June 6, 2026… written the weekend before pricing. As this piece was finishing, the market ran a live stress test on its central claim. On Friday the Nasdaq fell 4.18%, its worst session since April 20251, led by a violent unwind in chip and AI names after Broadcom declined to raise its AI outlook, while a blowout jobs report pushed traders to price in a Fed rate hike2. Two of the assumptions holding up the $1.75 trillion price, that AI capital earns its return, and that capital stays cheap, wobbled in a single session, four trading days before the deal prices on June 11. The risk this article is about did not wait for the IPO. It began rolling in early, in public, on schedule.
On June 12, SpaceX becomes a public company. It will price 555.6 million shares at a fixed $135, raise about $75 billion, and land at a $1.75 trillion valuation, the largest IPO in history3, eclipsing Saudi Aramco’s 2019 record several times over.
That is the headline. The headline is not the question.
The question is what you are actually paying for. One fundamentals-only lens, Morningstar’s, values the proven launch-and-satellite core at roughly $780 billion. Treat that as illustrative, not gospel; it deliberately excludes the moonshots, which is exactly the point. The market is asking $1.75 trillion. The roughly trillion-dollar wedge between the two is not a business that exists today.
It is optionality — Starship, defense logistics, robotics, orbital compute, frontier AI — priced as if it had already been earned. And optionality, unlike a satellite already in orbit, does not show up on a balance sheet until someone proves it out.
What the Fundamentals Actually Say
Start with what is demonstrated, straight from the S-14.
In 2025, SpaceX did $18.7 billion in total revenue. Starlink and the broader connectivity segment carried $11.4 billion of it… a genuinely excellent, fast-growing, cash-generative business. Against that, the consolidated company posted a net loss of about $4.94 billion.
How does a launch-and-connectivity franchise lose five billion dollars? It doesn’t. The AI division does. Formerly xAI — Grok, X, the data centers — it was folded into SpaceX earlier this year. In 2025 it produced $3.2 billion of revenue, lost $6.36 billion at the operating line, and absorbed $12.73 billion of capital expenditure.
The merger converted a profitable rocket-and-internet company into a heavy loss-maker overnight. By design.
Strip xAI back out and you’re still asking the rocket-and-Starlink business to re-rate from ~$800 billion to ~$1.5 trillion… on fundamentals that lost five billion dollars last year.
Back the AI unit out at its own $250 billion acquisition mark and the core is implied at roughly $1.5 trillion — about double what a fundamentals-only lens produces for it. You don’t have to accept any single analyst’s number to see the point: before a word about the future, the market is underwriting execution it has not yet seen.
The Money Is Already Spent
Here is the part most coverage skips. This is an all-primary deal — the cash goes to the company, not to selling shareholders, and insiders including Musk are locked up for 366 days. That sounds investor-friendly. Read it again.
According to reporting on the filing, roughly 78% of the proceeds, about $62.8 billion, is already spoken for, committed to obligations tied to the xAI merger and its backers plus vendor bills. That leaves under $18 billion of genuinely discretionary capital.
Now hold that against the burn: the AI unit consumed over $20 billion in capex across 2025 and the first quarter of 20265 alone. The largest capital raise in history covers a fraction of a single forward year of AI spend.
“No insider is selling” is true. It is not the same as “the money funds the future.” Most of it backfills what has already been spent and promised.
The Number Beneath the Number
To justify $1.75 trillion, someone has to draw the growth curve. That someone is Goldman Sachs — which happens to be a lead bookrunner on the deal.
Per the Financial Times, Goldman is telling prospective investors the AI division grows from $3.2 billion in 2025 to $322 billion by 2030, a hundredfold increase, with total company revenue reaching $474 billion. A hundredfold in five years is roughly a 150% compound annual growth rate, sustained, off an already multi-billion-dollar base. Almost nothing at that scale has ever done it. At $322 billion, SpaceX’s AI unit alone would be larger than Amazon Web Services is today.
For a division that lost $6.36 billion last year and whose chatbot, Grok, runs a distant fourth or fifth behind ChatGPT, Gemini, and Claude in actual usage.
The precision is the tell. False precision is how a motivated number gets dressed up as a forecast.
This is not a forecast. It is an anchor. When the fundamentals support $780 billion and you need to defend $1.75 trillion, you do not project forward to reality… you solve backward for the story that closes the gap.
The “$322 billion” exists because the valuation needs it to.
It Isn’t Access. It’s Distribution.
The genuinely unusual feature of this IPO is who gets to buy. Most large offerings reserve 5–10% for retail. SpaceX is reserving as much as 30%6, and Fidelity has dropped its eligibility threshold to a $2,000 account7 from as high as half a million. The framing is democratization. The structure is something else.
The strategy, per Yahoo! Finance8, is a bet that investors loyal to Musk are less likely to dump shares after the debut, retail as built-in price support. Offering more of a company to the public at a valuation that nearly doubled off the last private mark is not a gift to small investors.
It isn’t access. It’s distribution.
And days before the debut, the institutional backstop declined to show up. S&P Dow Jones Indices refused to relax its rules for megacap IPOs9, keeping its twelve-month seasoning, profitability, and 10%-float requirements intact — and SpaceX, unprofitable and barely 4% floated, meets none of them. The committee’s reasoning could have been lifted from this article: exempting a company simply for being large, long-private, and still unprofitable made little sense.
The S&P 500 is the giant forced-buyer, more than $20 trillion tracks it, against roughly $1.4 trillion for the Nasdaq-100, and it won’t be a buyer until mid-2027 at the earliest, if a dual-class structure doesn’t bar it for good. Strip out the passive bid and the question sharpens: who holds the price up?
The answer the deal is counting on is the same loyal retail it reserved 30% for.
And the demand headline is softer than it reads. The book is reportedly twice covered10, roughly $150 billion of orders against $75 billion of stock, which sounds bullish until you sit with it. Two times is modest for a deal this hyped; genuinely hot IPOs run five to twenty times, and institutions routinely pad orders severalfold expecting to be cut back, so a merely-twice-covered book may flatter much thinner real conviction.
More important, the price is fixed.
At a set $135, oversubscription can only tell you the deal clears at the number SpaceX chose… there is no bidding the valuation up, only agreeing to it, and it says nothing about whether intrinsic value sits above or below $1.75 trillion. And the demand justifying the premium over the proven core is drawn by the same orbital-AI vision this piece questions.
A book twice covered proves the marketing is working. It doesn’t prove the price is right.
And the cruelest part is the math underneath it. Yes, there is a record $8.28 trillion sitting in money-market funds11. But that is flight-to-safety money… parked because investors are nervous about the Fed and the world, the opposite posture of someone buying the most expensive IPO ever. It skews old and institutional; only about $3.1 trillion is retail12.
Meanwhile the young, Musk-following retail crowd the marketing leans on can least afford to participate in size. Forty-seven percent of Gen Z and millennials live paycheck to paycheck13, with cost of living their top concern for the fifth straight year. The cohort invited to the party is structurally the one that can only afford to watch, while the gains accrue to whoever already holds the assets.
The Premium Runs on Borrowed Power
Every dollar of the AI premium assumes the compute buildout continues at pace. That buildout runs on electricity. And the electricity is the problem.
In April, the NAACP, represented by the Southern Environmental Law Center and Earthjustice, sued xAI14 under the Clean Air Act over a fleet of unpermitted methane gas turbines in Southaven, Mississippi, powering the Colossus 2 data center across the state line in Memphis. The notice cited 27 turbines. After being warned, the company added six more, triggering a May emergency motion to shut them down15 at 33. State regulators later confirmed the real number had reached 46.
Two facts make this structural rather than cosmetic.
- First, the loophole is closing: the EPA ruled in January 202616 that the trailer-mounted “mobile” turbines are in fact stationary sources requiring full federal permits.
- Second, there is precedent… at the first Colossus site, legal pressure already forced xAI to remove unpermitted turbines and permit only fifteen17.
The turbines are not a line item. They are a time arbitrage on the grid. Utility interconnection for hundreds of megawatts takes years; trailer-mounted gas turbines bring compute online now. That speed is exactly what Goldman’s 150% growth curve capitalizes. Force full permitting, emissions controls, and environmental review, and you do not kill the buildout, you slow it and make it more expensive, which pushes the revenue inflection to the right and compresses its present value twice over.
This is a fundamentals discount, not a litigation footnote.
They are telling you what matters most by what they are willing to break the law to keep doing.
Stacking from 27 turbines to 46 during active litigation is revealed preference. It tells you fast power is the single most binding constraint on the entire plan. And the more load-bearing the shortcut, the more a credible threat to it should discount the story built on top.
The Grid Is the Real Constraint
The bull narrative has always been about chips: buy more GPUs, train bigger models. The real bottleneck is power. A frontier data center now needs 500 to 1,000-plus megawatts, a small city’s worth of round-the-clock electricity, and the grid cannot deliver it on demand. PJM, the largest U.S. grid operator, reports that data-center projects entering service in 2025 took an average of more than seven years to reach operation18: roughly three years in the interconnection queue, then four more waiting on transmission lines and substations. The GPUs are the easy part. The gigawatt is the hard part.
That single fact reframes the turbines. xAI didn’t build an unpermitted power plant because it likes risk. It did it because the legal, permitted path to power is measured in years, and the model-training race is measured in months.
The turbines aren’t a sign of abundance. They’re a symptom of a grid that can’t keep up.
And the spending never stops. Staying at the frontier means multiple new data centers, repeated GPU refresh cycles, continuous power and networking investment… plausibly tens of billions of dollars a year, indefinitely, against a segment that doesn’t yet generate comparable free cash flow. That makes the whole enterprise structurally dependent on what comes next: new equity, fresh debt, strategic partners, and capital markets that stay friendly.
Remove any one and the buildout stalls.
This is the setup for a classic capital-cycle trap. You commit billions based on today’s scarcity. By the time the capacity comes online, years later, the scarcity that justified it may be gone. The fiber-optic boom of 1999–2001 is the template: enormous capacity laid against extrapolated demand, then a decade of “dark fiber” and a wave of bankruptcies. Demand did eventually catch up.
But only the builders who hadn’t over-levered were still standing to enjoy it.
Then Build It in Space
Raise the grid constraint and the standard reply arrives on cue: then build the data centers in orbit, where the sun never sets. It is not science fiction anymore. Starcloud has flown an NVIDIA H100 in orbit, Google’s Project Suncatcher launches prototype TPU satellites in 202719, and SpaceX has filed with the FCC for up to a million orbital-compute satellites20. Of every player in the race, SpaceX is the best positioned — it owns the launch, the optical mesh, and the satellite line.
But the power advantage just trades one constraint for a harder one. In vacuum there is no air and no water to cool with; heat can only be radiated away, and Starcloud’s own gigawatt-scale design needs cooling panels measured in kilometers21. Add radiation degradation, the launch mass of tens of thousands of tons of servers, and the fact that you cannot hot-swap an obsolete GPU in orbit, and gigawatt-scale space compute becomes a 2035-and-beyond moonshot, Google’s word, not mine.
Which makes it the perfect emblem of the whole problem. Orbital compute is real optionality, and SpaceX may genuinely win it. But “best positioned to maybe win a race that may not be winnable on the relevant timeline” is a lottery ticket, not a line item.
Orbital compute is a lottery ticket, not a line item, and the price is paying for a winning ticket it doesn’t hold yet.
Nothing about a 2030 revenue curve can rest on it. When the dream is offered as a reason the price makes sense, it is borrowing from 2035 to pay for today.
The Risk the Price Hasn’t Rolled In
Here is the deeper problem, and the one the valuation quietly assumes away. The risk isn’t that the data centers sit idle. It’s that their output gets commoditized.
If open-weight and efficiency-optimized models, many of them coming out of Chinese labs, reach 95 to 98 percent of frontier capability at a fraction of the compute cost, the price of intelligence collapses even as the volume of it explodes. The clusters stay busy. They just sell a commodity at commodity margins.
The bull’s rebuttal is real and worth stating fairly: cheaper intelligence tends to create more total demand, not less. That’s Jevons’ paradox, and it’s probably right. But filling the building is not the same as earning a premium on what comes out of it.
Jevons fills the building. It doesn’t guarantee pricing power over what comes out of it.
So the spend is justified only on two conditions: that xAI stays at the moving frontier, and that the frontier keeps commanding a premium over the cheap-and-good tier. Both are contestable. “Ninety-five percent of capability” quietly assumes the gap stays closed… when at each new capability tier, it tends to reopen.
In valuation terms, none of this shows up in next year’s revenue. It shows up in the terminal multiple. The entire premium rests on the market granting the AI segment a monopoly-style, software-style multiple. Commoditization is the thesis that it’s actually a capital-intensive compute utility… and utilities don’t trade at software multiples. That re-rate, from narrative multiple to commodity multiple on the segment carrying the whole premium, is the single largest hidden assumption in the price.
And fiber wasn’t the first time the world confused a technological triumph with a financial one. Railroads remade a continent and bankrupted their financiers. Airlines made the planet smaller and handed shareholders decades of dismal returns. The pattern has almost nothing to do with whether the technology works: transformative infrastructure routinely succeeds for society while destroying the equity that funded it, because competition and overbuild compete the returns away faster than the assets can earn them. This is the risk that outranks all the build risk above it, you can win every permit, energize every megawatt, and still not clear your cost of capital.
Railroads remade a continent and bankrupted their financiers. The world keeps the surplus. Shareholders rarely do.
And it hasn’t been rolled in. The $1.75 trillion is anchored to projections that presume the opposite; explosive demand, durable pricing power, permissive everything. The price doesn’t need to reflect commoditization to clear the IPO. It needs to reflect it to be right. The distance between those two is not a rounding error. It’s the trade.
What You’re Actually Buying
None of this means SpaceX is a bad company. The opposite. Starlink is one of the best infrastructure businesses of the decade, and the launch franchise is close to a monopoly on getting mass to orbit. At $800 billion to $1 trillion, this is a compelling asset with real upside.
At $1.75 trillion, you are no longer buying the asset. You are buying three separate bets, priced as if they were one.
- That xAI can build it — against a grid that quotes seven-year waits and a capital appetite that has to be re-fed every year.
- That it can power it — through courts now being asked to switch off the turbines the timeline depends on.
- That it can earn on it — a non-commodity return, in a market racing to make frontier capability cheap.
Each is a real probability, and each sits well short of certainty. The mind wants to grade them one at a time — build 80%, power 80%, economics 80% — and conclude “pretty likely.”
But a valuation doesn’t get to hold them separately; it pays for all three at once.
Eighty percent across the board compounds to barely better than a coin flip: 0.8 × 0.8 × 0.8 ≈ 51%. Drop each to a still-respectable 70% and the joint odds fall to about 34%. (The three aren’t perfectly independent — a capital boom lifts several at once — but correlation cuts both ways, and the market is pricing the conjunction at something close to 100%.)
The valuation treats the product of three uncertain bets as a single sure thing.
Here is why the market can pop the stock anyway: none of these resolve on day one. They resolve in steps — a permit ruling, a first margin print, a competitor benchmark that does 97 percent of the job for a tenth of the cost. Each step is a moment the price is forced to roll in what it skipped. A thin float and a loyal retail base can carry the open. They cannot carry the disclosures.
At $800 billion, it’s an infrastructure asset. At $1.75 trillion, it’s a wager that three things land on schedule… and the first one is already in court.
The disciplined move is the boring one. Sit on your hands. Let the story meet its first quarter of public scrutiny. The gap between what the price assumes and what it has been forced to prove will not close on the debut…
and that gap, not the pop, is where the decision actually lives.
Not investment advice. Figures drawn from public filings and reporting as of June 2026; the IPO had not yet priced at the time of writing.