Musk remains Musk; his software prowess still falls short. Yet, Musk is no ordinary individual; he transcends the conventional.

Article by | She Zongming

Musk has encountered setbacks—in the realm of AI large models.

From its high-profile inception in July 2023 to its ignoble exit in May 2026, xAI's lifespan spanned less than three years.

In March of this year, Musk conceded his error on X, stating, "xAI was flawed from the outset," but added, "Now, we're rebuilding from the ground up."

▲Musk admits his mistake.

However, another luminary, Turing Award winner Yann LeCun, cast doubt on his endeavor. On June 18, in an interview with CNBC, LeCun remarked, "Frankly, xAI has failed because the founding team has departed."

Musk has triumphed—in the commercial spaceflight arena.

Following its monumental IPO, SpaceX's market value soared past $2 trillion on its debut, positioning it among the world's most valuable companies.

Even LeCun, after directly dismissing xAI, acknowledged SpaceX's impressiveness: "SpaceX is performing exceptionally well."

What does this signify?

Allow me to provide a more direct, candid, and succinct response than Doubao:

Musk remains Musk; his software skills still lag. Yet, Musk is no ordinary person; he defies the norm.

01 /

The past month or so has likely held significant importance for Musk, marked by a rapid succession of major events:

The dissolution of xAI and its integration into SpaceX;

Leasing 220,000 NVIDIA GPUs to competitor Anthropic;

Losing the century-long lawsuit "Musk v. OpenAI";

The successful inaugural test flight of Starship V3;

SpaceX's seamless listing, with its market value once surpassing $3 trillion;

SpaceX announcing the acquisition of Cursor's parent company for $60 billion...

Viewing these events collectively, they metaphorically represent a highly condensed version of Musk's entrepreneurial journey:

On one hand, he repeatedly falters in the software domain, where he lacks strength; on the other, he conquers new frontiers in the hardware realm, where he excels.

Take xAI, for instance. When Musk founded it in July 2023, xAI embarked on an auspicious path.

Led by the world's wealthiest individual and boasting a star-studded founding team of 11 top AI scientists from DeepMind, Google, and OpenAI, one could sense the formidable power of Musk's "Avengers against OpenAI."

Coupled with the massive real-time social data from the X platform and the 220,000 NVIDIA GPU computing cluster Colossus 1, OpenAI would weep, and Anthropic would fall silent.

At its zenith, xAI's valuation skyrocketed to $250 billion, with over $40 billion in funding.

Yet, in the end, Grok's ambitions were lofty, but its fate was... thinner than the GPT, Claude, and Gemini series.

Unless the "Big Three of Silicon Valley" expands to include a fourth, xAI will remain a "first-tier outsider."

▲All is fleeting, just a passing spark.

The founding team also disintegrated: within three years, everyone except Musk left, without exception.

What does it mean to be a lone wolf? This is it.

In a fit of frustration, Musk disbanded xAI, filled with "disappointment over its failure to meet expectations."

He also leased Colossus 1, which had a utilization rate of less than 11%, to the computing-hungry Anthropic—even though three months earlier, he had criticized Anthropic for having warped values and being "evil." He might as well have added to the xAI team: "Given the opportunity (computing power), you couldn't deliver."

▲Calling Anthropic evil in competition, but saying it's "beneficial to humanity" in cooperation?

However, a closer examination of Musk's software endeavors reveals that xAI's failure was no accident.

He created X.com, the predecessor of PayPal, but was ultimately ousted by the board due to technical architecture flaws and internal conflicts, and X.com's self-developed software system was completely abandoned.

He aspired to transform the X platform (formerly Twitter), which he acquired at great cost, into another WeChat, but the reality was: the stability of basic services declined sharply, and the paid blue verification logic failed...

In the pure digital realm, Musk is certainly not an "omnipotent deity."

But true visionaries never dwell on past explosions. Musk could easily say, "Let's compare hardware," and then add, "SpaceX + Tesla + Neuralink, what do you think?"

Just the "chopsticks catching rocket" maneuver of Starship is enough to inspire endless admiration from countless successors.

02 /

The question arises: Why can Musk, the contemporary world's top hardware engineering innovator, excel in rockets, electric cars, and energy storage but falter in AI?

Yet, this is a flawed question. The reasons for someone's success in certain areas often conceal the reasons for their failure in others. Bits and atoms are fundamentally different.

Examining all of Musk's successful projects, they are all concentrated in the hardware engineering field, where physical boundaries are clear, rules are defined, and progress can be quantitatively measured and iterated. The first principles are his unbeatable weapon.

The core of this methodology is to step outside industry conventions, break down complex systems into fundamental physical units, and thoroughly reconstruct cost curves through vertical integration, simplified structures, and large-scale production.

Must rockets use high-end aluminum-lithium alloys as the only material? Must aerospace components be niche, high-priced, and disposable?

Musk declared: I won't use expensive carbon fiber; I'll use low-cost stainless steel to build the Starship's body, self-develop 80% of the core components, and implement vertical recovery and reuse of the first-stage rocket. What do you think?

The result is that SpaceX has drastically reduced launch costs through rocket reuse, with the Falcon 9's first-stage booster surpassing 30 reuses. If Yang Dong from "The Three-Body Problem" were an aerospace scientist, she might say: Traditional aerospace science has failed.

Tesla has reshaped automotive manufacturing with integrated casting and vertical integration, and Neuralink's microelectrodes, implantation surgical equipment, and chip modules have entered mass production—all victories for hardware.

Simply put, rocket thrust, material strength, and thermodynamic limits are all within Musk's expertise.

In fact, all hardware projects share a common trait: physical boundaries are clear, failure causes are traceable, and optimization paths are quantifiable.

Musk's "Five-Step Algorithm" starts by questioning requirements, precisely eliminating redundant components and processes, and using industrial-scale production to reduce costs... It's like a fish in water in the physical world.

▲Musk is becoming the top spokesperson for "first principles."

But AI is not governed by Newtonian physics; it's a chaotic field with no fixed formulas, no clear decomposition, and no linear optimization paths. Intelligence emerges uncontrollably from massive data training.

Asking Musk to succeed in xAI as he does in SpaceX is like asking Sun Wukong to fight Sha Wujing in the Flowing Sand River for three hundred rounds.

SpaceX can break down rockets into modules like aerodynamics, engines, and landing systems. If a test flight explodes, veers off course, or crashes, you just find the corresponding hardware defect and fix it. But Grok doesn't work that way—you can't fix its weak logical reasoning, frequent hallucinations, or poor programming by replacing a module.

Data annotation, pre-training, and alignment fine-tuning form a multi-layered, entangled, and fuzzy coupled system. Model defects arise from the superposition of thousands of variables in data distribution, parameter architecture, and alignment strategies. Trying to decompose it into fundamental units and correct them one by one? That's humorous, Bro.

Not only are the underlying logics completely conflicting, but the organizational cultures are also incompatible. Musk imposed SpaceX's militaristic management style on xAI, causing scientists to leave for Anthropic, OpenAI, and Google.

When LeCun handed down xAI's death sentence, a key reason was that xAI now struggles to attract top AI talent because Musk "had not treated the original team members properly."

Media reports depict Musk as a notorious workaholic who has publicly praised intense work multiple times. When busy, he works 120 hours a week, sleeps 3-4 hours a day, and works seven days a week. The extreme work culture of 80-120 hours per week he imposed at X directly drove many researchers to breakdown.

▲The big boss sleeps on the floor.

A former xAI executive revealed that team members almost always stayed up late, with many living at the company for weeks without seeing their families.

This is clearly misaligned with AI's R&D pace—AI large model development requires a relaxed, long-term, and KPI-free research atmosphere.

(Ps: Does Musk also believe in "no strategy is the best strategy"? Such fanatics make great idols but terrible leaders...)

03 /

Musk can't excel in AI, at least not in the sense of cutting-edge foundational models.

But if you think Musk is "doomed to fail," you're being naive.

After disbanding xAI and merging it into SpaceX, many see the "disbanding of xAI" as a symbol of failure but overlook the "integration of xAI into SpaceX" as a dimensional upgrade.

SpaceX now boasts a market value exceeding $2 trillion. Its ability to tear a massive gap in the tech power landscape dominated by the "Magnificent 7" AI giants and become a "gigantic valuation black hole" in the U.S. stock market lies in its scarcity and uniqueness—there are many AI companies in the U.S. stock market, but only one comprehensive tech giant with "rockets + satellites + AI."

Simply viewing SpaceX as a rocket launch company cannot justify its $2 trillion market value.

But if you see it as a combination of space economy infrastructure + global communication network infrastructure + AI infrastructure, the perspective naturally broadens.

Interestingly, SpaceX's prospectus estimates that the company's total future addressable market could reach $28.5 trillion, with aerospace launches and Starlink communications accounting for only about $2 trillion, less than 7%. The remaining 93% comes from AI businesses.

▲SpaceX has drawn a big pie.

Keith Snyder, a senior Wall Street analyst, calculates that to support SpaceX's $2 trillion total market value, AI must have extremely high long-term growth expectations, with AI's long-term option premium accounting for 71% of the overall valuation.

By disbanding xAI, Musk is not abandoning AI; he's just abandoning "self-developed general-purpose large models" and positioning AI as a supporting auxiliary system for SpaceX's space hardware infrastructure. When AI can leverage SpaceX's hardware infrastructure advantages, its role becomes significant.

Its role is mainly reflected in two points: 1. Leasing space computing power and satellite data to global AI firms to generate stable cash flow; 2. Relying on self-developed vertical AI to optimize rocket launches and satellite operation efficiency.

Musk isn't giving up when he can't win; he's just switching to a higher-dimensional battlefield.

The AI industry cannot develop without several key production factors: computing hardware, clean energy, global data, and cross-regional transmission networks... Seeing this, Musk might chuckle: I have them all.

First is space computing capacity. Ground data centers are constrained by electricity, cooling, and land costs. Global power production growth cannot keep up with the expansion of large model computing power. Space is different—solar radiation is unaffected by atmospheric attenuation, with annual equivalent power generation hours 5-8 times that of the ground.

Relying on Starship's reusable large-scale launch capacity, SpaceX plans to mass-launch AI computing satellites and build orbital data center clusters. By 2030, its space computing deployment will reach the hundred-gigawatt level, gradually approaching the total computing power of global ground data centers.

Second is a global, full-domain data collection network. Traditional AI models can only train on internet text, lacking real-time spatial data.

But Starlink has tens of thousands of satellites in orbit, covering the globe without dead zones, collecting real-time surface imagery, environmental sensing, and global communication data streams, forming a full-domain spatial coverage capability unmatched by ground-based internet.

It matters less that Musk can't create the strongest AI. What matters is that in the future, companies that want to create the strongest AI will need to purchase launch, computing power, data, and transmission services from SpaceX—by then, Musk can fully embrace the title "Jiren Xu, aka the Iron Man."

▲Musk makes a bold statement.

Following the spin-off, xAI has shifted its focus; it no longer operates as an independent entity in the general AI market. Instead, it has evolved into a specialized AI division dedicated to supporting Starship simulations, satellite scheduling, orbital computing power management, and Starlink data analysis.

Ultimately, leveraging AI capabilities effectively is crucial for establishing a positive feedback loop: "renting out computing power and data externally while simultaneously enhancing internal intelligence and efficiency."

This is likely the foundation of Musk's bold declaration in March of this year: "In the coming years, SpaceX's accomplishments will far exceed those of all artificial intelligence companies combined."

04 /

When we assert that Musk has surpassed AI, we mean he has shifted the competitive landscape from the "software development layer" to the "infrastructure layer," marking a civilizational leap.

In the 1960s, astronomer Nikolai Kardashev introduced a scale for measuring cosmic civilization levels based on "total energy consumption"—known as the Kardashev Scale. Type I civilizations harness planetary-scale energy, Type II stellar-scale, and Type III galactic-scale.

Within this framework, the goal of AI companies—achieving Artificial General Intelligence (AGI)—essentially aims to maximize information utilization efficiency for a Type I civilization. In contrast, SpaceX's endeavors propel civilization toward the threshold of Type II.

This is not to diminish the importance of AI; on the contrary, AGI could be humanity's final invention or the culmination of all inventions. However, Musk's vision extends beyond this, encompassing a broader coordinate system.

Historian Jill Lepore once coined the term "Muskism," equating it with Neo-Capitalism and Interstellar Capitalism. In her view, visionaries like Musk draw inspiration from science fiction, believing that technology and engineering can resolve all political, social, and economic challenges.

Recurrent themes in science fiction include Earth's destruction, machine domination, interstellar wars, apocalyptic thinking, heroic redemption, and the creation of new civilizational systems.

Indeed, Musk's approach of "posing problems and solving them" operates within the horizon of galactic civilization. He perceives humanity as trapped within a "simulated matrix game of a higher-dimensional civilization" and views humans as biological bootloaders for silico-based life, all from the vantage point of galactic civilization.

Currently, there exists a fundamental difference in hierarchical goals between SpaceX and AI enterprises: the ultimate aim of general AI often revolves around reshaping our modes of production and life; SpaceX's core mission, however, is to construct interstellar survival infrastructure and expand the living space for human civilization.

While AI is formidable, it has its limits: it primarily relies on existing data for induction, reasoning, and simulation, unable to transcend physical laws or create entirely new living environments. SpaceX, in contrast, focuses on accomplishing civilizational-scale projects that AI cannot achieve.

Musk's ambition is to create a "multi-planetary civilization" system that enables humanity to become an interstellar species. Within this system, AI serves merely as a component, akin to an engine in a car or solar panels at a Martian base.

▲Image sourced from the SpaceX official website.

Thus, while both SpaceX and AI companies are sketching future scenarios, SpaceX's vision is clearly grander. If Anthropic and OpenAI represent Plus or Pro versions, then SpaceX is the Ultra Max version.

It is worth noting that the revaluation of tech company market capitalizations driven by this AI wave can no longer be confined to classic valuation frameworks such as price-to-earnings (P/E) ratios, discounted cash flow (DCF) analysis, or margins of safety. Their pricing logic has diverged from financial fundamentals and is now more narrative-driven.

SpaceX's "science fiction narrative" operates at a higher dimension than that of pure AI large model companies. Given this, it's not hard to understand why its valuation anchor continues to be pushed higher by long-term visions such as "space AI = low-latency Starlink network × distributed orbital computing power," and why Musk's personal net worth roughly equals the combined wealth of the world's second to fifth richest individuals.

Thus, on the path of "monetizing science fiction narratives," Musk is once again far ahead.

While xAI has indeed been outmatched by Anthropic and OpenAI in the AI realm, even they would have to acknowledge "respect" when facing SpaceX—after all, SpaceX has already transcended the atmosphere.

It has taken to the skies; you can't help but acknowledge its lead.