Flying into the universe, vast and boundless.

Written by: siqi

On June 12, on the day SpaceX officially went public, Musk chose to go to the Starship base in Texas and, together with hundreds of employees, rang the Nasdaq opening bell remotely.

At the event, he humorously remarked, "If someone had told me years ago that this day would come, I would probably think that person was high. Because back then, I thought this company would fail."

On this day, SpaceX officially landed on Nasdaq with an issue price of 135 dollars, raising about 75 billion dollars; it opened with a rise, and at one point surged over 176 dollars, briefly reaching a market value of 2 trillion dollars.

From being forced to start a company in 2002 due to the inability to buy rockets, to completing the largest IPO in human commercial history today, this 24-year journey is filled with numerous counterintuitive and consensus-defying stories.

This company nominally builds rockets, but the rocket business does not make money; its most notable achievement is recovering rockets, yet its valuation is supported by two other stories — Starlink, and the "space computing power" that has just been included in its prospectus.

We have compiled 15 of the most representative stories to help you develop a more comprehensive understanding of SpaceX.

1. The starting point of SpaceX came from a "public relations act" of a business newcomer

In 2001, Musk, who had just cashed out from PayPal, wanted to personally invest in a public relations project called "Mars Oasis": spending 10 to 20 million dollars to send a mini greenhouse to Mars, taking photos of green plants growing on red soil, to persuade Congress to increase NASA's budget.

But he was stymied by shipping costs. European rockets were too expensive, and his attempts to buy retired intercontinental missiles in Moscow were dismissed as an outsider.

In subsequent public speeches and media interviews, Musk stated that after facing setbacks, he believed that what was holding back humanity's journey to Mars was not public will or congressional budgets, but the price of rockets themselves. Thus "helping NASA raise funds" transformed into "making rockets cheap by himself."

In 2002, SpaceX was officially founded.

2. The first six years, the company was in a "failure" mode

From 2002 to 2008, the first three launches of Falcon 1 all failed.

During that era, all the know-how for building rockets was locked within the national aerospace system. SpaceX couldn’t buy blueprints or hire people. Musk later humorously noted that he became the company's chief engineer because "no good people were willing to come."

Even harsher were the physical properties of rockets: they could not be thoroughly tested on the ground; the only way to learn was to launch, explode, and try again. The "triple failure" was the tuition that a company paid to learn aerospace with live ammunition — but this tuition came at tens of millions of dollars, whereas Musk's money was enough for only four tries.

3. The fourth launch succeeded, opening the "commercial space" era

On September 28, 2008, the fourth launch of Falcon 1 succeeded — the first liquid-fueled rocket developed with private funding entered Earth orbit.

Before this, "space" was assumed to be a government game: money came from the government, and the work was done by the system.

Three months later, NASA awarded a 1.6 billion dollar International Space Station cargo contract (CRS) to this company that had just survived a near-failure. The day marked the official birth of "commercial space" as an industry.

4. New play in commercial space

Traditional aerospace procurement was "cost-plus": contractors charged as much as they spent, with the government adding a profit margin — the more you spent, the more you earned, and no one had the incentive to save money.

Nasa provided SpaceX with a fixed-price contract in the commercial cargo program (COTS/CRS): a flat rate, and any savings were yours, while any overruns were borne by them. This seemingly tedious procurement clause was the real starting point for an institutional transformation in commercial space, as it made "making rockets cheap" a profitable business for the first time.

The enduring cost obsession that SpaceX later maintained was half nature and half driven by this contract.

5. Reusable technology: Getting the client to willingly pay for "unreliable technology"

On December 21, 2015, the Falcon 9 first stage successfully landed back on Earth for the first time — exactly 13 years after the company was founded.

Before this, SpaceX's obsession with recovery went through a long process of experiments and failures: during the first two flights of Falcon 9 in 2010, attempts were made to recover the first stage with parachutes — the rocket disintegrated before it even deployed the parachute upon re-entering the atmosphere. From 2013 onwards, they switched to a retro-thrust scheme, making nearly ten attempts over the next two years: some hard crashed into the sea, some exploded or tipped over on barge decks, and none returned intact.

But most of these experiments were not self-funded tests; rather, they were conducted during customer-paid launches — the same rocket performed both mission and experimentation. The customer's payload was sent into orbit in the first half of the mission, satisfying the financial exchange; the first stage rocket, after delivering the cargo, was in accordance with industry practices viewed as trash thrown into the ocean, allowing SpaceX to practice landings on the way back.

Musk's "calculation" was: if it blew up, it was just garbage; if it worked, it would rewrite space history. Thus, SpaceX effectively used NASA's orders as grants, completing their education in "reuse" for free. Now, the Falcon 9 has a success rate of about 99.4%, with 165 launches and recoveries planned for 2025, having only lost 3 times.

Falcon 9 rocket performing commercial launch | Image source: SpaceX

6. Today’s SpaceX: Starlink making money to sustain AI

The prospectus shows that in 2025, SpaceX's total revenue will be 18.7 billion dollars, with a net loss of 4.9 billion dollars.

But when broken down by segment, the story is entirely different: the connectivity segment where Starlink is located contributes about 4.4 billion dollars in operating profit per year, being the only profitable section of the company; the space segment associated with rockets incurs a small loss of about 660 million — mainly due to an investment of about 3 billion dollars in Starship development.

The real black hole is the consolidated xAI: it incurs an annual operating loss of about 6.4 billion dollars, consuming all profits from Starlink and more.

In other words, if looking only at the "old SpaceX" (rockets + Starlink), it is already a profitable company; what brings it back to "loss" is instead the investment in AI for the next story.

7. Starlink is Musk’s "internal client" that prepares for reusable rockets

In January 2015, Musk publicly announced the Starlink plan, which consists of thousands of low-orbit small satellites that make up a "broadband network in the sky," selling internet service to ground users — especially in places at sea, wilderness, and remote areas where fiber optic cables and base stations cannot reach.

In December of the same year, Falcon 9 successfully landed for the first time. This means that before "cheap rockets" were proven, "customers for cheap rockets" had already initiated internal projects.

This was not a coincidence, but two halves of the same arithmetic problem: the global rocket launch market is only worth 5 to 6 billion dollars per year, and has not changed significantly in the past decade. Therefore, cheap lift capacity in this market cannot be satisfied; conversely, to lay out a network of several thousand to tens of thousands of satellites globally, there is no way to calculate without cheap lift capacity.

8. Starship has not yet succeeded, but its "buyer market" has already gone through a round

A similar early bet is happening with the next generation heavy rocket, Starship.

In 2014, SpaceX broke ground for the Starship base in Boca Chica, Texas — that was the year when Falcon 9 hadn't yet successfully recovered a single time. The previous generation had not landed, yet construction on the next generation had already begun.

What’s more noteworthy is the change of clients: the original narrative of Starship was "humans" — Mars colonization, space travel, a theme Musk had talked about for many years; but after the concept of space computing power emerged, the primary clients for Starship quietly shifted to "data centers."

The logic remains unchanged: Falcon 9 has a near Earth orbit capacity of about 20 tons, with Starlink as the client; Starship is planned for a capacity of 100-150 tons (near Earth orbit, planned value), which tourists cannot fully utilize, but the equipment needed by space data centers might.

With each increase in rocket size, Musk has to "create" a larger commercial client for them.

Starship V3 launches for the first time, carrying a total of 33 Raptor V3 engines | Image source: SpaceX

9. "Chopsticks holding rockets"

On October 13, 2024, during the fifth test flight of Starship, two mechanical arms on the launch tower grasped the slowly descending booster mid-air, flooding the internet.

Previously, Falcon 9 proved that rockets could "return" and also "fly again" — but every time they needed to go out to sea for recovery and return to the factory for renovations, with cycles counting in weeks, fundamentally it was still about "repair and reuse." Whereas Starship aims for something else: like an airplane, landing, inspection, refueling, and launching again.

The landing legs are dead weight, taking up payload capacity; if they land far away, it requires transportation. Allowing boosters to return directly to the embrace of the launch tower means that where they land is where they'll take off again — squeezing the intermediate steps to their limits, changing the turnaround goal from "weeks" to "hours."

The so-called "chopsticks holding rockets" actually points out what SpaceX sees as the ultimate form of rockets: from being recoverable, to "operating like flight schedules."

10. No need for a "domestic Starlink," but definitely a need for "domestic lift capacity"

The "Chinese version of Starlink" is a popular narrative, but there’s a commonly overlooked fact: Starlink solves the problem of "areas inaccessible by ground base stations" — the sea, wilderness, and sparsely populated areas; whereas China happens to have the strongest ground communication network in global coverage, so the experience of Starlink-like services domestically is inherently limited.

The real proposition lies at another level: satellites have many uses beyond communication — remote sensing, navigation, and future space computing power; all require sending large quantities of things cheaply and frequently into space.

In other words, China can avoid duplicating the "product" of Starlink, but cannot bypass the "lift capacity" behind Starlink. For Chinese commercial space, the question of "whether to have a network in the sky" is not the most central issue; "whether there are hands to weave the net" is.

11. Breaking the "never IPO" pledge

SpaceX was once Silicon Valley's most resolute "never IPO" company. Musk's public reasoning was that the short-termism of capital markets was incompatible with the long-term goals of Mars.

A turning point occurred in the fourth quarter of last year: Starlink's user growth and revenue per user show cap limits, while the capital expenditure for this new story, "space computing power," is so large that only the public market can bear it.

The prospectus disclosed that just in the first quarter of 2026, capital expenditures for AI businesses would exceed the combined total of the space and connectivity segments.

Therefore, going public is not a celebratory end, but a financing action for the next round of heavy bets.

12. Space computing power is a "consensus," but the details remain unknown

While the concept of space computing power is new, it indeed represents a super consensus quickly reached in the tech industry over the past six months, with hardly anyone publicly dissenting.

However, when questioned further, almost all technical details seem to lack common answers:

What do space data centers look like? There is no public product definition. What data will it process, and where does that data come from? Nobody knows.

Using the classic three elements from the AI industry — algorithms are running wildly on the ground, but "data" and "computing power deployment" are still blank in the context of space. Pre-training or inference? The two have completely different requirements for power supply, cooling, and networking, and correspondingly different satellite designs. A direction priced at trillions lacks even a consolidated product form.

Of course, looking from another angle, this just means that there are still many empty seats at the table yet to be occupied.

13. Silicon Valley has poured real gold and silver into "space computing power"

Silicon Valley's major giants have not only verbally supported space computing power.

Musk reshuffled assets involving trillions and merged SpaceX with xAI. The prospectus states that the deployment of on-orbit data centers is slated for as early as 2028.

Google has initiated the "Suncatcher Project": publishing technical papers and planning to launch two prototype satellites equipped with self-developed TPUs, while negotiating launch contracts with SpaceX.

Bezos's Blue Origin submitted an application to the U.S. Federal Communications Commission in March 2026 for the "Daybreak Plan" involving 51,600 data center satellites.

Former Google CEO Schmidt acquired the rocket company Relativity Space in 2025, openly stating that the purpose was to send data centers into orbit.

NVIDIA-backed Starcloud has already delivered an H100 chip into orbit and completed in-orbit model training by November 2025.

Acquiring companies, merging assets, filing for licenses, and launching satellites — the infrastructure race has already begun.

Silicon Valley startup Starcloud puts NVIDIA H100 GPU into orbit in November 2025 | Image source: Starcloud

14. The cold cost calculations

A space engineer has publicly estimated that building a 1 GW orbital data center (about 4,300 satellites, including five years of operation) would cost over 50 billion dollars — approximately three times that of an equivalent ground installation.

To overturn this calculation, the industry commonly estimates that the launch cost must be reduced to about 200 dollars per kilogram, but today Falcon 9's corresponding launch cost (note that "launch cost" and "entry cost" are not entirely the same) is about 2,000 to 3,000 dollars per kilogram. The two differ by at least one order of magnitude.

This order of magnitude requires a secondary reusable heavy rocket like Starship to make up for cost reductions. Therefore, the timeline for space computing power is deeply tied to the progress of Starship. The truth of the story will ultimately be tested with welders and launch pads.

15. A bigger story

Looking back at SpaceX’s growth history, it was forced to become a side character due to being unable to buy cheap rockets; then it created its own client (Starlink); now it has booked a larger client (space computing power) for the next generation rocket.

Over the past 24 years, it has turned two stories about rocket reuse technology and Starlink that no one believed in before into reality. Today, the yet-to-be-fulfilled promise of space computing power has appeared on the public market with a target valuation of about 1.75 trillion dollars.

This story is larger than the previous ones, and the price is also higher.