Wall Street thinks it bought rockets and broadband. The more interesting bet is whether SpaceX now controls more of AI's physical layer than almost anyone else.
The more revealing number was not on the launchpad. It was in the prospectus.
Ahead of this week’s record IPO, SpaceX disclosed that Google has agreed to pay it roughly $920 million a month for compute capacity at its Memphis campus, access to something on the order of 110,000 GPUs running into 2029. A company the market still files under "space" just signed one of the largest publicly disclosed AI compute contracts in the world. That doesn’t prove SpaceX's more exotic ambitions will work, but it proves something more immediate: investors are no longer pricing SpaceX as just an aerospace company. The rockets got the headlines. The receipt told the story.
The Market Saw Rockets. It Missed The Stack. #
The familiar version is easy: the largest IPO in history, shares priced at $135, a first-day close near $161, a valuation around $1.75 trillion, Starlink carrying the revenue story and Musk mythology doing the rest. All true, and all incomplete.
The sharper read is that public investors just bought into a company that controls more of AI’s physical layer than almost anyone else. Not every layer. SpaceX doesn’t own the chip fabs or the power grid, and pretending otherwise is where this argument gets sloppy. But it controls an unusual amount of the machinery sitting between intelligence and the real world.
That matters now because the AI race has changed shape. For three years it was mostly a software contest of models, weights and benchmarks. That contest has not ended, but model differentiation is narrowing while infrastructure scarcity is widening. The binding constraint is no longer intelligence. It’s power, land, cooling, chips, bandwidth and time.
AI is also leaving the browser, moving into vehicles, satellites, drones, factories and defense networks, where intelligence has to sense, communicate and act in the physical world. That kind of AI doesn’t run on a clever model alone. It runs on hardware, networks, energy and the capacity to deploy all of it at scale.
In the last AI cycle, everyone wanted the best model. In the next one, they may need the best landlord.
What SpaceX Actually Controls #
SpaceX’s rocket business gives it deployment capacity few competitors can compete with. Starlink is a communications network already in orbit, with some of its most strategically important customers in defense, maritime and aviation, not only on living-room couches. xAI, which SpaceX acquired in an all-stock deal in February, brings the intelligence layer in-house. The Memphis data centers, and that Google contract, are terrestrial, real and generating revenue today. The orbital compute ambition, with test units targeted for 2027, remains just that: an ambition.
The power is not in any single piece. It’s in the bundle. Having sat through enough industrial-AI procurement reviews to know how these decisions get made, I can tell you that is the part buyers underestimate. You rent one capability because it’s the best option on the table, and within a budget cycle or two you are designing around the whole platform, because unwinding it costs more than staying. In consumer markets, owning that infrastructure is an advantage. In defense markets, it starts to look like sovereignty.
The Risk Is Not Small #
None of this means the valuation is sane. Orbital data centers may turn out to be physics fan fiction; cooling, radiation and launch economics are not footnotes, they are the problem. The governance is uncomfortable too: Musk's role on both sides of the xAI deal, dual-class voting that leaves public shareholders with exposure but little control, and one man now load-bearing across several enormous companies. China's Spacesail is racing to build a rival stack. And even the marquee compute deal carries an asterisk: Google itself calls it "bridge capacity," a short-term arrangement rather than a permanent moat.
A $1.75 trillion valuation prices in a great deal of optionality. The problem with selling optionality is that eventually the market asks which option is becoming revenue.
The Lesson Is Bigger Than Elon #
Strip away the personality and the takeaway outlasts the news cycle. If SpaceX is right about where value is migrating, the same question lands on every AI player: the model labs, the hyperscalers, the telecoms, the robotics firms and the defense contractors. The warning is blunt: you can own the model and still rent the business.
Wall Street thinks it bought into the biggest space company in history. What it actually bought is a bet on who owns the ground the next technology cycle will run on. The rockets were never the point.