Tesla CEO Elon Musk personally laid out his vision for a "space artificial intelligence (AI) data center" ahead of SpaceX's initial public offering (IPO). While SpaceX warned in its prospectus of the technical uncertainties of the business and the possibility of failing to commercialize it, Musk publicly underscored its feasibility to persuade investors.
On the 8th (local time), Musk posted a video shot at the Starlink terminal factory in Bastrop, Texas, on X (formerly Twitter) and introduced a plan for a solar-powered orbital AI data center. He said space AI satellites do not require entirely new technology and explained that much of the design, communications, and mass-production expertise accumulated while developing next-generation Starlink satellites can be leveraged.
SpaceX engineer Ian Dahl also appeared in the video and unveiled the initial design of "AI1," described as the first AI satellite. AI1 targets up to 150 kW of compute performance per satellite, with a wingspan of about 70 meters. That is comparable to one of Nvidia's high-performance AI racks for ground data centers. Musk said AI satellites have a simpler structure than existing Starlink satellites, making them favorable for mass production, and that inter-satellite links can use the laser communication network already used by Starlink.
The core competitive edge Musk touted is power and cooling. While ground data centers face growing burdens securing grid power and using cooling water, orbital satellites can directly harness solar power that is unaffected by the atmosphere and clouds. SpaceX believes that, based on solar cell costs, the unit price of electricity can be driven far below U.S. wholesale power prices. For cooling, the plan is to use radiative cooling that dissipates heat into space instead of using water.
Musk argued that the point when space AI computing costs fall below those of ground data centers could arrive sooner than expected. He also set a goal of putting 1 GW per year of AI compute infrastructure into orbit by the end of 2027, then scaling it by tenfold annually to reach terawatt (TW) levels. He acknowledged, however, that the target timeline carries uncertainties.
To back this, SpaceX is building a "Gigasat" factory in Bastrop as a production base. On a site of more than 1,000 acres, the company plans to vertically integrate production of satellites and solar materials, and, by later bundling in-house AI chip production with large-scale satellite launches, build a space AI infrastructure ecosystem. It has reportedly applied to the Federal Communications Commission (FCC) for approval to launch up to 1 million low Earth orbit AI satellites.
Market assessments are divided. While it is true that AI data centers face power shortages and cooling cost issues, many point out that the cost of launching servers into space and the difficulty of maintenance remain excessively high. Some researchers believe that, rather than moving general-purpose AI computing straight into space, limited-use cases that reduce ground transmission burdens—such as processing satellite observation data or edge computing within space networks—are more likely to materialize first.
SpaceX's prospectus did not hide these risks. The company said the space data center business includes unproven technologies and high complexity and may fail to achieve commercial results. Musk's explanation is, in effect, a direct rebuttal to that warning.
As the investment case for the SpaceX IPO expands from rockets and Starlink to AI infrastructure, space data centers have emerged as both the company's biggest swing at future value and its most contentious variable.