"The $100 billion ARR in December is not a question mark," Elon Musk told analysts on SpaceX's first-ever earnings call. "That's what we would achieve if we basically did nothing."
Two days later, the stock closed at $108.27 — down more than 50% from its June high and below the $135 price Musk himself had set.
Start with those two acts of price-setting, because the whole story lives in the gap between them.
In June, SpaceX named its own IPO price: $135 a share, no range, no negotiation. The company had run its testing-the-waters meetings, gauged the appetite, and then simply told the market what the number was. Goldman, Morgan Stanley, and three other banks papered the deal. Investors got a take-it-or-leave-it wager.
That is not how a large IPO normally prices. New issuers offer a range so the company and its advisers can read demand at different levels — so the market gets a vote. SpaceX removed the vote before the roadshow even launched.
Then the market ran its own price discovery anyway. The stock hit $150 on debut, ran to $225 within days, and spent the next eight weeks getting marked down to $108. The difference between the two numbers is simply the difference between who was allowed to say no.
The difference between the two numbers is simply the difference between who was allowed to say no.
The trigger for the collapse was the first earnings report, and the number that did it was capital expenditure.
SpaceX spent $18.4 billion on capex in the second quarter. Revenue was $7.8 billion. Capex more than doubled the entire top line, and well over 80% of it went to AI — the $15.8 billion figure buried under a business that, a year earlier, was spending a sixth of that.
Analysts had modeled roughly $13.2 billion. The miss wasn't small, and it wasn't in the direction that flatters a newly public company.
Here is the thing to hold onto: none of this was a surprise about the business. The prospectus had already shown $10.1 billion of Q1 capex, $7.7 billion of it AI. The surprise was watching the ramp accelerate in real time, on a company that had priced itself as if the ramp were finished.
Now walk the "synergies," because each one is really a capex commitment wearing a growth story.
The Nvidia partnership: SpaceX will design the compute payload for its Starmind AI satellite around Nvidia's chips, and Musk declared on the call that the company would use Nvidia exclusively going forward — Vera Rubin is "the best architecture," full stop. A prototype ships early next year, mass production in 2027. That is spending, not revenue.
The Tesla chip plant in Texas: a facility built with Tesla to meet compute demand. Another factory, another multi-year outlay, framed as vertical integration.
The data-center-in-space pitch: orbital AI satellites from 2028, sold as ultra-low-energy compute powered by the sun. Beautiful. Also, by definition, years of building before a dollar comes back.
Musk's own framing gives the game away. Data centers, he said, are "a trivial problem compared to making reusable rockets." He wants 20 gigawatts of capacity — power and cooling — by the end of next year, conceding some projects slip but expecting "something close to 15GW." Every gigawatt is a check.
The bull case, delivered by CFO Bret Johnsen, is that this capex converts fast: "less than a one-year payback" on AI compute. He pointed to $6.7 billion of cloud services revenue contracted in the first weeks of the quarter, ramping from October. Stack it up and you get the $100 billion ARR headline.
But look at what the AI unit actually did in Q2. It generated $2.56 billion in revenue and lost $1.26 billion at the operating line — after a $2.47 billion loss the prior quarter on $818 million of revenue. The compute business is bleeding while the ARR promise is being made about the future.
And the revenue that exists is telling on itself. The deals with Anthropic (up to $1.25 billion a month), Google (up to $920 million a month), and Reflection (up to $150 million a month) indicate SpaceX built far more capacity at its Memphis Colossus site than it can use internally, so it's reselling the overhang. One nod worth making: SpaceX is Anthropic's compute landlord, which means some of that "recurring revenue" traces back to another cash-burning AI company's financing — a dependency, not a moat.
Reselling capacity for short-term revenue is, as the reporting puts it, detached from the broader Mars-and-Grok ambition. It's the near-term story papering over the long-term one.
So the company loses money three ways. The space segment posted a $542 million net loss on $962 million of revenue. AI lost $1.2 billion. Only Starlink — $1.6 billion in the quarter — actually earns. The consolidated net loss was $143 million in the quarter and roughly $2 billion across the first half, against a cumulative deficit that stood near $41.3 billion before the IPO even happened.
This is the risk that the fixed-price IPO transferred. Musk set $135 with no discovery, sold the tradeable float to retail and early holders, and locked in public shareholders as the people who would absorb whatever the capex curve did next. They said yes to a price they weren't allowed to negotiate. Then the capex number arrived and they had no recourse except selling.
And here is where the asymmetry turns from clever to structural.
On the Thursday after earnings, 911 million shares unlocked — about 7% of the company, and more than the 639 million sold in the IPO itself. Another 319 million become eligible August 20. Roughly 700 million in September. Close to that again in October.
Musk's stake? More than 6 billion shares, locked until June 2027. He also controls over 82% of the voting power — call it 85% with the governance protections baked into the prospectus — meaning the person who priced the deal keeps command of the company and keeps his shares frozen while everyone else's float triples into a falling market.
That is the trade laid bare. The people who set the price aren't the ones absorbing the repricing. The float that can flee is the float that didn't get a vote; the block that got the vote can't flee for another year and a half.
Mizuho's caveat is fair — shares becoming eligible to sell doesn't mean the full tranche hits the tape. But the sellers are lining up. Falcons safety Jessie Bates III, who paid $150,000 in 2022 at a $127 billion valuation, told CNBC he plans to sell everything to "lock in gains." Rainmaker's Greg Martin thinks the stock's near-term path is now driven more by the unlock than by any fundamental.
Which is the quiet admission underneath the whole thing: for the next several months, SpaceX is a supply-and-demand event, not a space company.
Musk's answer to all of it is that investors are "underestimating" him, that Starlink might "operate most of the world's internet," that $1 trillion in revenue arrives by 2030 — a year sooner than he said six weeks ago. In 2025, total revenue was under $19 billion.
You can believe every word of the vision and still see the mechanical problem clearly. A company priced itself with no discovery, committed to a capex program more than double its revenue, and handed the downside to the shareholders it wouldn't let negotiate — while the man who named the price sits locked, in control, until 2027.
The 50% drop and the 911-million-share unlock are the same event from two angles: the market clawing back a risk premium it was never allowed to price in the first place.