The most simplified breakdown of the SpaceX IPO on the internet

Why does it matter? Because 'SpaceX' is actually three businesses stapled together — and only one makes money

SpaceX is going public at $1.75 trillion — 100 times revenue, the largest IPO in history. Shaan spent a few hours trying to figure out whether to buy it, building a Claude-generated deck run through a Charlie Munger filter. Sam combed the S1 for the stuff nobody's talking about. What came out: the business generating actual cash is a four-year-old satellite internet service, the bull case has almost nothing to do with rockets, and the single biggest risk factor is that one specific person dies.

• Starlink: $11B revenue, 40% EBITDA margins, 10M subscribers, no real competition • The real bull case is SpaceX becoming the Saudi Arabia of AI compute, from orbit • Land permitting is so broken that orbital data centers are genuinely the easier path • At $1.75T, this is less a stock pick than a structured bet on one person's longevity

Starlink is the part that makes money — XAI burns it, and the rockets are the railroad in between

Four years old. That's the number Shaan keeps landing on — $11 billion in revenue, 40% EBITDA margins, 10 million paying subscribers, no real competitors, recurring revenue, a product half the planet needs. All of this is buried inside a rocket company like found money.

The moat is structural: SpaceX launches its own satellites on its own rockets at 40% of total launch volume. Nobody can replicate that flywheel. Direct-to-Cell extends the play further — Starlink signal straight to your phone via T-Mobile, no dish required. Either it launches as its own carrier or it becomes a $5–$10/month add-on to every cell plan on Earth with guaranteed coverage everywhere. The telecom market is $2 trillion and completely undifferentiated — Patrick Mahomes decides your carrier, not the product. Starlink is the first one with an actual difference.

Claude's Munger summary nailed the structure: "Starlink is the part that makes money. XAI is the part that burns it. The rockets are the railroads in between."

SpaceX's real bull case is becoming the Saudi Arabia of AI compute — from orbit

Saudi Arabia made trillions owning oil. The world ran on energy; they were the largest, lowest-cost producer. Shaan's frame for the next 20 years: the world runs on compute. Every business, robot, car, and appliance will have a compute need. The question is who's the Saudi Arabia of compute.

Space-based data centers run on free solar power. No cooling costs — space is already "basically freezing cold," with natural heat dissipation from chip radiation. Ground infrastructure burns money on power, cooling, and real estate simultaneously. Elon's projected advantage: 50% to 200% lower cost per AI token than anything built on land. If that number is right even by half, SpaceX becomes the entity everything else routes through. He's currently the only one with the launch capacity to build it at scale.

The physics gets contested. But betting against Elon on "this will never work technically" has been a losing trade every single time.

It's easier to build a data center in orbit than to get Alameda County to approve one in your backyard

Not hyperbole — Shaan quotes the SpaceX investor deck directly: it is literally easier to design and launch the heaviest rocket ever built than to navigate local permitting for a data center on American soil.

The bottleneck for AI infrastructure isn't capital, engineering, or equipment. It's zoning boards. Anti-data center community opposition has reached the point where fully resourced operators can't break ground. The money exists. The talent is available. Equipment is sitting reserved and idle. The roadblock is a public comment period.

So the response is to go somewhere with no county supervisors, no hearings, no permits required: orbit. This isn't idealism driving the thesis — it's regulatory arbitrage at civilizational scale. The AI infrastructure buildout is going off-planet not from futurism, but because domestic permitting became the binding constraint.

Grok couldn't attract enough users to fill the world's largest GPU cluster, so Elon rented it to Anthropic and Google for $2B+ a month

Grok has 100 million users. ChatGPT has a billion. Elon built Colossus — the largest GPU cluster on Earth, bigger than Google, bigger than Meta — and then discovered he didn't have the demand to fill it. Shaan's frame: he bought a huge mansion with no friends to come over.

So he turned it into Airbnb. Anthropic — directly competing with Grok — signed for over a billion dollars a month. Google — previously the world's best data center builder — signed for another billion. Two deals, combined $20B+, both announced in the last two months. Google is paying Elon for compute access.

The contracts have 90-day cancellation clauses. Nobody knows if this is permanent. But the principle holds regardless: when a product fails, audit the infrastructure you built on the way there. The asset is sometimes worth more than the product would have been.

At $1.75 trillion and 100x revenue, the biggest risk in the analysis isn't competition or timeline delays — it's that Elon dies

The AI Charlie Munger call was clean: "A wonderful business at a silly price." You could admire it from afar. Didn't need to own it.

Shaan and Sam take the tension seriously. Traditional valuation tools break here — P/E ratios, price-to-sales, both irrelevant. The real frame is the "price-to-Elon ratio": Elon companies trade at roughly 10x what comparable fundamentals would otherwise support, and the market has been right to price it that way.

The problem: at $1.75T, the present value of future cash flows stretches 30+ years. All of it assumes one specific person keeps running the company. Sam's conclusion, stated plainly: "The biggest risk is that he dies." Not Starship delays. Not regulatory risk. One human being's continued existence is the key variable. He's bad at timelines — both agree on that. But he's good at doing it eventually.

Elon retained 42% of a rocket company after 20 years of fundraising — the founder who built 'a folder in the cloud' owns 4%

This number stopped Sam cold. After two decades raising capital for the most expensive business on Earth — hardware that literally explodes when it fails — Elon still owns 42% of SpaceX with 85% voting control.

Shaan's contrast: Aaron Levy from Box went public at 4% founder ownership. Box is cloud storage. "A folder in a cloud." No explosions. 4%.

The standard rationalization for founder dilution is that hardware requires it, that capital-intensive businesses demand repeated rounds of painful dilution. SpaceX is the counterargument that ends the conversation. If you can hold 42% building rockets over 20 years of fundraising, there is no credible excuse for extreme dilution building software.

The highest-returning investment strategy of the last 15 years required zero sophistication — one right call, then nothing

The optimal move, it turned out, was to concentrate all the capital into one person's companies and stop there. Luke Nosek worked this out at Founders Fund and eventually left to formalize it as Gigafund. At the time it seemed crazy — and crucially, unsophisticated. Shaan flags the second thing as the more interesting problem.

The 15-year scorecard: buying Google, Facebook, and Amazon in 2010 and sitting on your hands beat nearly every active strategy. Bill Gates would be the world's richest man if he'd simply not sold his Microsoft stock. Chamath's entire post-Facebook career — funds, SPACs, sports teams — all of it underperforms simply holding his Facebook shares from when he was an employee.

"We seek out sophistication. We seek out activity. But simplicity is actually a more powerful thing if you get it right." Gigafund has one of the largest stakes in the SpaceX IPO. The boring move won again.

Elon's pay package vests only when a million people are living on Mars — base salary in the meantime is zero

$750 billion in total grants, two simultaneous conditions: SpaceX must hit $7.5 trillion in market cap — the most valuable company in history — and there must be a permanent, self-sustaining Mars colony of at least 1 million people. Both. Neither optional.

The consolation prize, the AI CEO Award, requires 100 terawatts of compute per year from non-Earth data centers. The entire current US electrical grid produces 1 terawatt. The smaller award requires 100 times the US grid, from orbit, from infrastructure that doesn't yet exist.

Base salary: zero. When you design compensation, ask whether the vesting conditions actually encode the outcome you care about or just a financial proxy that can be gamed. These don't hedge. No Mars colony, no payday — regardless of everything else built along the way.

The real bet isn't on SpaceX stock — it's on whether the entity that builds the AI infrastructure layer gets to charge rent on everything that runs through it

The companies renting Colossus today might be paying the first invoice to the most consequential utility on Earth. Starlink is already the connectivity layer for the places no one else reaches. If space data centers land, SpaceX ends up at the bottom of the stack that everything AI depends on — extracting rent from every inference, every query, every digital worker.

The Saudi Arabia analogy almost undersells it. Saudi Arabia owned the oil. Not the pipelines, not the tankers, not the refineries simultaneously. If this plays out, SpaceX owns all of it.

The rails get built first. Then everyone pays to ride.

---

Topics: SpaceX, IPO, Elon Musk, Starlink, space technology, AI compute, data centers, investing, founder equity, valuation, Grok, Colossus, satellite internet