Space Exploration Technologies Corp., widely known as SpaceX, is aggressively pursuing its orbital data center project, with plans to launch satellites equipped with artificial intelligence chips as early as next year. The initiative aims to construct a space-based computing network comprised of millions of solar-powered satellites. In its initial public offering documents, the company candidly acknowledged that the project involves “significant technical complexity, unproven or non-existent technology,” and that its commercial viability remains uncertain.

Company founder Elon Musk assured investors this month that this is not a distant-future endeavor. Following a highly publicized IPO in June, SpaceX has painted a picture of a trillion-dollar market opportunity. Musk stated on social media platform X on Monday that the first batch of AI satellites, powered by Nvidia chips, is slated for launch in the fourth quarter of 2027, with “mass deployment” targeted for 2028. However, company documents, technical literature, and analysis from aerospace engineers indicate that SpaceX must overcome three interconnected challenges to achieve scaled operations: launching rockets at an unprecedented frequency and low cost, dissipating heat from high-energy computers in the vacuum of space, and regularly replacing satellites to keep pace with rapid chip technology advancements.

Christopher Smith, a senior aerospace engineer at the University of California, Berkeley's Space Sciences Laboratory, described the logistical challenge as “enormous.” He calculated that maintaining a million-satellite constellation, which SpaceX calls “Starmind,” would require launching more than nine Starship rockets daily, assuming each rocket carries 60 satellites and chips are replaced every five years. The Starship rocket is specifically designed to carry larger, heavier satellites and aims to reduce space transportation costs. Yet, the rocket has not achieved full reusability and has suffered multiple test flight failures, with only eight successful launches to date. SpaceX's most prolific current rocket, the Falcon 9, completed 165 launches last year, averaging one every two days. George Lodos, a space systems architect at MIT, believes an initial AI satellite constellation is feasible to deploy by 2028, with the key question being scale.

Heat dissipation in space presents another critical technical bottleneck. Although space temperatures are extremely low (around minus 270 degrees Celsius), the near-perfect vacuum lacks the particles necessary for thermal conduction and convection. Satellite design images published on SpaceX's website reveal plans for deployable radiators with a total area of 160 square meters—equivalent to two badminton courts. Compared to the passive cooling systems used in previous Starlink satellites, the AI data center's peak power consumption of roughly 250 kilowatts requires a more robust cooling solution. SpaceX plans to employ a pumped liquid cooling loop to transfer heat from the central computing unit to the radiators, similar to the ammonia cooling system used on the International Space Station. Mackenzie Sandberg, a thermal engineer at Advanced Cooling Technologies who previously worked at SpaceX, stated that the performance requirements are extremely demanding but achievable. Electronic equipment in space is also vulnerable to high-energy particles, which can cause data errors such as “bit flips.” Aerospace engineers note that radiation shielding is the primary defense for protecting electronics; concentrating circuits into a single unit can reduce the required shielding material, thus saving valuable weight.

Satellite replacement poses another significant challenge. Current Starlink satellites are replaced roughly every five years, deorbiting and burning up in the atmosphere. However, spacecraft reentry incineration can release metallic particles into the atmosphere, the long-term environmental impact of which remains unclear. Jonathan McDowell, a former astrophysicist at Harvard University and satellite launch expert, pointed out that scaling up existing launch plans by a hundredfold would have enormous, unknown consequences for the atmospheric environment. Additionally, the large solar panel arrays on orbital data centers—which McDowell calls “700-square-meter debris collectors”—increase the risk of collisions with space debris. Even if engineering challenges are resolved, SpaceX must still make the economics competitive with terrestrial data centers.

The company has historically reduced costs by controlling its supply chain and is replicating this strategy for Starship and the orbital data centers. SpaceX has begun constructing a natural gas pipeline to its Texas launch facility to accelerate fuel transport—each Starship launch requires approximately 3.4 million liters of liquid methane. Simultaneously, SpaceX is partnering with Tesla and Intel to build a $17 billion chip manufacturing plant in the state. In its IPO filing, the company admitted that the orbital AI program depends on “obtaining a sufficient quantity of AI chips, far exceeding what is currently available.” Some experts question whether placing data centers in orbit is the optimal path to building computing capacity. Berkeley's Smith remarked that the logistical challenge of launching millions of large spacecraft at costs competitive with ground-based projects “seems insane,” but he also cautioned against dismissing the plan entirely, noting, “We've seen Elon defy convention before.”