SpaceX has long been celebrated as the company that would take humanity to Mars, but its first quarterly earnings report as a public company tells a very different story. By revenue, SpaceX is now primarily a telecom company and a provider of rented computing power. The space sector, the part that launches rockets and dreams of interplanetary travel, contributed just over 10 percent of the company’s total revenue and failed to break the billion-dollar mark in the quarter. The numbers make one thing clear: SpaceX is barely space and mostly X.
The telecom arm, known as Starlink, generated $4.2 billion in revenue and was the only segment of the business that did not post an operating loss. Starlink’s satellite internet service has become a reliable cash cow, with millions of subscribers worldwide. During the earnings call, Gwynne Shotwell outlined plans to expand into a phone service that would compete directly with AT&T, Verizon, and T-Mobile. That move could position Starlink as a major player in the broader telecommunications market, not just a niche provider for rural users and travelers.
But the real financial engine, and the source of most of the spending and hype, is what the company graciously calls “AI.” This segment includes the leasing of data center capacity to artificial intelligence companies, a business that has grown so large that it now dwarfs the rocket-launch side of SpaceX. In the second quarter alone, SpaceX spent $15.8 billion on AI-related infrastructure. Analyst Alexander Potter expects that spending on the neocloud business, which leases data center capacity to AI firms, will climb to $65 billion next year, a $17 billion increase over his previous estimates.
A Data Center Built for Grok That Didn’t Work Out
The pivot to AI wasn’t originally the plan. Musk built SpaceX’s Colossus 1 data center in Memphis to train Grok, his in-house AI model that has gained notoriety for referring to itself as MechaHitler and for issues around generating inappropriate images. But xAI, the company behind Grok, struggled to operate the complex. Latency problems made it difficult to train models effectively, and the facility’s mix of newer and older chips created bottlenecks. Eventually, the decision was made to rent out the compute capacity to other companies instead.
On the earnings call, Musk admitted that only 10 percent of the compute SpaceX builds will go to Grok. That is a striking admission for a company that spent billions on infrastructure ostensibly designed for its own AI. Instead, SpaceX has signed deals with Google, Anthropic, Reflection AI, and Cursor, the AI coding startup that Musk later chose to acquire. CFO Bret Johnsen said these agreements put the company on a trajectory to reach $100 billion in annualized revenue run rate (ARR) by December. Musk was even more bullish, claiming that the $100 billion ARR figure is not a question mark and that the actual number could be higher.
But revenue is not profit, and building data centers is expensive. The bare-metal business of renting out compute is capital-intensive and subject to relentless technological obsolescence. Chips that are cutting-edge today may be outdated within a few years, forcing companies to spend constantly on upgrades. Moreover, compute is effectively a commodity, meaning that as more data centers are built, prices for chip capacity tend to fall. The more supply comes online, the harder it becomes for any single provider to command premium rates.
The Space Data Center Fantasy
Musk has claimed that he took SpaceX public because he wanted to build data centers in space. That idea, while visionary to some, seems like a monumental engineering challenge. If the company had trouble running a data center on the ground, it’s logical to question whether it could succeed with a facility in orbit, a feat that has never been attempted. Nevertheless, SpaceX has proposed an orbital data center consisting of as many as 1 million satellites to the Federal Communications Commission.
The application is notably light on technical details. There is no information about satellite size, deployment schedule, or how such a constellation would be maintained. This lack of specifics has led many analysts to view the proposal as more of a public relations exercise than a serious engineering initiative. Musk has also floated a more elaborate vision involving a chip-producing company called Terafab, which he says will manufacture one terawatt of chips every year, and a billion Optimus robots that will do the work once he figures out how to make them function properly.
The end goal, Musk claims, is to build a mass accelerator on the Moon. That idea sounds like it belongs in a science fiction novel, and Musk himself often draws on the tropes of mid-20th-century sci-fi writers. But there is little evidence that any of these projects are feasible in the near term. Scientists have already raised concerns that space data centers are a bad idea, and the history of Musk’s projects suggests caution. Remember the Hyperloop? It never got built. The motto should be: It’s all vaporware till it ships.
Rockets as a Support Business
If we strip away the grand talk, what remains is a company that launches rockets mostly for its own benefit. SpaceX is its own biggest customer for rocket launches, using them to deploy Starlink satellites and to support its growing data center ambitions. There simply aren’t enough external customers who want to buy rocket launches, which explains why the space segment’s revenue was so small. The rockets themselves may be impressive, but they are increasingly a supporting player rather than the star of the show.
In the meantime, SpaceX has become a significant customer for Tesla, buying $295 million worth of Megapack battery storage units. The company has also reportedly purchased a number of Cybertrucks, a vehicle that has been widely regarded as a flop in the automotive market. Tesla’s stock is down 25 percent since January, and Musk’s cross-selling between his various companies has raised eyebrows among investors who question whether these transactions are based on merit or on Musk’s desire to prop up his other ventures.
Political Connections and the Lockup Expiry
Musk’s political connections could also play a role in the company’s fortunes. He has reportedly planned to spend $100 million on the midterm elections, and it’s not unreasonable to think that such spending could yield favorable treatment for his businesses. Federal and state governments could potentially ease permitting for data centers or provide other benefits. But terrestrial data centers have become wildly unpopular across the political spectrum due to their massive energy consumption and water use, so it’s unclear how much political capital Musk can actually spend on that front.
There is a more immediate concern for shareholders, though. SpaceX’s insider lockups start expiring on August 6, which is tomorrow. If insiders decide to sell their shares, as short-sellers expect, the already slumping stock could fall even further. Nasdaq has changed its rules to accommodate SpaceX, meaning that any losses from insider selling will be felt by everyone who holds an index fund containing the stock. That’s a risk that goes beyond the company’s core operations.
In the end, the first public earnings report reveals a company that is fundamentally different from the one Musk often describes in his public statements. It is a telecom company with a successful internet service, a speculative AI infrastructure bet, and a rocket business that serves mostly itself. It may not be the path to Mars in six years, but it could be a profitable business if the AI and telecom segments continue to grow. The challenge is managing the enormous costs and competitive pressures of the data center industry, where even giants like Amazon and Microsoft have had to fight for market share. SpaceX may have a unique funding base, but in the bare-metal compute market, it is just another player.
Source: The Verge News