SpaceX, an AI company in the guise of a rocket company

The largest IPO in human history might really be on the way.

However, after reading through the entire SpaceX prospectus, the only two words that come to mind are "disconnect."

On one hand, Elon Musk is still telling the world's most alluring story: colonizing Mars, becoming a multi - planetary civilization, and leading humanity into the universe.

This story is also propelling SpaceX towards the largest IPO in human history — aiming to raise $75 billion and aiming for a valuation of up to $2 trillion.

On the other hand, though, SpaceX is pouring money into AI like crazy.

In the first quarter of 2026, SpaceX's capital expenditure on its AI business reached a staggering $7.723 billion, more than twice the combined total of its space and communication segments.

Moreover, on the commercial front, everything is shifting towards AI. Starting from May 2026, Anthropic will pay SpaceX approximately $1.25 billion per month for computing power services.

Based on this figure, just Anthropic alone can bring in about $15 billion in annual revenue for SpaceX.

This figure is almost 80% of SpaceX's total revenue last year. That is to say, selling AI computing power might soon become SpaceX's most important source of revenue.

This isn't the end. What's truly astonishing is the valuation logic. According to the TAM (Total Addressable Market) calculation provided by SpaceX in its prospectus, its total market potential reaches $28.5 trillion.

Among this, AI alone accounts for $26.5 trillion, which is 12 times the size of the space and network connectivity market.

As a result, an extremely absurd yet entirely reasonable situation has emerged:

Based on this market potential, about 93% of the future value of this world - renowned rocket company will ultimately come from AI.

It seems that no matter how alluring the space story is, it can't compare to the tangible profits from AI computing power.

Spent $7.7 billion in the first quarter: SpaceX heavily invests in AI

In the past three years, SpaceX's revenues were $10.387 billion, $14 billion, and $18.674 billion respectively.

Looking at its business structure, it mainly consists of three segments: Starlink, Aerospace, and AI.

Currently, Starlink is the most profitable business. In 2025, Starlink's revenue reached $11.387 billion, accounting for 60.9% of the total revenue, and its operating profit was as high as $4.423 billion.

The Starlink business involves sending "base stations" into space.

As of March 31, 2026, SpaceX had deployed approximately 9,600 Starlink satellites in low - Earth orbit, accounting for about 75% of the total number of maneuverable satellites in orbit globally.

The broadband service provided by Starlink has an average latency of about 25 milliseconds, and the average peak - hour download speed for residential users reaches 225 Mbps. This means that in many scenarios, the experience is close to that of terrestrial broadband.

The business model is also clear. Users purchase terminal equipment and then pay a monthly network service fee. As of the end of March 2026, Starlink had approximately 10.3 million global users, a year - on - year increase of 105%.

SpaceX's second business segment is aerospace, which is the most well - known. In 2025, the aerospace business had revenues of $4.086 billion and a loss of $657 million.

What the aerospace business does is to reduce the cost of rocket launches and make rocket launches as frequent as airplane take - offs and landings.

Before the emergence of SpaceX, the long - term average launch cost in the global aerospace industry was about $18,500 per kilogram.

According to NASA data, when SpaceX launched the "Falcon 9" for the first time, it had already reduced the orbital insertion cost to about $2,700 per kilogram, a full 85% reduction compared to before.

With the subsequent launch of the "Starship", the rocket launch cost will further decrease.

SpaceX's last business segment is AI.

The so - called AI business is to sell AI computing power, which is currently the most aggressive and money - burning direction for SpaceX.

In 2025, the AI business had revenues of only $3.2 billion but a loss of up to $6.355 billion. In the first quarter of this year, the loss in the AI business reached as high as $2.5 billion, accounting for more than half of the company's total loss.

In terms of capital expenditure, AI has become SpaceX's largest area of investment.

In the first quarter of this year, the capital expenditure for the aerospace business was $1.052 billion, and for the Starlink business it was $1.332 billion. Meanwhile, the capital expenditure for the AI business reached a whopping $7.723 billion, more than twice the combined total of the space and communication segments.

At this point, a question arises: why is a rocket - building company pouring the most money into AI?

The space business can't support a $2 - trillion valuation

To answer this question, we have to start with SpaceX's own business structure.

Many people might think that SpaceX's most alluring aspects are its aerospace and Starlink businesses. However, the problem is that although these two businesses are grand, they are not the kind that can achieve "exponential growth" in the short term.

Let's start with aerospace. In the aerospace business, there are two core indicators: orbital insertion mass and number of launches.

The former can be understood as the total amount of materials that SpaceX has transported into orbit within a certain period, representing SpaceX's industrialized transportation capacity.

The greater the mass that can be sent into orbit, the more Starlink satellites SpaceX can deploy.

In the past few years, SpaceX's transportation capacity has shown rapid growth. In 2023, the company's cumulative orbital insertion mass was 1,210 metric tons.

It increased to 1,699 metric tons in 2024 and further to 2,213 metric tons in 2025. In just the first quarter of 2026, it had already achieved an orbital insertion capacity of 556 metric tons.

Although the transportation capacity has grown rapidly, the vast majority of this capacity doesn't generate revenue and is mainly for its own use.

In 2025, out of the total orbital insertion mass of 2,213 metric tons, about 1,901 metric tons were internal payloads, mainly for the continuous expansion of the Starlink constellation.

On the other hand, it's also difficult for the rocket business itself to experience "explosive growth" in the short term.

The reason is simple. The aerospace industry is naturally a capital - intensive industry with strong engineering constraints. These factors determine that it can't expand infinitely like an internet business.

This can also be seen from SpaceX's number of launches in the past three years. In the past three years, SpaceX launched rockets 96 times, 134 times, and 165 times respectively, with a compound growth rate of 31.1%.

The aerospace business is expanding slowly, and Starlink also has its own bottlenecks.

The logic of Starlink is more like that of a "global operator". The two core indicators affecting its commercialization are the number of users and ARPU.

As of March 31, 2026, the number of Starlink's global service lines had reached approximately 10.3 million, a year - on - year increase of about 105% compared to 5 million a year ago. However, the monthly ARPU of Starlink decreased from $91 to $66.

This means that if Starlink wants to rapidly expand its user base, it will inevitably lead to a decline in the revenue per user.

So, neither the aerospace nor the Starlink business has the geometric revenue growth potential of AI.

In addition to the revenue growth rate, another problem with the aerospace and Starlink businesses is the lack of long - term growth potential.

Currently, the TAM (Total Addressable Market) of the global commercial launch market is at most a few hundred billion dollars.

Even if we include Starlink's "space internet", the entire market space is only about $1.6 trillion.

But now, SpaceX is valued at $2 trillion. That is to say, this market space can hardly support SpaceX's current valuation.

Therefore, Elon Musk has to find a story with no real ceiling. And that story is AI.

93% of the value comes from AI

Compared with the first two businesses, the story of selling AI computing power is much more straightforward.

According to the information disclosed in the prospectus, starting from May 2026, Anthropic will pay SpaceX approximately $1.25 billion per month for computing power services.

Based on this figure, just Anthropic alone can bring in about $15 billion in annual revenue for SpaceX.

This figure is almost 80% of SpaceX's total revenue last year.

Moreover, this is just the beginning.

This contract will last until May 2029. Considering the production ramp - up discounts in the first two months, the cumulative value of the entire contract period may exceed $40 billion.

More importantly, SpaceX clearly stated in its prospectus that it expects to sign more similar large - scale AI computing power service agreements in the future.

If the terrestrial computing centers can bring direct revenue to SpaceX, then the space data centers will take SpaceX's growth potential to a whole new level.

In Elon Musk's view, AI will eventually hit a wall. Because the Earth's energy, power grid, land, and the construction speed of data centers are all limited.

But space is different. 99.8% of the energy in the solar system comes from the sun. The solar energy in orbit is almost continuous and uninterrupted.

Therefore, Elon Musk proposed the concept of "orbital AI computing".

The core idea is to deploy large - scale AI computing satellites in orbit in the future and migrate some data centers to space.

In this way, AI computing will no longer be completely restricted by the terrestrial power grid, land, and the construction cycle of data centers. In the long run, there is even a chance to significantly reduce the computing cost per token.

According to Elon Musk's previous calculations, the cost of Chinese solar panels has dropped to $0.25 per watt.

In orbit, the power generation efficiency is about five times that on the ground, and at the same time, a large amount of energy storage and battery costs are saved.

Overall, the theoretical unit electricity cost could even be only one - tenth of that on the ground.

According to Elon Musk's plan, the Starship is likely to give priority to serving orbital - level data centers in the future.

SpaceX is even preparing for 10,000 launches per year, or even 20,000 - 30,000 launches per year.

This figure sounds crazy. But in Elon Musk's view, compared with the global aviation industry, this is still a low - frequency system.

According to Elon Musk's vision, in five years, the AI computing power deployed and operated by SpaceX each year may even exceed the sum of all other systems on Earth.

What's even more amazing is that only SpaceX can actually achieve this.

Because it not only has the transportation capacity but also the world's most mature large - scale satellite engineering system today.

In the past few years, Starlink has proven that SpaceX has the ability to manufacture, launch, and operate large - scale low - orbit satellites.

Its V1 and V2 Mini satellites have been verified in launch vibration, shock, overload, noise, and vacuum environments, with an average operational reliability of 99.9%.

At the same time, through Starlink, it has established inter - satellite laser links, mesh networks, global ground stations, and constellation management capabilities. These capabilities can be directly applied to AI computing satellites.

Since the data generated by AI computing satellites ultimately needs to be transmitted back to the ground, and Starlink happens to have global communication capabilities, the current Starlink satellites themselves have already become an important part of the orbital AI computing network infrastructure.

In a sense, today's Starlink is more like a precursor version of the future "orbital AI infrastructure".

With AI computing power and AI products such as Grok and xAI, SpaceX has formed a complete AI business ecosystem.

In its vision, every Starlink user may become a subscriber to xAI products and the Grok AI assistant in the future; every enterprise using Starlink's high - speed network may be converted into a traffic entry point for its digital advertising ecosystem; and the "AI computing satellites" deployed in orbit in the future will directly provide computing power services in space.

According to the prospectus, the total addressable market (TAM) for AI is ultimately as high as $26.5 trillion, which is 12 times the size of the aerospace and Starlink markets.

Among this, it includes $240 billion for space data centers, $76 billion for consumer subscriptions, $60 billion for digital advertising, and $2.27 trillion for enterprise - level AI applications.

This means that if calculated based on this market space, about 93% of SpaceX's future value will ultimately come from AI.

From this perspective, rather than calling SpaceX a space company, it's more like a super AI infrastructure company in a rocket - shaped shell.

This article is from the WeChat public account "Silicon - Based Observation Pro", written by Silicon - Based Jun, and published by 36Kr with permission.