Decoding SpaceX IPO: How Musk Integrates AI into Rockets

Contributing Author | Zhou Hengxing

“If you have a decisive technological advantage, you can win with minimal casualties.”

On a Saturday morning in July 2025, a Chinese AI entrepreneur walked into the headquarters of xAI in Palo Alto.

This is the heart of Silicon Valley, not far from Tesla’s headquarters. Unlike the old warehouses and industrial buildings favored by many tech companies in San Francisco, this place resembles a typical Silicon Valley R&D campus: low, restrained, with large glass facades supported by steel columns, outside are neatly trimmed bushes and quiet parking lots. Extending outward from this area are Stanford University, Sand Hill Road, AI laboratories, autonomous driving teams, and robotics companies. Compared to the neighborhoods in San Francisco that have been repeatedly transformed by the last internet boom, this is the true hinterland of Silicon Valley: calm, orderly, even a bit boring. The people here still believe that technology is not just applications on screens, but a force that can directly transform cars, factories, energy, robots, and the physical world.

In the United States, Elon Musk is no longer just the entrepreneur who “changes the world.” He is at the center of political controversy, a target of daily media pursuit and attack, and someone many Silicon Valley liberal elites no longer openly embrace. But in China, he still holds a rare prestige.

Musk has read Sun Tzu’s “The Art of War” many times, believing it is full of wisdom. He also added in a podcast that perhaps the book should include a chapter on “technology.” “If you have a decisive technological advantage, you can win with minimal casualties.”

In the AI era, the “technological advantage” Musk seeks increasingly concretely falls on talent. He has repeatedly expressed respect for Chinese tech companies and admiration for Chinese tech talent. Once, I forwarded him a video of a Chinese company's robot demonstration; he liked it and replied, “I have great respect for Chinese competitors. China has so many smart and hardworking people.” During the wave of startups in the AI industry last year, Musk also quickly noticed this group of young Chinese products.

That day, his two-hour conversation with this Chinese entrepreneur surprised him a bit—Musk was very relaxed and personable, quite different from the media persona. They discussed models, computing power, X’s data, and the future of human civilization.

The day after the conversation, Musk sent him an acquisition offer.

But the entrepreneur did not accept the olive branch on the spot. A few months later, he joined another Silicon Valley company.

At the time, this seemed just an unsuccessful recruitment attempt. But eight months later, it appeared as an early footnote.

On March 13, 2026, Musk posted a rare message on X. He said that xAI had “not built it right the first time” and was rebuilding from the ground up. He then apologized to those who had been wrongfully rejected or not even given interview opportunities over the past few years.

For someone who almost never publicly apologizes, these two posts were unusually stark; even more so given the timing. Just six weeks earlier, SpaceX had completed the acquisition of xAI entirely with stock: SpaceX valued at about $1 trillion, xAI at about $250 billion, and the combined valuation at $1.25 trillion. That is, Musk had just incorporated a $250 billion AI company into his rocket company, then told the world: this company was built wrong the first time.

More than two months later, on May 20, SpaceX filed an S-1 prospectus with the SEC, planning to list on Nasdaq under the ticker SPCX. According to the prospectus and market reports, SpaceX aimed to raise about $75 billion, seeking a valuation between $1.75 trillion and $2 trillion, with a tentative listing date of June 12—just two weeks before Musk’s 55th birthday.

The assets in SpaceX’s IPO prospectus include not only Falcon 9, Starlink, but also X’s data, xAI’s models, Colossus’s computing power, the vision of orbital data centers, and Musk’s long-standing AI ambitions after OpenAI.

This isn’t a carefully crafted capital magic trick from the start. It’s more like Musk gradually shifting the battlefield back to what he knows best— the physical world— after setbacks in the AI arena.

This is also the thread this article aims to follow: how Twitter, from a failed acquisition, became a data asset for xAI; how xAI, from Musk’s counterattack on OpenAI, turned into an internal mess with talent drain and product pressure; and how SpaceX, at this moment, took over xAI, turning the AI competition from software and models back into stories of rockets, satellites, chips, energy, and orbital data centers.

The Other Side of the $44 Billion Deal

On the afternoon of October 27, 2022, Musk walked into Twitter’s San Francisco headquarters holding a white washbasin, and posted on X: “Let that sink in.”

“Sink” here means both “to sediment” and “water basin.” It’s classic Musk humor: unpretentious but memorable.

At that time, the public focused on the pun and the washbasin, while Wall Street paid attention to the $44 billion price tag. Almost from the moment the deal closed, many analysts considered it Musk’s worst gamble in his career. It was originally a deal he tried to back out of. In April 2022, he signed the agreement; in July, he attempted to withdraw; then Twitter sued him in the Delaware Court of Chancery. Ultimately, he was forced back to the negotiating table to complete the acquisition at the original price.

In the first six months after the acquisition, all signs supported the judgment that “Musk bought wrong.” Twitter underwent massive layoffs, advertisers withdrew in droves, Fidelity repeatedly lowered X’s internal valuation report—at its lowest, below $10 billion, less than a quarter of the purchase price. To outsiders, Twitter had transformed from a global public square into a platform mired in political controversy, with declining ad revenue and a chaotic brand image.

But Musk didn’t just buy a social media company; he bought a real-time text generation system.

The daily arguments, news dissemination, political debates, code discussions, memes, rumors, clarifications, insults, and praise on X formed a data landscape that was not carefully curated but extremely close to authentic human language. From late 2022 to early 2023, one of the most lacking elements in the entire AI industry was precisely this kind of data.

At that time, almost everyone mocked Musk for paying too high a price for Twitter, but few realized that this deal could have another explanation in the AI era.

Five months later, on March 9, 2023, a company called xAI was registered in Nevada. It was five days before OpenAI released GPT-4.

Almost simultaneously, Musk signed a public letter initiated by the Future of Life Institute, calling for a pause on training AI systems more powerful than GPT-4. The letter later sparked controversy over signatures and stance, but one fact was clear: Musk was publicly urging the industry to slow down while secretly putting his own AI company on the road.

Initially, xAI appeared as a small team of over ten researchers and engineers from top labs like DeepMind, OpenAI, Google, and Microsoft. Igor Babuschkin from DeepMind, involved in AlphaStar; Jimmy Ba, co-author of Adam optimizer paper; Christian Szegedy from Google; Tony Wu focusing on reasoning.

Officially, xAI’s mission was “to understand the nature of the universe.” It sounded very Musk, and grand enough. But inside Silicon Valley, few ignored the deeper implication: eight years earlier, Musk and Sam Altman co-founded OpenAI. He invested funds, reputation, and resources, only to be gradually excluded from the company’s most critical power structure. When he looked back, OpenAI had transformed from a non-profit lab into a multi-billion-dollar AI empire.

Musk was not surpassed by a brand-new company. What surpassed him was what he had helped create himself.

Therefore, xAI is not only a new company but also a return journey.

The Illusion of Victory and Internal Fractures at xAI

In 2024, Colossus went live in Memphis, Tennessee. It was built inside an abandoned appliance factory. xAI later claimed that this supercomputing cluster was built in 122 days, then expanded to 200k H100 GPUs in 92 days. Reuters reported at the end of 2024 that Colossus had about 100k GPUs and planned further expansion. Whichever figure you accept, it’s an extremely Muskian engineering feat: stacking hardware first, then outpacing traditional data center build times with speed.

From any perspective, Colossus was a real engineering miracle. It proved Musk could still change the pace of hardware development with extreme speed. By the end of that year, he had three assets: X’s real-time text data, xAI’s research and engineering team, and Colossus’s computing power. Individually, each had issues; together, they already formed a new capital story.

On March 28, 2025, Musk announced on X: xAI would acquire X.

This deal, without using cash, shifted multiple narrative layers.

It rewrote X’s valuation story. On March 28, Musk announced that xAI would acquire X entirely with stock, providing figures: xAI valued at $80 billion, X at $33 billion. The latter roughly equals a $45 billion enterprise value minus about $12 billion in debt. A few months earlier, Fidelity’s valuation of X had been less than $10 billion. Through this deal, the story of “overpaying for Twitter” was re-priced within the AI context.

X’s debt was incorporated into xAI’s balance sheet, mixed with xAI’s high investments, transforming what was once a Twitter acquisition burden into a strategic integration of a “fast-growing AI company.”

More critically, xAI officially gained access to X’s real-time data and distribution scenarios.

Musk reintroduced the failed social media acquisition into the AI narrative. Twitter was no longer just a platform abandoned by advertisers, burdened with debt, and devalued. It became xAI’s data asset and application portal.

Since then, xAI’s funding curve continued upward. In summer 2025, xAI secured a $10 billion debt and equity financing, including debt financing and strategic equity investments. By January 2026, xAI announced the completion of a $20 billion Series E.

Capital still followed Musk, but internal fractures within the company grew more apparent.

In August 2025, Igor Babuschkin posted a long article on X announcing his departure from xAI. He didn’t specify reasons, only likened his farewell to “proud parents driving away their child to college.” This was polite, gentle, leaving much unsaid. Colossus was built by him, and after it became xAI’s most important engineering asset, he left.

In the following half-year, most of the original 11 co-founders who started with Musk had left; by March 2026, only engineers Manuel Kroiss and Ross Nordeen remained relatively low-profile.

For an AI company relying on top-tier research culture and long-term model accumulation, the continuous departure of founders was itself a signal.

Meanwhile, xAI’s products had not yet achieved market dominance. Grok failed to surpass Anthropic’s Claude Code and OpenAI’s Codex in the AI programming market. Claude Code drove rapid revenue growth for Anthropic, making xAI’s programming tools more passive. By March 2026, Musk even publicly admitted that xAI’s programming tools “cannot compete.”

If only looking at funding curves, xAI was becoming one of the most valuable AI companies globally; but in terms of products and team, it seemed more like an organization hollowing out.

By the end of 2025, Musk’s AI ambitions faced repeated setbacks. He had helped found OpenAI but was pushed out of its core power structure; when he re-established xAI, OpenAI had become the global AI tide’s center, and Anthropic was rapidly gaining advantages in programming. Although xAI had X’s data, Colossus’s compute power, and Musk’s personal capital appeal, it lacked overwhelming advantages in models and products. Worse, the original co-founders he recruited from top labs were gradually leaving.

This was not Musk’s familiar rhythm of victory. Over the past two decades, his greatest strength was turning seemingly impossible hardware goals into engineering problems: rockets that could be recovered, mass-produced electric vehicles, global satellite internet. But AI was different. The core competition in AI was not just compute and data, but also research culture, product judgment, and organizational stability.

Moving the AI Battlefield to Space

In December 2025, during an internal meeting at xAI, Musk proposed a new concept: orbital data centers.

He described a network of large satellites forming a computing power grid. These satellites would orbit close to Earth, with solar panels constantly facing the sun, providing power for AI training. His idea was that ground data centers faced power shortages, land approval issues, transformer queues, natural gas turbine permits, and cooling bottlenecks—all could be bypassed by moving compute into space.

The emergence of orbital data centers was not just a sudden sci-fi idea. It was more like Musk’s search for a new terrain after setbacks in the AI battlefield. On the ground, he faced OpenAI’s model advantages, Anthropic’s product breakthroughs, Google and Microsoft’s cloud infrastructure, and a team of xAI founders gradually leaving. But if the battlefield moved to space, the rules changed. Models would no longer be the only core; launch capacity, satellite manufacturing, orbital operations, solar power, and hardware engineering would become new barriers. And these were precisely Musk’s strengths, as well as SpaceX’s.

Musk was not suddenly discovering that AI should go to space; he was shifting the battlefield to where he had the best chance of winning after failing on the ground.

Things then accelerated. On January 20, 2026, Musk told a group of global elites at Davos that space would become the “least costly place to deploy AI,” and predicted it would happen within two to three years.

On January 30, SpaceX submitted an application to the FCC, requesting permission to launch and operate a “SpaceX Orbital Data Center system” with up to 1 million satellites. The application indicated these satellites would orbit at 500 to 2000 km altitude. On February 4, the FCC Space Bureau accepted the application for review. The document lacked detailed engineering plans but clearly indicated the direction: Musk aimed to redefine the future of AI from a model competition to an infrastructure race.

On February 2, SpaceX announced it would acquire xAI entirely with stock. According to market reports, SpaceX’s valuation was about $1 trillion, xAI’s about $250 billion, and the combined valuation $1.25 trillion. This was one of the largest mergers in history. It followed the same logic as the March 2025 xAI acquisition of X: all-stock deal, no cash, no external minority shareholders capable of changing the outcome, as both buyer and seller were ultimately controlled by Musk.

But this time, the significance was greater. xAI was no longer an independent company; it became a division of SpaceX. Grok, X, Colossus, the orbital data centers, and all AI commitments related to xAI would be incorporated into SpaceX’s IPO prospectus and valued by the public markets.

xAI’s challenge to catch up with OpenAI and Anthropic on the ground AI battlefield was growing harder; to gain new narrative space, it had to shift competition from model capabilities to hardware and infrastructure. Once the battlefield moved to space and compute infrastructure, SpaceX became the most suitable shell. Thus, xAI was merged into SpaceX, which, along with its satellite internet and rocket business, prepared for an IPO.

On February 25, Oltmann was asked at a press conference in New Delhi about this idea. He almost laughed and said that putting data centers into space was “ridiculous” given current realities. He explained that the costs of launching and repairing GPUs in space were prohibitive. He said, “It’s a long way off.”

This was a direct clash of two worldviews. Oltmann represented the cost calculations, product rhythm, and model ecosystem of the software world; Musk represented the push of hardware—building systems first, then lowering costs through scale and engineering iterations. Musk had often won such debates in the rocket and EV eras. But whether AI was the same problem remained to be seen.

Before this debate was resolved, Musk created the greatest uncertainty for xAI’s story. On March 13, 2026, he posted on X that xAI had “not built it right the first time” and was rebuilding from the ground up, apologizing to those who had been wrongfully rejected in recent years. A newly valued $250 billion AI company was suddenly publicly admitting “it was built wrong.”

From an outside perspective, this seemed like a frank confession; from a capital narrative, it was also a risk transfer. xAI’s failure was no longer solely its own; it was embedded into SpaceX’s larger story. Musk no longer needed to prove that xAI could beat OpenAI and Anthropic as an independent AI company; he only needed to show that within SpaceX’s hardware ecosystem, xAI could still become part of a larger future infrastructure network.

Eight days later, he threw out another hardware story: TeraFab.

This was Tesla and SpaceX’s advanced chip manufacturing plan, initially described as a roughly $20 billion project. Musk later said Tesla and SpaceX would build two advanced chip factories. The details remained uncertain, and SpaceX’s S-1 hinted that TeraFab might not succeed; the company still heavily relied on external chip suppliers.

Relying solely on SpaceX’s orbital data center needs would hardly support such a massive chip manufacturing plan. Only by including Tesla’s FSD chips, Optimus robot chips, and Dojo training chips could the factory’s demand story be complete. Financially, too: although Starlink had become a cash cow, supporting long-term chip investments still depended on Tesla’s revenue scale and manufacturing capacity.

Thus, TeraFab appeared as a chip factory on the surface, but in reality, it further welded Tesla and SpaceX’s fates together, providing a new narrative foundation for Musk’s trillion-dollar compensation package challenged by shareholders and courts: Tesla was no longer just an electric car company but part of AI, robotics, chips, and space infrastructure.

This “layered” industrial layout was more like a counterattack built step-by-step after xAI’s setbacks: if xAI couldn’t beat OpenAI and Anthropic alone, then put X’s data in; if data and compute power still weren’t enough, merge xAI into SpaceX, turning AI competition from models into space infrastructure; and if SpaceX’s orbital data centers still needed chips, cash flow, and manufacturing, then bring Tesla into the story. Each merger was like reinterpreting unresolved issues from the previous stage within a larger narrative.

The Three-Stage Rocket in SpaceX’s IPO Prospectus

On May 20, 2026, SpaceX submitted an S-1 prospectus to the SEC. The 277-page document planned to list on Nasdaq under the ticker SPCX, aiming to raise about $75 billion, with a target valuation of $1.75 trillion to $2 trillion. The IPO was scheduled for June 12, potentially surpassing Saudi Aramco’s 2019 IPO record.

The prospectus divided SpaceX into three segments: Space, Connectivity, and AI. On the surface, these appeared as three parallel business lines; but in detail, they resembled a three-stage rocket. The first stage was launch services, providing the most basic engineering capability; the second was Starlink, turning launch capacity into cash flow; the third was AI, pushing cash flow, compute, and space infrastructure into a more distant and uncertain future.

The first segment was Space, the launch business. By 2025, SpaceX’s Falcon 9 launch frequency and reusability had far surpassed traditional competitors. It lowered commercial launch prices to levels difficult for traditional providers to match, gradually capturing most of the global commercial launch market. But more noteworthy was that one of SpaceX’s biggest customers was itself: many Falcon 9 launches were used to deploy Starlink satellites. This created a closed loop that external competitors found hard to copy—subscriber fees from Starlink supported launch costs, more satellites meant more subscribers, and more subscribers supported more launches.

The second segment was Connectivity, i.e., Starlink. This was the real cash machine of the entire SpaceX empire. According to the prospectus and market reports, Starlink had become one of SpaceX’s main revenue sources and contributed most of the group’s profitability. In 2025, SpaceX’s combined revenue was about $18.7 billion, with Starlink as the primary growth engine. In other words, Starlink transformed a conglomerate with loss-making AI businesses and X’s debt into a seemingly profitable company.

Starlink was a genuine industrial achievement. OneWeb had tried satellite internet and gone bankrupt; Iridium had succeeded but mainly in narrowband communication. Starlink delivered global high-speed broadband, with terminal costs low enough for the US military, shipping companies, remote villages, islands, and front-line war zones.

But Starlink also faced pressures. As user numbers grew, it needed to lower prices, subsidize terminals, and expand capacity in more countries and complex markets. It was trading price and capital expenditure for scale. For an independent company, this might be normal expansion; but for a group still supporting high-investment businesses like AI, chips, and orbital data centers, whether Starlink’s cash flow could sustain internal high investments long-term was a question investors had to consider.

The third segment was AI, i.e., xAI plus X. This was the most imaginative and most money-burning layer. The prospectus showed that AI had become one of SpaceX’s largest capital expenditures, but it had yet to prove comparable profitability to Starlink. The outer two layers represented Musk’s most genuine engineering achievements over the past two decades: rockets really fly, satellites really orbit, users really pay, and government contracts are real; but the deeper you go, the closer you get to an unproven gamble. AI was money-burning, and orbital data centers were even more speculative.

The prospectus used very cautious language when discussing the “Orbital AI Data Center.” A few months earlier, Musk publicly said that putting AI compute into space was “almost obviously the right thing to do”; but in the S-1, lawyers wrote that this business “may never be commercially viable.” Musk’s legal team was warning future investors: this story was grand, but the company could not guarantee it.

The prospectus also disclosed SpaceX’s control structure. The company used dual-class shares: Class A with one vote per share, Class B with ten votes per share. Musk held a significant economic stake but had absolute voting power advantage. It also stated Musk could only be removed by Class B shareholders, which were essentially controlled by him. Even after going public, SpaceX remained a company designed around Musk’s control.

This was also the most honest part of the SpaceX IPO: it was not just selling Falcon 9 launch capacity and Starlink cash flow, but Musk himself. Investors were really pricing whether he could reshape the next twenty years of infrastructure.

The IPO’s Three-Front Battle

On May 18, 2026, the Oakland Federal Court.

Musk’s lawsuit against Altman and OpenAI entered its third week. Nine jurors took less than two hours to rule: Musk’s claims were beyond the statute of limitations. Judge Yvonne Gonzalez Rogers then adopted the jury’s recommendation, dismissing his case.

The jury did not decide whether Altman and Brockman violated their promise to develop AI “for the benefit of humanity.” The answer was colder and more technical: Musk sued too late.

In the final days of the trial, Musk, along with Trump-led business delegations, visited Beijing. Back in the US, he announced on X that he would appeal, claiming the court’s decision was based solely on the statute of limitations, not the merits.

Two days later, SpaceX officially filed its IPO prospectus.

His lawsuit against OpenAI was like a pre-IPO counterattack: if the mission controversy was dragged back into court, its listing narrative would be embroiled in debates over founding promises, control, and commercialization paths.

The most important companies of the AI era were fighting for the same capital market gate.

SpaceX was leading. It had already filed the S-1, planning to list under SPCX on Nasdaq, aiming to raise about $75 billion, with a valuation around $1.75 trillion. According to the latest Reuters report, SpaceX set the IPO price at $135 per share, with trading expected to start on June 12.

OpenAI and Anthropic almost simultaneously followed. In late May, media reports revealed OpenAI was working with Goldman Sachs and Morgan Stanley on a confidential IPO prospectus, targeting a $1 trillion valuation and accelerating plans. On June 1, Anthropic announced it had just completed a new $65 billion funding round, with a valuation close to $965 billion, and had filed a confidential IPO application with the SEC.

Altogether, the potential fundraising exceeded $200 billion— the densest IPO window in history.

Who goes public first will set the valuation anchor for the public market. SpaceX’s early S-1 submission was, in a way, proactively establishing a reference point: $1.75 trillion, the price for hardware and infrastructure in the AI era. OpenAI followed with nearly 10k monthly active users and the label “AI product king”; Anthropic brought a clear path to profitability and a dominant position with enterprise clients.

The relationship among the three was more than “competitors.” Musk was a co-founder of OpenAI, later ousted, turned against, and sued. Anthropic’s founder, Dario Amodei, left OpenAI and founded Anthropic, itself a product of internal splits. The rise of Claude and Claude Code made xAI passive in the programming tools market. More subtly, Musk’s SpaceX IPO prospectus listed Anthropic as a competitor of xAI but also disclosed that Anthropic rented Colossus’s compute power.

According to the prospectus, if this arrangement continued until May 2029, Anthropic would pay about $1.25 billion per month to rent Colossus, totaling over $45 billion. A few months earlier, Musk had publicly criticized Anthropic on X; now, Anthropic was one of xAI’s most important clients.

The subtlety was: xAI failed to beat Anthropic in products but had already rented its compute to the competitor. Colossus, once a symbol of xAI’s engineering myth, had become the infrastructure of a rival. Musk later said on X that the current lease was only a 180-day agreement, with both sides holding a 90-day cancellation right. It was not a highly certain long-term contract, but it was already one of the most prominent sources of revenue in SpaceX’s AI income curve.

The IPO prospectus was cautious about this transaction, stating it allowed the company to monetize unused infrastructure. The phrase “unused compute” was key, admitting that xAI had not fully utilized Colossus’s capacity. For investors, this showed xAI’s ability to turn hardware into revenue but also raised a sharper question: if an AI company’s most valuable asset was not its models but the compute power it could rent to competitors, was it a model company or an infrastructure company?

The three IPOs represented three diverging futures from the same origin: one was ChatGPT, one was Claude, and one was xAI embedded in a rocket company. Each carried part of the original DNA, competing for inheritance in the same market.

The concern among analysts was liquidity. About $8 trillion sat in money market funds; SpaceX’s $75 billion raise was only about 1% of that. The space seemed ample. But as hundreds of billions flowed into new stocks over a few months, institutional portfolios had to rebalance—buying SpaceX, OpenAI, and Anthropic meant reallocating from elsewhere. That “elsewhere” was likely Nvidia, Microsoft, Google, Apple.

The outcome of Musk’s lawsuit, from another perspective, actually helped OpenAI. The jury’s decision not to rule on whether OpenAI betrayed its founding mission simply cleared the biggest legal uncertainty before IPO—if Musk had won, OpenAI’s restructuring and IPO path could have faced greater unpredictability.

Musk lost a round in Oakland, but this loss, in a sense, made his SpaceX IPO cleaner—if the lawsuit dragged on until the end of the year, it would also become a question investors asked during roadshows.

Since that dinner at the glamorous hotel in 2015, these three companies had traveled eleven years. Now, eleven years later, they lined up at the same capital market gate.

The Final Trump Card

After SpaceX’s IPO prospectus was made public, Wall Street quickly began discussing a bigger question: if Musk had already put X into xAI, and then xAI into SpaceX, could Tesla be next?

This question remains speculative. It’s based on analyst predictions, media reports, and internal discussions, not an actual transaction. It fits Musk’s recurring pattern over recent years: when a company’s story becomes too small or needs new assets, cash flow, or control arrangements, he tries to reconfigure them.

X provided data for xAI; xAI gave SpaceX an AI story; SpaceX provided rockets, satellites, and orbital data centers. What could Tesla contribute?

The answer is the gateway to real-world AI.

Tesla is not just an electric car company. At least in Musk’s narrative, it has long been a robotics company, an energy company, a chip demand source, and a real-world data portal. FSD, Optimus, Dojo, fleet data, batteries, factories, and global supply chains could all be integrated into the same AI story. The TeraFab plan further linked Tesla and SpaceX’s chip supply chain: if future AI compute required its own chip manufacturing system, relying solely on SpaceX’s orbital data center needs would be insufficient; Tesla’s robot, autonomous driving, and training chip demands had to be included for the story to be complete.

But Tesla was also one of Musk’s weakest-controlled companies. It was a publicly listed company, with no dual-class shares—each share one vote. Musk remained the soul of the company but held only about 12.8% of shares. In 2024, he publicly stated on X that if he could not hold about 25% of voting rights in Tesla, he would not continue pushing AI and robotics inside the company.

That statement was a pressure on the board and shareholders at the time. Looking at it before the SpaceX IPO, it revealed a long-term structural issue: Musk wanted Tesla to play a role far beyond a car company, but its governance remained that of a typical public company.

SpaceX was entirely different. The IPO prospectus showed that even after going public, SpaceX would retain a dual-class structure: Class A with one vote per share, Class B with ten votes per share. Musk would have absolute voting power, and could only be removed by Class B shareholders, which were essentially controlled by him.

If the two merged, Tesla’s cars, FSD, Optimus, Dojo, cash flow, and manufacturing capacity could all be folded into a Musk-controlled superstructure. Publicly, it would be explained as an integration of AI, robotics, chips, energy, and space infrastructure; internally, it could also reshape Musk’s control over the entire empire.

In Musk’s narrative, Tesla was the gateway to real-world AI, the predecessor of robotics, part of the energy network, and a major source of future chip demand; SpaceX provided launch capacity, satellite networks, and orbital infrastructure. Add in xAI’s models, X’s data, and Colossus’s compute power, and a superstructure controlled by Musk gradually took shape: cars, robots, factories, and chips on the ground; satellites, communication networks, and future data centers in orbit; connected by AI models, user data, and energy demands.

Final Words

In 2013, I first visited Tesla’s factory in Fremont, California. The factory was filled with red robotic arms, and cars moved slowly along the assembly line. Sitting across from me was Musk, then just 42. He told a story: on Earth, SolarCity produced solar energy, Tesla stored and consumed clean energy to reduce reliance on fossil fuels; beyond Earth, SpaceX built reusable rockets and starships to send humans to Mars, providing a “multi-planetary backup” for civilization.

At the time, I was new to the industry and couldn’t tell if he was talking about a business plan or science fiction. Thirteen years later, looking back, many of Musk’s seemingly “boastful” words have mostly become reality.

But AI has introduced new cracks into this story.

In the hardware world, Musk’s strength was always breaking down seemingly impossible engineering goals into parts, using supply chains, organizational strength, cost control, and tight schedules to push reality. But AI isn’t just that. It also requires research culture, model training experience, product sensitivity, and a top-tier research team willing to stay long-term.

Musk seems to dislike the term “researcher.” He often says his companies only have “engineers.” The problem with xAI was precisely here: researchers don’t seem to like it. Musk has obtained X’s data, Colossus’s compute power, and the market’s valuation, but three years later, most of the original co-founders had left, Grok failed to beat Claude Code and Codex in programming, and Musk himself had to publicly admit that xAI was “not built right the first time.”

xAI’s setbacks remind us that in the AI era, the most scarce resources are not just hardware. They also include organizational order, research judgment, and understanding of software product rhythm. Musk can build Colossus, design orbital data centers, and plan TeraFab, but he still needs to prove he can turn these hardware advantages into truly leading AI capabilities.

xAI did not become a real rival to OpenAI as he expected, so he didn’t continue to push it as an independent AI company to the market. Instead, he merged it into SpaceX, letting rockets, satellites, orbital data centers, and future chip factories share its valuation.

In May 2026, Musk’s friend, XPRIZE founder Peter H. Diamandis, said on X: “He doesn’t need to run the best AI models; he needs to control the best hardware.” Musk replied: “You’re right. But our AI will eventually become great. Whether it can be the strongest remains to be seen; but I will never give up.”

That reply was very Musk. He didn’t fully deny failure, nor did he dwell on it. He simply extended the timeline, changed the reference point, and reinserted the problem into his familiar narrative. “xAI is only 3 years old, half the age of Anthropic, a quarter of OpenAI. Let’s see in three years.” He also reminded Diamandis that when SpaceX was founded three years earlier, it hadn’t achieved anything worth bragging about; after six years, it experienced three consecutive launch failures, nearly burned through all its money, and was almost declared dead. The story only began to change after the fourth successful launch.

In my thirteen years of following Musk, I have always held a complex feeling towards him. Of course, Musk makes mistakes—often big ones—but I still respect him because he embodies a rare engineering romanticism of this era: he truly believes the physical world can be remade, and is willing to stake his companies, wealth, reputation, and time on seemingly absurd engineering goals.

In the month of SpaceX’s IPO, Musk turned 55. Steve Jobs was approaching the end of his life at that age; Huang Renxun, at 55, was experiencing a dark hour, and Wall Street mocked him for the wrong direction.

Many of Musk’s stories, before being proven, seemed like overhyped imagination. But many times, he turned near-fantastical visions into emotionally resonant moments in human engineering history.

I often recall a scene: on April 8, 2016, Falcon 9 executed the CRS-8 mission to resupply the International Space Station. After separation, the first stage re-entered the atmosphere, ignited to slow down, and then, as if by magic, landed steadily on a drone ship in the Atlantic. It was humanity’s first successful recovery of an orbital rocket booster at sea.

The drone ship used for sea recovery was called “Of Course I Still Love You,” named after a spaceship in the science fiction novels of Iain M. Banks. Musk inscribed such a name on a drifting drone ship, waiting for a rocket, several meters tall and just returned from space, to land—this act itself was very much like him: half engineering, half science fiction; half precise calculation, half childhood fantasy that refused to die.