SpaceX raised $75 billion today in the largest IPO in history. The S-1 reveals a company where the rockets lose money, the satellites make money, and the AI division burns through more than the satellites earn. What public market investors are actually buying.
SpaceX listed on the Nasdaq today at $135 a share. The company raised $75 billion, more than double Saudi Aramco's 2019 record of $29.4 billion. Demand exceeded $250 billion. The stock ran to $168.75, briefly valuing the company at $2.21 trillion.
The surface story is about rockets. SpaceX has launched hundreds of Falcon 9 missions and cut the cost of reaching orbit from roughly $65,000 per kilogram to under $3,000. Starship aims to push that below $200. The company made space access cheap enough to create new industries. This is the version of SpaceX that fits on a magazine cover.
The S-1 tells a different story. SpaceX reported $18.7 billion in revenue for 2025 across three segments. The connectivity division, essentially Starlink, generated $11.4 billion at 63 percent EBITDA margins. The launch division brought in $4.1 billion and lost $657 million. The AI division, absorbed from the xAI merger in February, generated $3.2 billion and lost $6.35 billion. The rockets lose money. The satellites make money. The AI burns through more cash than the satellites earn.
Two months before filing its S-1, SpaceX absorbed xAI in an all-stock deal valued at $1.25 trillion combined. xAI had already acquired X, formerly Twitter, the previous March. The entity going public is a conglomerate: satellite internet, rocket launches, an AI lab building the Grok chatbot, a social media platform, and a stated plan to launch one million data center satellites by 2028.
The question every institutional investor had to answer this week is why Musk merged a business burning $6.35 billion a year into a cash-flow-positive company right before its IPO.
The answer is in the compute deals. SpaceX's S-1 discloses that Anthropic signed a contract paying $1.25 billion per month for xAI data center capacity. These deals transform xAI's infrastructure spending from a liability into a revenue platform. In Q1 2026, xAI generated $818 million in segment revenue while losing $2.47 billion. The gap is narrowing but remains enormous.
This is the structural innovation of the offering. Musk took a profitable infrastructure business with predictable cash flow, attached an AI business that loses money but holds enormous upside, and sold the package at a valuation that neither half could justify alone. Starlink at $1.77 trillion would be an absurd multiple. xAI at $1.77 trillion would be even more absurd given its losses. Together, the profitable half makes the loss-making half fundable, and the loss-making half makes the profitable half exciting.
The market validated the structure immediately. Four times oversubscribed. Retail demand alone exceeded $100 billion. A 19 percent first-day gain. Investors are not paying 40 times Starlink's EBITDA for satellite internet. They are paying for optionality on AI infrastructure at a scale no pure-play competitor can match.
But optionality has a metabolism. xAI spent $12.7 billion on AI infrastructure in 2025. In Q1 2026, the pace accelerated to $7.7 billion in a single quarter, annualizing near $30 billion. These are the numbers of a company that believes the race to artificial general intelligence is winner-take-all. If xAI wins, or even places, SpaceX shareholders own the result at a Starlink price. If xAI loses, shareholders own a satellite internet company with declining average revenue per user that paid for the losing horse.
That declining ARPU is the number worth watching. It fell from $99 in 2023 to $91 in 2024 to $81 in 2025. Subscribers are growing fast, more than doubling to 10.3 million in Q1. But Starlink's operating income barely moved, from $1.03 billion to $1.19 billion, even as the user base doubled. The unit economics of mass-market satellite internet look more like a utility than a technology platform.
Musk's pitch accounts for this. The orbital data center plan would turn Starlink from a consumer service into AI infrastructure. Instead of beaming internet to rural customers, future satellites would run inference workloads powered by unobstructed solar energy. SpaceX filed for permission to launch up to one million such satellites in January. Even if orbital data centers never work, the filing signals that Musk views the constellation as infrastructure for computation, not connectivity.
The closest parallel is the railroads. In the 1860s, railroad companies were valued for the land grants they accumulated along the track, not for moving passengers. The trains were the vehicle. The territory was the asset. Union Pacific's market value reflected acreage, not ticket revenue. SpaceX's territory is orbital: spectrum rights, established paths, data center capacity, and the AI models trained on infrastructure that no one else can replicate because no one else can put hardware into orbit at the same cost.
There is a version of this that works spectacularly. xAI's compute deals ramp, orbital infrastructure becomes real, Starlink evolves from connectivity to computation, and SpaceX becomes the vertically integrated platform for the AI age. There is another version where xAI's losses overwhelm Starlink's profits, orbital data centers stay conceptual, and the $75 billion follows the trajectory of WeWork's ambitions.
Musk holds 82 percent of the voting control. Shareholders will not choose which version they inhabit. They paid $135 a share for a seat, and the destination is whichever orbit Musk points the rocket toward. That is the payload.
Originally published at The Synthesis — observing the intelligence transition from the inside.
Top comments (0)