SpaceX’s business is increasingly about Starlink, AI compute, and index-fund exposure

Once, I had some questions about why SpaceX, Elon Musk’s healthiest company, acquired xAI, his sickliest one. Now I have some questions about why we’re calling the whole thing SpaceX.

By revenue, the company looks primarily like a telecom business and a compute-rental business, according to SpaceX’s first quarterly earnings statement as a public company. The space sector did not break a billion dollars this quarter and contributed only a little over 10 percent of revenue. SpaceX remains its own biggest customer, and the company still lacks enough outside demand for its rockets.

The telecom side, which SpaceX calls “connectivity,” is Starlink, the satellite internet service. It brought in $4.2 billion in revenue and was the only part of SpaceX that did not post an operating loss. Gwynne Shotwell said on the call that SpaceX plans a phone service meant to compete with AT&T, Verizon, and T-Mobile.

Most of the spending and attention, however, is now concentrated in what the company calls “AI.” Analyst Alexander Potter expects spending on the neocloud business, which leases data-center capacity to AI companies, to rise to $65 billion next year, Bloomberg said.

The importance of leasing data-center space was hard to miss. It made more money than SpaceX’s rockets and also drove more spending — $15.8 billion on AI alone in the second quarter. By comparison, spending on the space and connectivity sectors was a little over $1 billion each. That business makes SpaceX a competitor to neocloud companies such as CoreWeave and Nebius.

Except that was not the original plan.

Musk built SpaceX’s Colossus 1 data center in Memphis for Grok, his AI model, but xAI had trouble running the complex and decided to rent it out instead. The center also used a mix of newer and older chips that created bottlenecks. Musk said on the earnings call that only 10 percent of the compute SpaceX builds will go to Grok.

Revenue is not profit, and building data centers is expensive.

SpaceX now has deals with Google, Anthropic, Reflection AI, and Cursor, an AI company Musk later chose to acquire. On the second-quarter earnings call, Bret Johnsen said those deals put the company “on a trajectory, including contribution from Cursor, to reach $100 billion of ARR, or annualized revenue run rate.” Musk said the “$100 billion ARR in December is not a question mark,” and added that actual ARR might be higher.

That is impressive, but revenue is not profit, and data centers are costly to build.

The bare-metal business of renting compute has several problems: inevitable obsolescence, construction risk, and the fact that compute is a commodity, which means companies compete on cost. The more data centers that are built, the more compute is available, and the less companies can charge for chips.

Musk has claimed that he took SpaceX public because he wanted to build data centers in space. He has also proposed an orbital data center made of as many as 1 million satellites to the Federal Communications Commission. The application is light on technical detail, which suggests it may be more about public relations than execution.

Musk has also described a future in which a Musk-owned chip producer called Terafab makes one terawatt of chips each year, and a billion Optimus robots do the work once he figures out hands.

The end goal, Musk says, is to build a mass accelerator on the Moon.

The column rejects those claims and says Musk often talks about a science-fiction future that sounds closer to mid-20th-century speculative writing than to current industrial reality. It also cites scientists who say space data centers are a bad idea, while arguing that feasibility remains in question. The piece adds that Hyperloop was never built.

If the science-fiction framing is stripped away, the company looks like one that launches rockets mostly for itself, has a relatively successful satellite internet business, and is running a risky, capital-intensive compute business after failures in its in-house AI efforts.

That is less dreamy than getting to Mars in six years.

SpaceX’s insider lockups start expiring on August 6, which is tomorrow.

The company also matters to Tesla because it is an important customer, buying $295 million in Tesla Megapack battery storage. SpaceX has also been buying Cybertrucks. Tesla stock is down 25 percent since January.

Musk may be able to charge a premium for compute because of his political connections. He has also spent heavily on politics and may use those connections to make terrestrial data-center building easier, although the column says data centers are unpopular across the political spectrum.

The article suggests that Musk’s emphasis on a Moon-based future may also be a way to rally investors before the lockup expiry. If insiders sell, the already weak stock could fall further, and the writer notes that Nasdaq changed its rules for SpaceX so those losses would hit index-fund holders as well.

The column ends with a dismissive science-fiction line directed at Musk.