On Binance, a synthetic contract called SPCX climbed to the exchange's third-largest perpetual futures pair by trading activity, behind only Bitcoin and Ether.
It gave retail speculators up to 5x leverage on a stock most of them couldn't actually buy—a derivative of a proxy of a dream, priced against SpaceX's implied $1.75 trillion valuation.
Then the underlying dream did something the fever hadn't priced in. It fell.
SpaceX debuted on Nasdaq on June 12 with a $75 billion raise, the largest IPO in history, eclipsing Saudi Aramco's 2019 record. Shares jumped nearly 20% on day one, closing at $160.95 and briefly touching a $2.25 trillion intraday valuation. Elon Musk became the world's first trillionaire, his net worth crossing $1.17 trillion.
By June 22, four sessions after peaking at $229.85, the stock had shed 35.73%. More than $600 billion in market value evaporated—over $400 billion of it in a single day.
The trigger was not a launch failure or a lost satellite. It was a press release.
Less than two weeks after collecting $75 billion in fresh equity, and while sitting on roughly $100 billion in cash, SpaceX announced it would borrow another $25 billion from the bond market.
That sequence—equity, then debt, then selloff—is not the story new public shareholders want to carry back to their investment committees. It reads as an admission that the largest tech listing on record did not reset the balance sheet for the next cycle of spending.
To understand why $100 billion in the bank isn't enough, look at the burn.
SpaceX's ratio of capital expenditure to revenue climbed from 44% in 2023 to 112% in 2024 to 207% in 2025. In the first quarter of 2026, the company earned $4.7 billion in revenue against $10.1 billion in capex. It is spending more than two dollars to earn one.
The prospectus lays out where the money goes: $120 billion planned across 2026 and 2027, and $350 billion cumulatively before 2030—most of it earmarked for AI.
Which brings us to the crack under the whole valuation.
The bull case for SpaceX was never really about rockets. It was about a vertically integrated "rocket + satellite + AI + data center" empire, with xAI's Grok as the software crown jewel bolted onto the space business. That was the trillion-dollar dream investors thought they were buying.
But look at what the AI segment is actually becoming.
Over the past weeks, SpaceX signed deals to rent out raw compute. Anthropic will pay $1.25 billion a month for the entirety of the Colossus 1 data center. Alphabet pays $920 million a month for about half as many processors. Reflection AI pays $150 million a month for capacity on Colossus 2.
Those three contracts total nearly $28 billion in annualized revenue. Impressive—until you ask what kind of revenue it is.
It is undifferentiated compute rental. A token is a token. SpaceX has become, in the Motley Fool's phrasing, a "neocloud"—an infrastructure-as-a-service provider selling capacity to labs that want as much of it as they can get.
The tell is what's happening to its own AI. Management expects Grok to remain a niche large language model, and hundreds of thousands of GPUs sit idle in its data centers—so xAI is leasing out the silicon rather than running its own models on it.
This matters because it repriced the business against the wrong benchmark.
If SpaceX is a software company with a frontier model, the comparisons are Palantir and the trillion-dollar AI unicorns. If it is a neocloud renting GPUs, the comparisons are CoreWeave and Oracle.
CoreWeave trades at about 4.2 times sales. Oracle at roughly 5 times. Both have enormous contracted backlogs—CoreWeave near $100 billion, Oracle's remaining performance obligations at a staggering $638 billion.
SpaceX trades at well over 100 times trailing twelve-month sales.
Do the arithmetic on that gap. Even if SpaceX triples its revenue over the next year on the back of these infrastructure contracts, its sales multiple would still sit far above anything reasonable for a business selling commodity compute.
The company's answer is orbital data centers—servers in space, launched cheaply by a fully reusable Starship, tapping the one cost advantage terrestrial neoclouds can't match. Management says the first ones could launch by 2028.
That is the thread worth pulling. The orbital-data-center bet is no longer a speculative flourish; it is the load-bearing justification for a compute business the market is otherwise repricing to commodity multiples. If Starship slips, the entire cost-advantage story collapses, and SpaceX is left as an expensive CoreWeave.
The orbital-data-center bet is no longer a speculative flourish; it is the load-bearing justification for a compute business the market is otherwise repricing to commodity multiples.
Meanwhile, the only segment reliably generating cash is Starlink.
Connectivity was 61% of revenue in 2025 and rose to 69.4% in the first quarter of 2026, with some banks expecting it to approach 80% for the full year. Nearly 9,600 satellites, 164 countries, more than 10 million subscribers.
But even the cash cow leaks. Starlink's average revenue per user has fallen three years running, so growth depends on constantly adding subscribers—which means constantly launching satellites and building ground stations, against accelerating competition from Amazon's Kuiper, OneWeb, and China's GW constellation.
So this is the honest shape of the thing investors bought: one profitable-if-declining-margin business, one launch operation that barely breaks even, and an AI segment that just quietly demoted itself from software moonshot to GPU landlord.
Which is why Scott Galloway's verdict lands so hard.
The NYU Stern professor argued the $1.75 trillion target had "no anchor." Not one investment bank on the deal could name a clean public comparable. Underwriters tried Boeing, AT&T, Palantir, GE Vernova, and Vertiv—none fit.
At $1.75 trillion, SpaceX would trade at roughly 94 times trailing revenue on $18.7 billion in sales. That is richer than any company in the S&P 500. Even Palantir, the index's most expensive stock on price-to-sales, trades near 67 times.
To bridge the gap, bankers told investors SpaceX's total addressable market was $28.5 trillion—roughly the size of the entire U.S. economy, including a projected $22.7 trillion in enterprise application revenue, about 30 times today's enterprise software market.
When the number needs the whole economy to justify it, the number has no anchor.
Galloway read the offering's structure as the confession. A typical IPO reserves 5% to 10% of shares for individuals; SpaceX set aside as much as 30%, funneled through Robinhood, Schwab, Fidelity, SoFi, and E*Trade.
The reason institutions weren't asked to absorb it, in this reading, is that institutions wouldn't.
The numbers around the float confirm it. Only 4.2% of shares outstanding traded on day one—a sliver that magnifies every move in both directions. Retail bought a net $405 million in the first five sessions, the strongest retail debut of any IPO in recent memory.
As JonesTrading's Michael O'Rourke put it once the selling started: "Sellers are back in control. Anyone in the world who wanted to buy this has bought it already."
That is the whole story in one sentence. A valuation with no comparable, held aloft by retail enthusiasm because the professionals passed, ran out of new buyers—and the first honest public referendum returned a skeptical vote.
The stock is still above its $135 IPO price. This isn't a collapse; it's a repricing.
But it is a specific kind of repricing. The market didn't punish the rockets, or the satellites, or even the idea of AI. It punished the switch—the discovery that a $2.25 trillion space-and-software dream had, on closer inspection, become a commodity GPU-rental business bolted to a maturing internet provider, funded by borrowing against cash it supposedly didn't need to spend.
The synthetic perpetual on Binance can keep the dream trading around the clock. The equity, for now, has to answer to gravity.