Does a company make its sale of the idea of orbital data centres without the satellites already there in the air being judged of it? When Elon Musk chose to merge SpaceX and xAI, he has joined two companies at different levels of maturity: one conducts the most prolific private launch and satellite missions in the world, whereas the other is an AI company, just developing its computer base. The pull is uncomplicated. SpaceX introduces rockets, production pace, and a worldwide constellation of broadband; xAI introduces experience in data-centre operations and a flagship model in Grok. The resulting pitch, which is “AI, rockets, space-based internet, direct-to-mobile device communications” increases the story SpaceX will have to tell in front of the public markets, and it changes the focus to the question of whether the computational capabilities could be generated off-planet in large scale.
The most provocative statement made in the merger is the orbital computing proposal: Musk discussed “launching a constellation of a million satellites that operate as orbital data centers,” and projected that in two to three years, the cheapest and “lowest cost way to generate AI compute will be in space.” According to engineers and infrastructure analysts, the concept can be feasible in principle, and is associated with long lead-times and numerous execution dependencies, such as launch economics, power generation, thermal management, and a communications architecture that maintains end-to-end latency and reliability within real-world tolerances.
To aerospace and telecom watchers, a less futuristic, but more immediate test is available in Starlink. One group of characters cuts across the story. Individuals well informed about the outcomes of Space X have estimated Starlink to be the primary revenue generator, of approximately 50-80 percent of a total revenue, with profitability that has made the constellation the working engine to the wider plans. That reliance also implies that any new AI in space business plan is hooked to an existing business that is already dealing with capacity, control, and orbital crowding on a daily basis.
In practice performance is not even across the board in a manner that is relevant to broadband customers and to any proposal of compute in orbit in future. In a latency study carried out worldwide and grounded on measurements of 2,400 users in 27 countries, it was discovered that round-trip time changes drastically by location and that persistent latency spikes may happen on a daily basis in many users. The paper explains those spikes as a result of the routing of traffic between inter-satellite links and down to ground systems, a fact that should remind people that orbital networking is a systems problem- not just a rocket-and-satellite problem.
Such systems pressures are applied to orbital operations themselves. In an end of 2025 safety report submitted to the U.S. regulators, SpaceX estimated approximately 300,000 manoeuvres of collision avoidance in 2025 by Starlink satellites, and the maneuver threshold used by the company is more conservative than the industry standard. The sheer figure highlights the challenges required by a mega-constellation and the need to be able to make independent decisions in real-time, to coordinate with other operators, and an operational diversity that increases with each satellite placed into orbit.
At the same time, satellite broadband is being considered a less exotic option and more of an infrastructure solution, especially to terrain that is difficult to serve. States in the U.S. have started to consider low-Earth-orbit service as a part of rural programs where fiber deployments proceed very slowly or are too expensive to deploy; some early awards have allocated a small but significant portion of the funds to LEO providers. That policy change makes the Starlink usage a more serious day job, as a connectivity utility, which has compliance standards and is independently tested, and not an experimental technology.
It is in that context that a joint SpaceX-xAI venture will discover that markets have two questions simultaneously: can orbital compute be technically viable and can the current constellation continue to scale without performance volatility and sustainability challenges emerging as the new limiting factor? The merger links the patchy speculative upside with the discipline of running thousands of spacecraft in an overcrowded band of low Earth orbit, wherein any additional capability is contingent upon the reliability of the network that already covers the bills.