a16z: Moving Computing Power to Space, Letting the Sun Bear Witness, SpaceX Valued at $7.5 Trillion

Original Authors: Marc Andreessen, Michael McGuiness, a16z

Compiled by: Yu Shiyi

Musk's compensation plan at SpaceX revolves around two goals.

The first goal: He will receive his first reward if the company reaches a valuation of $7.5 trillion and establishes a permanent human colony of at least 1 million people on Mars.

The second goal: He will receive a second reward if SpaceX operates data centers in space and those data centers consume at least 100 terawatts of power——a figure over 1,000 times the total power consumption of all data centers on Earth.

If neither goal is achieved, Musk gets nothing, except for the $54,080 annual salary he has been drawing since 2019.

The board members who signed this compensation package have witnessed one thing over the past two decades:

Musk has made seemingly impossible SpaceX predictions time and again, and those predictions have become reality time and again.

He once said SpaceX would put humans into orbit——before that, no private company had ever done it. Now, SpaceX routinely transports astronauts for NASA.

He once said SpaceX would land and reuse orbital-class rockets——before that, the entire industry treated boosters as expendable. Now, SpaceX has completed hundreds of recoveries and reuses.

He once said a satellite internet business could be worth tens of billions of dollars——before that, satellite internet was practically a graveyard for bankrupt companies. Now, Starlink's revenue has grown from zero to $11.4 billion in just a few years.

These predictions are often aggressive in their timelines, but almost never wrong in their direction.

And the original mission SpaceX wrote down in 2002 was to make humanity a multi-planetary species.

So, the board tied his compensation to this mission itself.

If this mission sounds like science fiction, it's perhaps because it indeed comes from science fiction.

Iain M. Banks and the Blueprint of "The Culture"

Iain M. Banks spent twenty-five years writing a world called "The Culture."

By most reasonable standards, it might be the best utopian society humanity has ever imagined.

There, humans live alongside super-intelligent AIs called Minds. The Minds are responsible for running orbital habitats as vast as small worlds. The relationship between humans and AI is not one of enslavement or competition, but partnership.

No one is forced to work.

No one goes hungry.

The Minds handle the staggering computational load required to run space cities.

And humans are responsible for continuing to be human.

This, it turns out, is a full-time job in itself.

The three autonomous drone ships SpaceX uses at sea to recover Falcon 9 boosters are all named after sentient starships from Banks' novels:

Of Course I Still Love You
Just Read the Instructions
A Shortfall of Gravitas

In an interview at the 2023 UK AI Safety Summit, Musk was asked: What should a good AI future look like?

He replied:

"Banks' Culture series is by far the best envisioning of an AI future. Nothing comes close to it for understanding what a fairly utopian, or proto-utopian, AI future could be."

He has been telling us, with the names of those landing platforms, exactly what he wants to build.

Caption: "Of Course I Still Love You" catches a Falcon 9 first-stage booster on April 8, 2016. This was the first successful drone ship landing in history, and the moment reusable orbital-class rocketry moved from theory to reality. The ship's name comes from a sentient starship in Iain M. Banks' Culture series. (Image: SpaceX)

But The Culture is not a frictionless paradise.

Banks' novels are full of war, intrigue, and moral complexity. It is a utopia because the Culture has solved the preconditions for survival so well that trillions of humans can finally attend to what Banks called "the really important things in life":

sports, games, romance, studying dead languages, barbarian societies, impossible problems, and climbing high mountains without safety nets.

Such a future has four prerequisites.

First, access to a meaningful fraction of a star's energy output——orders of magnitude more energy than human civilization produces today.

Second, large-scale physical intelligence: machines that can build, mine, refine, and repair anything without human intervention, and can do so anywhere.

Third, cheap digital intelligence that surpasses biological intelligence.

Fourth, the ability to move mass off Earth cheaply, frequently, and reliably. Because none of the above can scale on Earth alone.

Reasoning Backwards from the Future

Most analyses of SpaceX push forward from the present:

rockets, satellites, contracts, revenue.

But if you want to see what's really happening, a more useful method is to start from the end and work backwards.

Mars City

The operational goal is:

to build a self-sustaining city of 1 million people on Mars within the lifetime of people alive today.

The truly difficult part is "self-sustaining."

This means: if the ships from Earth stop coming, the city must still survive.

It must manufacture everything itself:

food, water, air, energy, medicine, machines, and eventually, more humans.

By SpaceX's own calculations, sending 1 million people and millions of tons of cargo to Mars within a few decades requires thousands of Starship flights; during each transfer window, more than ten launches per day are needed.

These windows are determined by Earth-Mars orbital mechanics, are only a few weeks wide, and open only once every 26 months.

Caption: A SpaceX rendering of a Mars city. (Image: SpaceX)

Moon City

A Moon city is a closer, easier dress rehearsal.

Permanently shadowed craters at the lunar south pole contain ice, and certain ridges receive continuous sunlight, making it a naturally suitable place for a base.

But Musk is not just talking about a scientific outpost.

He envisions building factories on the Moon to produce AI satellites and launching them one after another into space using mass drivers.

A mass driver is another concept Musk borrowed from science fiction. It is an electromagnetic launch system that leverages the Moon's gravity, which is only one-sixth of Earth's, and its lack of atmosphere to hurl solar-powered satellites into deep space on an industrial scale.

These satellites can be manufactured on the Moon because lunar regolith is, by weight, roughly 20% silicon and 10% aluminum——the two primary inputs for solar cells and satellite structures.

Musk explained: "If you want to go beyond 1 terawatt per year scale, you have to go to the Moon."

Caption: A SpaceX rendering of a mass driver at Moonbase Alpha, used to launch lunar-manufactured AI satellites, i.e., data centers, into orbit. (Image: SpaceX)

Orbital Data Centers

Musk's bet is:

In a few years, the most economically viable place to put AI data centers will be in space.

The bottleneck for AI is energy. Outside of China, energy supply growth is very limited, while demand for AI computing power is growing exponentially.

Solar panels in orbit provide 4 to 10 times the power of identical panels on the ground. The exact multiple depends on how sunny the ground location is.

The reason is simple:

In space, there is no atmosphere, no day-night cycle, no clouds, and no seasons.

NASA figured this out decades ago. Now, rockets are finally cheap enough to make it a reality.

Musk expects that in five years, the AI computing power SpaceX launches into orbit annually will exceed the total cumulative installed computing power on Earth.

This is why SpaceX merged with xAI in February.

Rockets and intelligence are becoming the same problem.

Starship: The Vehicle for Everything Upstream

Starship is the vehicle that makes everything upstream possible.

Starship V3, which made its first flight this year, is the largest and most powerful rocket humanity has ever built. It is taller than a 40-story building and has more than twice the thrust of the Saturn V that sent astronauts to the Moon.

According to NASA statistics, the historical cost to reach orbit was about $18,500 per kilogram.

In 2010, the first Falcon 9 reduced this cost by about 85%, to roughly $2,700 per kilogram.

In 2018, Falcon Heavy brought it down further to about $1,400 per kilogram.

And Starship, as the world's first fully and rapidly reusable spacecraft, aims to drive the cost down further to $100 to $500 per kilogram.

Spaceflight, which once cost billions of dollars per launch, is becoming a business measured in the tens of millions.

Starlink: The Cash Flywheel

Starlink is the cash flywheel that helps pay for everything else.

According to SpaceX's IPO documents, the connectivity business segment——almost entirely Starlink——brought in $11.4 billion in revenue in 2025, up about 50% year-over-year, with an adjusted EBITDA margin exceeding 60%.

As of March 2026, Starlink has 10.3 million users in 164 countries, operating on over 9,600 satellites.

Starlink started as a side project to fill the company's own launch capacity; now, it is becoming one of history's great consumer businesses.

In 2019, when a16z was conducting due diligence on SpaceX, several people told us this economic model would never work.

The reason was that Starlink's terminal antennas required antenna technology previously used only in F-22 fighter jets and Navy destroyers—technology that had never been mass-produced for consumers.

SpaceX's first batch of terminals cost about $3,000 to manufacture, yet sold for only $499.

But they eventually drove down manufacturing costs and proved the doubters wrong.

Falcon 9: The Workhorse Buying Time for the Future

Falcon 9 is the workhorse rocket buying time for everything else.

It is the only orbital-class booster on Earth reused at scale. A single booster can typically fly over twenty times before retirement.

In 2025, SpaceX launched 83% of the world's total orbital mass.

Despite everyone else having a half-century head start, SpaceX now launches more payload into orbit than the rest of the world combined.

This is the stack from top to bottom.

Generations from now, a Civilization-like future sits at the top.

Falcon 9 and Starlink sit at the bottom, paying today's bills.

Each layer makes the next one possible.

SpaceX CFO Bret Johnsen described the feeling inside the company:

"[Musk] has created a culture where you first set a goal that initially seems incredibly audacious, and then step by step, you find yourself moving toward something that is absolutely achievable...

Take going to Mars as an example. When I first arrived in 2011, whenever people talked about Mars and a multi-planetary species, others would roll their eyes. Now when we talk about it, the reaction has become: 'Which year?'

I think Elon's greatest strength is that he sets these goals and builds very sound business models around every piece of intellectual property needed to achieve the ultimate objective."

The Idiot Index and the Algorithm

Musk did not initially set out to start a rocket company.

In 2001, at age 30, Musk was thinking about what he wanted to do after selling PayPal.

He had always been interested in space. When he went looking for NASA's plan to send humans to Mars, he was shocked to find: there was no such plan.

So he devised a scheme:

send a small greenhouse to Mars and transmit photos back to Earth.

His thinking was: if people saw a green sprout emerge on the dead red planet, it might reignite public interest in space and generate the political will to fund a real Mars program.

He just needed a rocket to send the greenhouse there.

Later that year, he went to Moscow, looking to buy a refurbished intercontinental ballistic missile. This was the first of two Moscow trips.

Those meetings were reportedly filled with vodka and posturing.

Adeo Ressi, Musk's best friend from his University of Pennsylvania days, went along. He told Esquire in 2012:

"We would all walk into a small room, and everyone had their own bottle of wine in front of them."

The Russians did not take Musk seriously.

On one occasion, a chief designer even spat at Musk and his team as a gesture of contempt.

The second trip to Moscow was in February. Musk asked how much one missile cost.

The reply: $8 million each.

Musk countered: $8 million for two.

Musk's aerospace advisor Jim Cantrell recalled the response was something like:

"Young man, no."

And implied he had no money at all.

Musk judged they were not serious and got up to leave.

Cantrell thought the trip was over.

On the flight back, he and Mike Griffin, who later became NASA Administrator and was also along as an advisor, ordered drinks and toasted finally leaving Moscow.

Musk sat in the row ahead of them, hunched over a laptop.

Then, he turned around:

"Hey, guys, I think we can build this rocket ourselves."

He showed them a spreadsheet listing the raw materials needed for a rocket: aluminum, titanium, copper, carbon fiber, and the cost of each.

Those material costs were only 2% of the quoted price.

As Musk later said:

"Clearly, you just need to figure out clever ways to take those materials and shape them into the form of a rocket."

Within months, Musk decided to risk $100 million to start a rocket company. This was more than half of the roughly $180 million he received from the PayPal sale.

SpaceX was thus founded in a warehouse in El Segundo, California.

He extended invitations to five people for the founding team.

Three declined, including Cantrell and Griffin.

The two who said yes were:

Tom Mueller, who later became Vice President of Propulsion and employee number one;
Chris Thompson, employee number two, responsible for operations and production.

Musk later joked:

"SpaceX in 2002 basically consisted of a carpet and a mariachi band. That was it. As you can see, I'm a dancing machine."

Years later, Musk called the diagnostic tool behind that spreadsheet the "idiot index."

If the ratio of a component's cost to its raw material cost is high, then either you are an idiot, or you are working with idiots.

It sounds like a joke, but it is the foundation of SpaceX's strategy.

Every part SpaceX procures comes with an "idiot index" calculation.

One of the company's most legendary early stories involves Steve Davis.

Davis joined SpaceX straight out of Stanford as the company's 14th employee. His task was to source an actuator to control the steering of the Falcon 1 rocket's upper stage.

When he reported that a traditional aerospace supplier quoted $120,000, Musk laughed.

Musk told him the part was no more complex than a garage door opener and gave him a $5,000 budget to build it from scratch.

Biographer Ashlee Vance recorded that Davis spent nine months iterating the design, ultimately producing a working actuator for just $3,900.

When Davis sent Musk the technical breakdown of this victory, Musk replied with just two letters:

"Ok."

To push the idiot index toward its theoretical minimum, you must vertically integrate and control the process end-to-end.

But vertical integration creates fixed costs that only pay off at high volume.

And achieving high volume in the rocket business meant breaking how the industry had always operated.

Traditional launch providers, like ULA and Arianespace, treated every mission as a custom project.

Customers specified the orbit, payload, and integration requirements, and the launch provider designed a bespoke mission around the satellite.

This model assumed:

only a few launches per year, with extremely high costs per mission.

It made scaled manufacturing impossible.

SpaceX did the opposite.

They published a Falcon User's Guide, defined the rocket's exact specifications, and told customers:

please design your satellite to fit our rocket.

At the time, this was considered radical and cost SpaceX some early business.

But it unlocked the manufacturing flywheel.

Standardization and reusability reinforce each other.

Because every Falcon 9 is the same, a recovered booster can become a finished, qualified product ready to fly again.

The first Falcon 9 booster to fly twice completed its re-flight in 2017.

By 2020, a single booster could fly five times.

By 2021, ten times.

Today, the record holder has flown 35 times.

This reusability has changed the economics of spaceflight, and it is hard to see how competitors can catch up.

In 2021, Musk estimated that Falcon 9's marginal launch cost (excluding overhead allocation) to send 15 tons to orbit was about $15 million under optimal conditions. He said this was roughly one-half to one-third the cost of alternatives.

Today, SpaceX launches a rocket with a reused booster every two to three days, while competitors launch only a handful of custom rockets per year.

But SpaceX's advantage is not just economies of scale, vertical integration, and better strategy.

It also has speed and culture.

Traditional aerospace companies eliminate uncertainty through analysis.

NASA once used a polite phrase to describe Boeing's commercial crew program:

"Employing a mature systems engineering approach, investing upfront in engineering studies and analysis to mature the system design before manufacturing and testing."

Measure twice, cut once.

SpaceX reversed that order.

The company builds many cheap prototypes, pushes them to failure, learns from the failures, and iterates quickly.

The Starship test program has probably produced more spectacular explosions than any rocket program in history.

But each failure is a data point where reality deviated from the model.

This contrast is very clear to those who have worked in both worlds.

Garrett Reisman is a NASA astronaut who flew on two Space Shuttle missions. In 2011, he left NASA to join SpaceX as a senior engineer.

He described the prevailing NASA view of SpaceX back then:

"They're cowboys; they're dangerous; they're going to get people killed."

But what truly changed his perspective was seeing how SpaceX worked.

"They were doing in a month what NASA might take a year to do. We were stunned."

The clearest example is the Falcon 1 program.

Between 2006 and 2008, SpaceX launched four Falcon 1 rockets from a small atoll in the Pacific called Kwajalein.

The first three all failed.

But each failure was different and provided learning:

First, a fuel leak;
Second, abnormal propellant slosh;
Third, residual engine thrust causing a separation collision.

By September 2008, the company only had enough money for one more launch.

And this was not the only company Musk had standing on the edge of a cliff.

Tesla, the electric car company he was simultaneously building, was also weeks away from bankruptcy.

He had to decide: whether to concentrate his remaining PayPal cash on one company or split it between the two.

Musk recalled:

"That was a very tough decision. In the end, I decided to split my remaining money and try my best to keep both companies alive. But that could have been a very bad decision, causing both companies to die together.

I never thought I would have a nervous breakdown, but I came really close."

He couldn't choose, because in his worldview, both missions were indispensable:

Tesla was to accelerate the world's transition to sustainable energy.

SpaceX was to make humanity a multi-planetary species.

Musk's fiancée at the time, Talulah Riley, said in the BBC documentary The Elon Musk Show: "All available resources had to go into the companies. He gave me the option to leave. He said: 'The hardest part is coming next, you don't have to stay and go through it with me.'"

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

The fourth launch succeeded.

That December, just weeks before SpaceX was about to run out of cash, NASA awarded the company a $1.6 billion cargo contract.

When NASA called to notify Musk, he was hit by a massive emotional release and blurted out:

“I love you guys.”

The pattern formed from failing fast and correcting fast later became the culture for every project at SpaceX.

This is why today SpaceX can iterate rapidly between Starship test flights, while traditional aerospace projects often take years to go from a flight anomaly to a vehicle redesign.

The reason this method is more effective than the alternative is:

For problems you don't yet fully understand, you cannot arrive at a perfect solution through thinking alone.

Reality is the only sufficiently qualified validator.

The key is to reduce the cost of consulting reality low enough so that you can consult it frequently.

SpaceX's “Algorithm”

The above is the SpaceX iteration loop told through stories.

But it also has a written version.

Over the past two decades, Musk has codified SpaceX's method into a five-step operational process, known internally as “The Algorithm.”

Tim Berry, who worked at SpaceX for ten years and once led the Falcon 9 and Falcon Heavy upper stage production team, said this method was “drilled into our brains.”

Walter Isaacson published its standard version in Musk's biography:

1. Question every requirement

Every requirement should come with the name of the person who made it.

You should never accept a statement like “this requirement comes from the legal department” or “this requirement comes from the safety department.”

You need to know who the real person who made the requirement is, and no matter how smart that person is, question it.

Requirements from smart people are the most dangerous, because people are least likely to question them.

Then, make those requirements less dumb.

2. Delete any part or process you can

You may have to add them back later.

In fact, if you don't end up adding back at least 10% of what you deleted, you haven't deleted enough.

3. Simplify and optimize

This step should come after step two.

A common mistake is to simplify and optimize a part or process that shouldn't exist in the first place.

4. Accelerate cycle time

Every process can be sped up.

But you should only do this after completing the first three steps.

Musk said he once made a mistake at the Tesla factory: spending a lot of time speeding up certain processes, only to later realize those processes should have been deleted in the first place.

5. Automate

Automation comes last.

The mistake Tesla made at the Nevada and Fremont factories was trying to automate at the beginning, rather than first questioning requirements, deleting parts and processes, and shaking out the bugs.

Most engineering organizations jump straight to step five.

They automate a process that shouldn't exist in the first place.

SpaceX runs these steps in sequence, every time, in every part of the company.

When the “Algorithm” has been run on a piece of hardware enough times, it starts to look like nothing else in the industry.

Caption: Three generations of SpaceX Raptor engines, V1 to V3. (Photo: SpaceX)

Raptor 3 is the product of a team iterating on the same engine for a decade.

It produces 22% more thrust than Raptor 2, weighs 40% less, and requires no heat shield.

The reason is that the plumbing and wiring harnesses that used to hang on the outside of the engine have been integrated into the engine's metal structure through 3D printing.

Musk said:

“The amount of work required to simplify the Raptor engine, internalize secondary flow paths, and add regenerative cooling for exposed components is staggering. It is approaching the known limits of physics.”

In the history of spaceflight, no known engine program has iterated this fast.

The Space Shuttle Main Engine flew essentially the same design for its final three decades.

The RD-180 that powers the Atlas V is a derivative of an engine designed in the 1970s.

SpaceX, in less than a decade, has done three completely new designs of the Raptor, each one substantially better than the last.

The same philosophy applies to people.

By mid-2018, Falcon 9 reusability had entered a reliable rhythm, and Musk turned his attention to the satellite internet constellation that would ultimately fund all the upstream work.

The Starlink team was based in Redmond, Washington, with many senior engineers coming from Microsoft, and the development pace was slower than Musk wanted.

In June, he flew to Redmond and fired the senior leadership team.

He then transferred young star engineers from the rocket division and gave them one year to launch the first operational satellites.

This is a brutal way to manage a company. From the media coverage of the firings, the division seemed to be collapsing.

But 11 months later, in May 2019, the first batch of Starlink satellites launched.

Musk cleared the bottleneck, then moved on to the next problem.

He manages everything this way.

In 2018, Tesla was in the “production hell” of Model 3 ramp-up, with its burn rate threatening survival. Musk literally moved into the factory.

Years later, he recalled:

“I lived in the Fremont factory and the Nevada factory for three years straight. I slept on the floor under my desk so that the whole team could see me during shift changes.

This was important, because if the team thinks their leader is off enjoying themselves somewhere, drinking Mai Tais on a tropical island, they get demoralized.

Because they could see me sleeping on the floor during shift changes, they knew I was there. It made a huge difference; they gave it their all.”

Later, he turned this into a company-wide rule:

The higher your position, the more visible your presence must be.

To find someone comparable to Musk's way of operating as a CEO, you have to go back to the era of industrialists in the late 19th and early 20th centuries:

Henry Ford, Andrew Carnegie, Thomas Watson, Andrew Mellon, Cornelius Vanderbilt.

What's unique about Musk's operating style is his relationship with work.

It is said that he shows up at each of his companies every week, identifies the single biggest problem, and fixes it.

52 weeks a year, he does this every week.

So in theory, each company solves 52 of its biggest problems that year.

An engineer who joined SpaceX from another aerospace company described the experience as:

“Like being thrown into a shocking zone of competence. Everyone around you is absolutely competent.”

Constellation

SpaceX looks like a company.

But a more useful way is to see it as the central node of a corporate constellation.

These companies are run by the same person, building toward the same long-term mission, and are almost impossible to disentangle from one another.

Over the past two decades, Musk has assembled a set of companies. Each one solves a constraint that would otherwise become a bottleneck for the others.

Now, they are beginning to compound on each other.

SpaceX's merger with xAI in February is a microcosm of what SpaceX is becoming.

If compute ultimately goes into orbit——this is Musk's bet——then SpaceX has the most credible path to deploy it at the scale AI requires.

Putting mass into orbit, and producing intelligence at scale, may be the two most critical capabilities of the coming decades.

Now, they reinforce each other under one roof.

xAI brings Grok, a frontier model uniquely positioned on real-time information through access to X's live data stream.

It also brings the engineers who built the Colossus 1 and Colossus 2 supercomputers. The speed of these engineers exceeded what many in the industry imagined possible.

Caption: Colossus 1. (Photo: xAI)

The construction of Colossus is worth pausing to examine closely.

xAI took over an old factory in Memphis and had 100,000 GPUs training within 122 days.

Once the racks started arriving, it took only 19 days to get the entire cluster running.

NVIDIA CEO Jensen Huang said this about Musk:

“From concept to building a massive, liquid-cooled, powered, permitted factory, and completing it in that timeframe, is superhuman.

As far as I know, there is only one person in the world who can do this.

What they accomplished is unique. No one has ever done it. 100,000 GPUs, as a cluster, was easily the fastest supercomputer on Earth in 2024.

This normally takes three years of planning, then equipment delivery, and another year to get everything running.”

A project that would take the rest of the industry at least four years took Musk and the xAI team four months.

In May this year, Anthropic agreed to pay SpaceX $1.25 billion per month to buy the full compute capacity of Colossus 1.

A few weeks later, in an amended IPO filing, SpaceX disclosed that Google will pay $920 million per month for access to 110,000 GPUs, roughly half the compute capacity that Anthropic is getting.

These two deals together represent about $26 billion in annual revenue.

And this is just two customers, paying for a business that didn't exist at SpaceX before it absorbed xAI earlier this year.

Chips, power, and land are all scarce.

SpaceX is becoming one of the few companies with enough AI infrastructure to both rent out compute externally and pursue its own ambition of building a leading frontier model.

What xAI gets from SpaceX is a more durable solution to the power constraint Musk believes will limit AI in the coming years.

Producing enough electricity to meet the intelligence demand he anticipates requires grid expansion, new power plants, and years-long permitting processes, and the industry doesn't have that much time.

In his view, orbital solar is the way out, because it is nearly limitless.

And SpaceX is the only company with a vehicle that can send compute up there at scale.

Whether he is right is one of the most important open questions in tech.

But SpaceX's IPO documents show the company takes this bet very seriously: it expects AI to be the company's largest future market, far larger than any other.

The space business that built the company almost looks like a rounding error compared to these ambitions.

Tesla: Another Core Piece of the Constellation

Tesla is another important component of this constellation.

Its integration with SpaceX runs deep, in a different way.

Tesla and SpaceX share a founder, talent pools, operational culture, and increasingly overlapping technology roadmaps.

Tesla provides three things to the SpaceX-xAI constellation.

First, chips.

AI5, AI6, and Dojo3 are all designed internally by Tesla.

Musk has made it clear that these chips are not just for cars, but are building blocks for a broader constellation computing stack.

AI5 handles autonomous driving inference.

AI6 is designed for Optimus and AI data centers.

Dojo3, paired with the planned AI7, is engineered for orbital computing power.

Second, robots.

Tesla's bet is that Optimus will become the physical AI layer for factories, warehouses, homes—environments meant to run without human labor—and ultimately serve the lunar and Martian cities Musk envisions.

Third, solar energy.

Musk says Tesla and SpaceX are each building toward 100 gigawatts of annual solar cell production capacity to support AI construction on Earth and in orbit.

Then there is TeraFab.

This April, Tesla disclosed it had begun ordering equipment for a research semiconductor fab within its Giga Texas complex.

Musk told investors on Tesla's Q1 2026 earnings call:

"We expect this to be roughly a $3 billion project, with capacity of perhaps a few thousand wafers per month."

SpaceX is separately funding a much larger facility, designed to reach a capacity of about 1 million wafers per month at maturity.

The reason is that no existing fab can scale at the speed Musk envisions.

And the scale he envisions is measured in gigawatts.

Musk said last week: "This is not something we are committing to do. This is something we will try to do, and think we can probably do: reach an annualized rate of about 1 gigawatt of space AI compute by the end of next year.

Then, vision-wise, scale by an order of magnitude each year.

That means, two and a half years from now, 10 gigawatts annualized in space. Three and a half years from now, perhaps 100 gigawatts.

Then, depending on global chip manufacturing and TeraFab progress, continue beyond that scale to 1 terawatt per year, which is 1,000 gigawatts.

That is twice the electricity consumption of the United States."

Caption: SpaceX's TeraFab is designed to reach an output of 1 terawatt per year, roughly twice the current U.S. electricity consumption. (Image: terafab.ai)

Comparing Musk to the industrialists of the Gilded Age captures something real, but also points to the differences.

Carnegie built steel.

Vanderbilt built railroads.

Each dominated one sector of that era's industrial base.

Musk is trying to do several sectors simultaneously:

space, energy, AI, robotics, tunneling, brain-computer interfaces, autonomous vehicles.

And bend them all toward a goal most consider highly fantastical.

Whether any of this will work is genuinely unknown; much of it may not succeed.

But the attempt itself has no historical precedent, and may become a rallying point for another century.

The World SpaceX Is Opening

Before its retirement in 2011, the Space Shuttle cost about $54,500 to deliver 1 kilogram of cargo to orbit.

Once Starship matures, Musk expects costs to drop to $100 per kilogram.

When the cost of accessing space falls by more than 500 times, every industry that can theoretically exist in space begins to become economically viable.

There are many such industries.

Caption: Starship and Super Heavy are designed to return to the launch site after flight and be caught by the launch tower, enabling rapid turnaround and re-flight without refurbishment. (Image: SpaceX)

The closest historical analogy might be the U.S. Transcontinental Railroad.

Before 1869, traveling from New York to San Francisco took six months by wagon, cost about a year's wages, and carried a very real risk of death.

After 1869, the journey took just one week.

The railroad itself was a stunning engineering achievement, but the real story is what it unlocked:

Sears Roebuck, meatpacking giants like Swift and Armour, Standard Oil, and eventually U.S. Steel—all born from the railroad boom and further consolidated into industrial empires.

If Falcon 9 is the space age's Transcontinental Railroad, then Starship might be the equivalent upgrade to the airplane.

The railroad opened a continent.

The jet age opened a planet.

Starship will open the solar system.

Industrializing the Moon

Since humans first gazed at the Moon, it has held scientific significance.

Now, it is beginning to hold economic significance.

Because it is an entire world made of industrial raw materials.

First, consider how to send things away from the Moon.

As mentioned earlier, the Moon has only one-sixth of Earth's gravity and no atmosphere, making mass drivers—not rockets—the natural way to transport cargo from the lunar surface.

This would completely change transportation economics.

Once the track is built, the marginal cost of delivering finished goods is primarily determined by electricity, not fuel.

And electricity on the Moon is sunlight.

A package is flung from the lunar surface, re-enters Earth's atmosphere with a heat shield, opens a parachute, and lands at a recovery site.

When throughput is high enough, the marginal cost no longer looks like spaceflight, but more like freight.

Next: what you manufacture there.

The same lunar regolith can provide the silicon and aluminum needed for solar cells and satellites, and can also serve as feedstock for an entire industrial base.

The space revolution of the 2030s and 2040s might look like this:

autonomous mining vehicles operating on the regolith around the clock;

smelters producing aluminum and silicon;

factories assembling satellites, solar panels, and the chips needed to run them.

Most industries on Earth have a lunar version waiting to be built.

SpaceX cannot build it all alone.

Those who build the "Lunar Alcoa," "Lunar Caterpillar," and "Lunar Union Pacific" will become the titans of the 21st century.

Caption: Starship HLS is the lunar lander SpaceX is building for NASA's Artemis program, designed to return humans to the lunar surface after more than 50 years and deliver the foundational modules needed for a permanent presence near the lunar south pole. (Image: SpaceX)

Computing Power in the Sky

By 2030, the bottleneck for AI may not be chips, but electricity.

The obvious response is to build more solar in Texas or Nevada.

But this hits a wall faster than people think.

1 terawatt of continuous solar power requires roughly 1% of the U.S. land area.

And new utility grid interconnection permits typically take a year or more.

To build Colossus in Memphis, xAI needed to deploy a temporary fleet of gas turbines, navigate state permitting processes, and establish a separate power hub across the state line in Mississippi just to bring 1 gigawatt online.

Scaling this to the hundreds of gigawatts needed for AI buildout is simply infeasible.

Even the gas turbines used as backup for solar have internal blades and vanes backlogged until 2030.

Caption: A Baker Hughes Frame 5/2C gas turbine generator. The cast blades and vanes inside such turbines are produced by a handful of specialized foundries, all backlogged until 2030. A single hyperscale data center requires dozens of these units. (Image: Baker Hughes)

The solution: move the computing power to where the sunlight already is.

Once Starship flies daily and orbital deployment becomes routine, this becomes easier.

And the economics will continue to improve along the cost curves of rocket launches, solar panels, and chips.

SpaceX CFO Bret Johnsen explained:

"We are ramping factory capacity and benefiting from declining silicon costs, so our costs will come down over the next few years.

If you look at ground-based solutions, the curves are moving in the opposite direction. Everything is getting more expensive: cooling methods, electricity rates won't drop, land and regulation become more challenging."

A common objection comes from those who hear "space data center" and imagine launching a building the size of Colossus into orbit.

But that is not the reality.

Early SpaceX investor Gavin Baker said: "It's roughly the size of a Blackwell rack, with solar wings maybe 500 feet long on each side. You place it in a sun-synchronous orbit so the solar panels are always in sunlight.

I've spent a lot of time at Starbase over the years and talked to many SpaceX engineers. I truly believe this is the most talented group of engineers on the planet, and they are very confident they have solved this problem."

Caption: AI Sat Mini is designed to capture the sun's energy. (Image: terafab.ai)

In fact, Musk believes AI Sat Mini is easier to manufacture than Starlink satellites.

He explained:

"You still have some laser links, but you don't need the incredibly complex antennas found on Starlink satellites.

Between the two, the easier one to design is the AI satellite.

AI satellites don't require any magic. We've already developed a lot of the technology for Starlink V3 satellites. Compared to what we are already doing, we don't see this as a particularly difficult problem."

He expects that within five years, SpaceX will be launching more AI compute capacity into orbit annually than the total cumulative installed compute on Earth.

The math here roughly works out to:

10,000 Starship launches per year, meaning more than one launch per hour around the clock.

By the late 2030s, as lunar mass drivers come online, the petawatt threshold comes into view:

that is 1,000 times the compute deployed in 2030, launched into deep space at a cadence of one satellite every few minutes.

Mars

The Mars trajectory was originally supposed to begin this year.

In September 2024, Musk announced that SpaceX would launch five uncrewed Starships to Mars during the November 2026 transfer window, carrying Optimus robots to test landing systems, search for ice, and begin building infrastructure for future crewed missions.

In May 2025, he said the probability of achieving that goal was fifty-fifty.

But earlier this year, things changed.

On February 8, Musk posted on X announcing that SpaceX would delay the Mars timeline and shift near-term focus toward a self-sustaining city on the Moon.

The reasoning:

Mars launch windows occur only every 26 months, with a flight time of 6 months; the Moon can be reached every 10 days, with a flight time of just 2 days.

He wrote:

“This means we can iterate much faster and complete a lunar city compared to a Martian city.

That said, SpaceX will also work toward building a Martian city and will begin doing so in about five to seven years, but the overwhelming priority is to secure the future of civilization, and the Moon is faster.”

On the surface, this was a pivot.

But in reality, this was the moment the path to a city of a million people on Mars became clear.

The orbital data center thesis became sharper in late 2025 to early 2026, giving the Moon a new role.

Achieving petawatt-scale orbital compute requires:

lunar mining, lunar refining, lunar manufacturing of solar panels, radiators, and satellite structures, and launching them into orbit using surface-powered mass drivers.

Industrial infrastructure at this scale requires a permanent population, and a permanent population requires a city.

That city can be funded entirely by the orbital compute industry, while simultaneously serving as a dress rehearsal for Mars.

Every problem SpaceX must solve to build a self-sustaining city on Mars will be encountered first in the lunar city:

Radiation shielding;
Life support;
In-situ resource utilization;
Governance of a permanent off-world population;
Supply chains across a gravity well.

Building the lunar city will teach SpaceX how to build a Martian city, using a much faster iteration loop.

The first uncrewed lunar surface landing demonstration is targeted as early as 2027.

According to Musk's public timeline, the lunar city will follow in less than a decade.

Mass drivers, lunar industrial construction, and lunar manufacturing of orbital compute infrastructure will launch in parallel.

Then, Mars.

But the hardest part isn't transporting people.

It's building the Mars-side infrastructure capable of absorbing them.

The lunar dress rehearsal will help.

Optimus will help too.

In his May 2025 Starbase Mars presentation, Musk repeatedly mentioned that early uncrewed Starships will carry Optimus robots to search for resources and begin building infrastructure for human arrival.

The company is building a 1-million-unit-per-year production line in Fremont and a 10-million-unit-per-year line at Giga Texas.

These robots are still in early production and have yet to perform meaningful useful work in Tesla's factories.

But the production capacity coming online in the next two to three years is critical for bootstrapping the initial Mars base.

Caption: SpaceX rendering of Optimus robots working on Mars, recreating the famous 1932 photograph "Lunch atop a Skyscraper" from the construction of Rockefeller Center. (Image: SpaceX)

A Conscious Sun

The mission statement SpaceX adopted after absorbing xAI in February is:

Scale to bring a conscious sun into being, to understand the universe, and to extend the light of consciousness to the stars.

This sentence depends on how you read it.

It is either the most absurd thing a serious company has ever written on a mission page,

or the most honest.

We believe it is the latter.

If you squint at the organizational chart, SpaceX is a launch provider with an internet subsidiary and a recently acquired AI lab.

If you squint at the technology roadmap, it is the only company on Earth assembling the complete front-end stack for a post-scarcity transition.

If you squint at the mission statement, it is one of the most operationally capable founders of our era, seriously attempting to push humanity through that bottleneck.

On the other side of the bottleneck, there are two possibilities:

One, we become an interstellar species, sharing the universe with the intelligent machines we built;

The other, we are merely a footnote on some rocky planet, a lineage that failed to make the leap.

When the first child born on Mars asks their parents why their family is here, Starship may have been flying routinely for thirty years.

In the factory around the corner, Optimus robots are working, running a descendant of Grok that has been self-improving for twenty years.

The compute power sustaining her city's operations comes from space-based data centers.

Those data centers were manufactured from lunar regolith by other robots and launched by a mass driver that has been flinging satellites into deep space at a cadence of one every few minutes for nearly a generation.

Her parents came to Mars aboard a vehicle named after a starship from an Iain M. Banks novel.

Because at some point in the early 21st century, someone who had read those books as a teenager decided to spend a lifetime turning them into reality.

Banks understood those who choose to go to Mars.

The Culture is paradise, but his most interesting characters are often those who leave paradise.

Civilization solved scarcity; what remains is the human desire for difficult journeys.

Even when paradise is next door, the frontier is still where meaning dwells.

Musk has said the recruitment pitch for early Mars colonists will be Shackleton-esque.

It comes from the famous recruitment ad for the 1914 Imperial Trans-Antarctic Expedition:

“Men wanted for hazardous journey. Low wages, bitter cold, long months of complete darkness, constant danger, safe return doubtful. Honour and recognition in case of success.”

The ad is almost certainly apocryphal.

But the story has been retold for a hundred years because it captures something true about those who choose to set out.

Why would anyone find this appealing?

Musk said: "Life cannot just be about solving one miserable problem after another. There need to be things that inspire you, that make you glad to wake up in the morning and be part of humanity. Earth is the cradle of humanity, but you cannot stay in the cradle forever. It is time to go out and become a starfaring civilization, to move out among the stars, and expand the scope and scale of human consciousness. I find that incredibly exciting. It makes me glad to be alive. I hope you feel the same way."

Caption: Starman, a mannequin wearing a SpaceX spacesuit, sits behind the wheel of Elon Musk's personal Tesla Roadster, orbiting the Sun. The car was the payload for the first Falcon Heavy test flight, launched on February 8, 2018. Its current orbit will pass near Mars roughly once per Earth year for the next million years or so. (Image: SpaceX)