will already have been flying every day for thirty years. The factories across the neighborhood will be run by Optimus robots, executing descendant models of Grok that have self-improved for 20 years. The compute that keeps her city running comes from data centers in space. Those data centers are manufactured using lunar regolith by other robots and launched into the sky by a mass driver. For nearly a generation’s worth of time, this mass driver has been launching them into deep space on a cadence of one satellite every few minutes. Her parents arrive on Mars in a craft named after a starship in an Iain M. Banks novel—because at some point in the early 21st century, someone who read those books as a teenager decided to spend their entire life turning them into reality.

Banks understands people who would choose to go to Mars. “The Culture” is paradise, but the most interesting characters in his writing are those who leave paradise. This civilization solves the problem of scarcity, and what remains is humanity’s desire for difficult journeys. Even if paradise is next door, the frontier is where meaning lies.

Musk once said that the recruitment pitch for early Mars settlers would be “Shackleton-style recruiting,” derived from the famous 1914 recruiting advertisement for the Transantarctic Expedition: “Recruit men to take part in a dangerous journey. Wages are low, and the cold is fierce, with months of complete darkness; danger is constant—whether you will be able to return safely is uncertain. If successful, you will receive honor and recognition.” This advertisement almost certainly did not exist as a real document, but it has been retold for a century because it captures some truth about those who voluntarily set out.

Why do some people find this appealing?

Musk said: “Life can’t just be one painful problem after another, solved one after the other. There must be something out there that inspires you—so that when you wake up in the morning, you’re happy to be a human being. Earth is the cradle of humanity, and you can’t stay in the cradle forever. It’s time to set out, to become a civilization navigating between the stars—entering the stars, expanding the range and scale of human consciousness. I find this extremely exciting. It makes me happy to be alive. I hope you feel the same.”

Starman, a mannequin wearing a SpaceX spacesuit, sits in the driver’s seat of Elon Musk’s personal Tesla Roadster, orbiting the sun. This car was the payload of Falcon Heavy’s first test flight on February 8, 2018. According to its current trajectory, it will pass near Mars about once every Earth year for roughly the next million years. (Image: SpaceX)

[Original Article Link]