SpaceX Prints AI Money While Texas Kills the Power Switch

Here is a sentence you probably did not expect to read today: SpaceX, the company that builds rockets and shoots them into space, now makes more money renting out servers than it does launching anything.

The company's AI compute division pulled in $2.6 billion in revenue this quarter — more than three times what it made in the same period last year. Meanwhile, the actual space business — the rockets, the launches, the whole reason anyone knew SpaceX's name in the first place — generated $962 million. That is not a rounding error. The company that put a car in orbit is now, functionally, a cloud computing provider that also happens to own some rockets.

The pivot happened faster than most people noticed. SpaceX had originally built out significant data center capacity to train its own Grok AI models. But Grok fell behind competitors, and rather than let expensive hardware sit idle, the company started selling that compute capacity to others. Deals with Anthropic in May and Google in June followed, putting SpaceX in direct competition with companies like CoreWeave — the so-called neoclouds that have become Wall Street darlings by renting GPU capacity to AI labs.

What makes this particularly interesting is how openly SpaceX leaned into the identity shift. In documents filed ahead of its record-breaking IPO — the largest ever — the company stated plainly that it expected AI to be the primary source of its value. Not Starship. Not Starlink. AI compute. Elon Musk backed that up on an investor call, claiming SpaceX is building AI infrastructure faster than anyone else on the planet.

That ambition is expensive. Capital expenditures hit $18.37 billion this quarter, a staggering number even by tech industry standards. The AI division itself lost $1.5 billion, though that loss is slightly narrower than last year. SpaceX overall lost $143 million for the quarter, an improvement, but still a loss.

The one part of the business actually making money is Starlink, the satellite internet service, which brought in $4.2 billion in revenue. That matters because Starlink is also the reason SpaceX needs Starship to work. The next generation of Starlink satellites are heavier, and only Starship can carry them in the quantities needed. SpaceX confirmed it has launched 20 of those next-generation satellites already, though a full deployment of 60 at once is still a work in progress.

SpaceX also agreed to acquire Cursor, the AI coding tool, which would give it an actual enterprise software product to sell. That deal has not closed yet, pending regulatory approval, and Musk was careful not to oversell the timeline.

Despite beating analyst estimates on revenue, the stock dipped after hours following an initial burst of enthusiasm. Markets, apparently, are not yet sure what to make of a rocket company that moonlights as a data center.

Less than a year ago, Texas Governor Greg Abbott was calling his state the epicenter of AI development. This week, he hit the pause button — hard.

Abbott has ordered a moratorium on all new power grid connections for data centers, directing regulators at the Public Utility Commission of Texas and grid operators at ERCOT to conduct a full audit of every data center project currently in the interconnection queue. Until developers hand over detailed information about their power needs, financing, ownership structure, and water usage, nobody is getting plugged in.

The numbers behind this decision are genuinely alarming. ERCOT's interconnection queue currently holds more than 1,800 projects representing over 474 gigawatts of connection requests. To put that in perspective, Texas's all-time record peak electricity demand is somewhere around 85 gigawatts. The queue represents more than five times that. And roughly 90 percent of those requests come from data centers.

Texas spent years making itself the most attractive state in the country for data center development. Cheap land, relatively abundant energy, fewer regulations, and aggressive tax incentives made it a no-brainer for developers. The state was on track to overtake Virginia as the largest data center market in the US. That strategy worked — perhaps too well.

The AI boom supercharged demand that was already growing quickly. ERCOT's own forecasts suggest that data center growth and related factors could cause statewide electricity demand to double its current record by 2032. Texas leads the country in adding new power generation, and it still cannot keep pace with what is being asked of it on paper.

The financial picture is also getting uncomfortable. A data center tax break that passed with bipartisan support back in 2014 — back when the stakes were much lower — has ballooned into more than $1 billion in annual tax breaks for developers. The state estimates it could forfeit $3.2 billion in sales tax revenue over just the next two years. That is a significant bill for a state that positions itself as fiscally conservative.

It is worth noting that many of the projects sitting in ERCOT's queue will never actually get built. Interconnection queues are notoriously bloated with speculative projects that fall apart for financial or logistical reasons. But even a fraction of 474 gigawatts materializing would be a serious stress test for the grid.

What Abbott is really doing here is buying time and information. The audit requirement is a way to separate serious projects from placeholder applications, and to force transparency on who owns what and how much state money they are collecting. Whether that is enough to actually manage what is coming is a different question entirely — but it is a more honest reckoning with the costs of the AI buildout than most states have been willing to have.