Once upon a time, SpaceX was easy to describe: it was the company that wanted to get people to Mars, and it happened to build rockets along the way. Those days are over. The company that just reported its first quarterly earnings as a publicly traded firm looks less like a space exploration venture and more like a telecommunications and AI infrastructure company. If anything, SpaceX has become a business powered by Starlink and by renting out enormous data centers to other AI companies. Calling it SpaceX, in that context, seems generous.

According to the earnings report, the 'space' part of SpaceX did not break $1 billion in quarterly revenue. That works out to just over 10 percent of the company's total sales. The rest comes from two entirely different businesses: a satellite internet operation called Starlink and a rapidly growing AI compute rental operation that has been described as a 'neocloud.' In other words, the rockets that made Elon Musk famous are now a side project for a company that is primarily selling connectivity and processing power.

Starlink is the quiet moneymaker

The strongest part of SpaceX is Starlink, which the company calls 'connectivity.' Starlink generated $4.2 billion in revenue in the quarter, making it the only segment of SpaceX that did not record an operating loss. That is a meaningful distinction: Starlink is not just the biggest revenue driver; it is also the only business unit that is actually profitable on an operating basis.

During the earnings call, Gwynne Shotwell, the company's president, outlined plans for a phone service designed to compete directly with AT&T, Verizon, and T-Mobile. The idea is to use satellites to provide coverage to ordinary mobile phones, cutting out traditional carriers in places where terrestrial networks are unavailable or unreliable. That would put Starlink in an even more direct collision course with incumbent wireless operators, and it suggests that the telecom side of SpaceX still has room to grow.

But even that telecom expansion was overshadowed by the much larger amount of attention, and money, devoted to artificial intelligence. The company is spending heavily on data centers and is positioning itself as a major supplier of computing power to AI startups. The 'lion's share' of spending and hype is now concentrated on what the company broadly refers to as AI. Whether the actual AI products, including Musk's Grok model, are the reason for that spending is another question.

AI compute is now the biggest business

SpaceX invested $15.8 billion in AI-related spending in the second quarter alone. By comparison, spending on space and on connectivity was just over $1 billion each. That enormous gap shows where the company's priorities lie. It is no longer primarily a rocket company; it is a builder and operator of massive computing infrastructure.

That makes SpaceX a direct competitor with neocloud providers such as CoreWeave and Nebius, companies that have made their name by leasing data center capacity to AI businesses. The high demand for GPU-powered computing has driven a boom in such services, and SpaceX has pushed into that market with surprising speed.

A key part of the story is Colossus 1, the data center Musk originally built in Memphis for his AI company, xAI. The facility was supposed to power Grok, but xAI ran into serious problems running the complex. There were latency issues that made it difficult to train in-house models, and the center used a mix of newer and older chips that created bottlenecks. Rather than solve those problems, Musk decided to rent the capacity out to other companies. On the earnings call, he said only 10 percent of the compute SpaceX builds will go to Grok.

That is a remarkable admission. The company is effectively admitting that its own AI model could not make full use of the data center, and that the better business model is to sell that computing power to outside customers.

Renting compute is not the same as selling rockets

The AI compute business is already producing major deals. SpaceX has announced agreements with Google, Anthropic, Reflection AI, and Cursor, an AI company that Musk later decided to acquire. According to the company's chief financial officer, Bret Johnsen, those deals put SpaceX on a trajectory to reach $100 billion in annualized recurring revenue, or ARR, including a future contribution from Cursor. Musk was even more bullish, claiming that 'the $100 billion ARR in December is not a question mark' and that the actual number could end up being higher.

Still, revenue is not profit, and building data centers is incredibly expensive. The bare-metal compute business comes with real challenges. Hardware becomes obsolete quickly. Construction projects are prone to delays and cost overruns. And computing is essentially a commodity, which means companies compete on price. As more data centers are built, more compute becomes available, and prices for chips and processing time tend to fall.

That is a very different economic model from the one SpaceX used when it was launching rockets for customers. Rockets are unique, highly engineered products with few viable competitors. Compute, by contrast, is interchangeable. A customer who rents time from CoreWeave can, at least in principle, rent time from Nebius instead. That puts SpaceX in a much more competitive market than the one it dominated as a launch provider.

Space data centers remain a distant dream

Musk has said that one reason he took SpaceX public was to build data centers in space. It is a strange argument, especially given that the company has struggled to run a ground-based data center. Building the same thing in orbit, with all the added complexity of launch and maintenance, would be far harder.

SpaceX has proposed an orbital data center with as many as 1 million satellites to the Federal Communications Commission. But the application is light on technical details. There is no clear information about satellite size or deployment schedule, which makes it look more like a public relations move than a serious engineering plan. Musk has also talked about a chip producer called Terafab that would produce one terawatt of chips every year, and about a billion Optimus robots doing the work, once the robots have functional hands.

At the end of that fantasy is the claim that SpaceX wants to build a mass accelerator on the Moon. It is the kind of old-fashioned sci-fi future that Musk often invokes, reminiscent of mid-20th-century pulp novels. But none of it has been built. Hyperloop was never built. The idea of a Moon accelerator is even further from reality.

The financial pressures are real

There is a more immediate reason Musk might want to promote an ambitious vision. SpaceX's insider lockups start expiring on August 6th. Insiders who have been holding shares since before the public offering will soon be able to sell them. If enough insiders sell, the stock, which has already been slumping, could fall even further.

Short-sellers expect that to happen. That is one reason the company might want to remind investors about a mass accelerator on the Moon. It gives people something exciting to focus on instead of the company's actual financial picture, which is dominated by a capital-intensive AI compute business that faces serious competitive pressure.

There is also a strange relationship between SpaceX and Tesla. SpaceX has become an important customer for Tesla, buying $295 million in Megapack battery storage. It has also bought a number of Cybertrucks, the electric vehicle that has been a major disappointment for Tesla. Tesla stock is down 25 percent since January, and the company has been looking for ways to find new revenue streams. Musk's habit of selling to himself has helped both companies on paper, but it does not change the underlying challenges they face.

Musk may also be using his political connections to gain an advantage for the AI compute business. He has said he plans to spend $100 million on the midterms, and those political ties could make it easier to build terrestrial data centers, though data centers have become unpopular across the political spectrum. Or, as some might put it, he could be offering a message to AI companies: it would be a shame if something happened to your access to compute. Perhaps you should buy mine, just to be safe.

All of this raises the question of what SpaceX really is. It is a company that launches rockets, but mostly for itself. It has a satellite internet business that is profitable but faces new competition. And it is running a risky, capital-intensive AI compute business that emerged from the failure of its own AI efforts. That is not the same thing as getting to Mars in six years. For now, the stock market is about to learn whether enough investors are willing to buy into the dream.

Source: The Verge News