SpaceX’s newly filed public offering has opened a window into one of Elon Musk’s most ambitious ideas yet: using the proceeds to help turn the company into an artificial intelligence infrastructure giant by launching vast numbers of data center satellites into orbit. The concept is designed to overcome the power, water, and land constraints that are increasingly limiting conventional AI buildouts on Earth. But as striking as the vision sounds, it is already drawing skepticism from analysts and industry specialists who see a long list of technical and economic obstacles standing in the way.
The debate is not entirely theoretical. Microsoft once pursued a similarly unconventional escape from land based data center limits by submerging a container sized facility off the coast of Scotland. That experiment, known as Project Natick, met its technical objectives but was ultimately abandoned because customers did not want it and the economics did not work. That history is now being cited as a warning for SpaceX, whose idea is even more extreme, more expensive, and far less mature.
The underlying question is no longer whether data centers can be placed in unusual environments. It is whether doing so makes business sense compared with simply expanding capacity on the ground. For now, many industry voices believe that answer remains no.
SpaceX wants to move AI beyond Earth
According to the source material, Musk’s plan is to use SpaceX’s future public market funding to support an effort to place as many as 1 million AI focused satellites into orbit. The broader logic is straightforward: if AI demand is growing faster than Earth can comfortably support through traditional data center expansion, then computing capacity could eventually be pushed into space, where solar power is abundant and terrestrial energy and water limits no longer apply in the same way.
The proposal fits Musk’s broader strategy of linking his companies into a more integrated technology ecosystem. SpaceX acquired xAI in February, bringing together launch capacity, satellite infrastructure, the Grok chatbot, and the X social platform under a wider AI narrative. If the company goes public at the scale expected, it could become one of the largest IPOs ever, potentially giving Musk a vast financial base to pursue projects that would be out of reach for most other companies.
Still, even by Musk’s standards, the idea is unusually aggressive. Building AI data centers in orbit would require not just cheaper launches, but also solutions for cooling, maintenance, radiation exposure, and hardware replacement in an environment that is far more hostile than anything on Earth.
Microsoft’s failed underwater project is the obvious warning
Industry experts point to Microsoft’s Project Natick as the clearest historical parallel. The company lowered a shipping container sized data center onto the seabed in 2015, hoping to use natural seawater cooling and nearby renewable energy to cut costs and improve efficiency. Technically, the experiment succeeded. Commercially, it did not.
According to people familiar with the project, customers showed little interest in expanding the concept because conventional land based facilities remained easier to upgrade, scale, and maintain. That flexibility became even more important as AI accelerated and the pace of hardware improvement increased. A sealed, fixed unit that cannot be easily modified loses appeal quickly in a market where chip performance changes year by year.
That same weakness now hangs over SpaceX’s orbital plan. Analysts say the locked for life design that worked in theory underwater would become even more restrictive in space, where repairing, swapping, or expanding hardware would be vastly more expensive and operationally difficult.
The economics look punishing even before launch scale
The biggest obstacle may not be engineering, but cost. Analysts at MoffettNathanson estimate that Musk’s goal of building an orbital AI network on the scale he has described could run into the trillions of dollars. To make space based data centers economically competitive, launch costs would need to fall from current low thousands of dollars per kilogram to the low hundreds, a gap that remains enormous.
That challenge becomes even more severe when the scale of launches is considered. The same analysts estimate that meeting Musk’s objective would require around 3,000 Starship launches a year, or roughly eight launches a day. That would demand a degree of operational reliability and frequency that no rocket system has yet demonstrated, especially one still dealing with major developmental setbacks.
Starship is central to the plan because it is supposed to be fully reusable and capable of carrying much larger payloads than the Falcon rockets. But the vehicle remains years behind schedule and has suffered explosive failures in some of its suborbital test flights. That leaves the core financial and logistical assumption behind orbital AI infrastructure resting on a launch system that is not yet proven at anywhere near the required scale.
Most experts see orbital AI as a niche, not a replacement
Even among analysts who believe space based data centers have a future, the dominant view is that they are more likely to serve as a specialized complement rather than a substitute for terrestrial facilities. The most plausible use cases involve in orbit infrastructure such as military constellations, research platforms, or space stations that benefit from processing data locally instead of sending it back to Earth.
That narrower vision is very different from Musk’s suggestion that orbital AI could become a major answer to Earth’s future computing shortages. Analysts argue that the world still has many less radical options available first, including more efficient AI chips, expanded solar power, improved water recycling, and modular nuclear generation. Even Nvidia chief executive Jensen Huang has said that the economics of orbital AI remain unattractive and that the industry should focus on solving computing challenges on the ground before moving to space.
That does not mean the concept is meaningless. It means its most realistic future may be as a niche layer of specialized infrastructure rather than the foundation of the next great computing boom. SpaceX’s vision is bold, but for now, the cautionary lesson from Microsoft’s abandoned underwater project still applies: proving something can work is not the same as proving it deserves to exist at scale.
