Launching 1 million satellites, what is Musk planning to do?

Elon Musk is once again “double-dealing.”

On February 3rd, Elon Musk, founder of SpaceX, announced on social platform X that his artificial intelligence company xAI will be acquired by SpaceX to create “the most ambitious innovation engine in the world.” Recently, a SpaceX shareholder responded to the media that the merger has been approved by the board of directors.

Obviously, this is not just a simple business merger. On January 30th, SpaceX submitted an application to the U.S. Federal Communications Commission (FCC) to launch “a satellite constellation with unprecedented computing power” to support on-orbit AI large models. This constellation could include up to 1 million satellites, forming a global orbital data center network. On February 4th, the FCC announced it had officially accepted the application and began soliciting public opinions.

What Musk truly aims to develop is a new growth point: space computing. Space computing refers to deploying computing resources in space, utilizing satellites and other infrastructure to perform data processing and intelligent decision-making. Musk has publicly stated that as ground-based AI infrastructure becomes increasingly intense, Earth will eventually be unable to support the computing power required by large models, making space-based computing an inevitable choice. So, can the space, which is “always sunny” and capable of providing unlimited energy in his view, truly become the future hotbed for large models?

Power Generation Efficiency Remains a Challenge

This is not Musk’s first high-profile asset restructuring. In March last year, xAI, which was just two years old, acquired social platform X, forming a closed loop of social media and model computing power.

The merger of xAI and SpaceX means the former will have stable launch capabilities, while the latter can secure long-term satellite launch orders, creating a complete “rocket launch + space computing + AI large model” closed loop. According to documents released by SpaceX, the constellation Musk applied for will directly capture solar energy through energy conversion systems to power data centers in orbit, supporting the “electricity-hungry” large models.

This concept is highly attractive. According to data released by the UK’s Barclays research team in 2025, the total capacity of large data center projects planned in the U.S. exceeds 45 gigawatts, and by 2030, it will surpass 200 gigawatts, accounting for 40% of the country’s total power generation. High-orbit space data centers can use high-intensity solar energy for extended periods, unaffected by the atmosphere, with power generation efficiency reaching five times that of ground-based solar.

“The currently recognized application of space computing is ‘day-to-day calculation’,” said Yu Guang, a senior domestic aerospace expert, in an interview with China News Weekly. Taking remote sensing satellites as an example, their traditional operation mode is to transmit remote sensing data, such as satellite photos, back to Earth via radio, then process the data to generate products, known as “day-to-day ground calculation.” Training and inference of large models are still done on the ground, meaning most computing power is deployed at core hubs like intelligent computing centers. If large models can go to space, these data can be processed and analyzed directly in orbit, with the ground only receiving the results, freeing up ground equipment space.

This application prospect has become a consensus domestically and internationally. At the “Star Computing·Intelligent Connection” space computing seminar held in January this year, He Baohong, chief engineer of the China Academy of Information and Communications Technology, pointed out that building a space computing system can not only effectively alleviate the transmission pressure of interstellar data and improve the timeliness of space information services such as remote sensing, communication, and navigation, but also help solve the multiple constraints faced by ultra-large-scale computing clusters, such as energy supply.

The core advantage of “day-to-day calculation” is efficiency. Yu Guang believes that large models require massive data exchange with databases or other large models. If connected via low-earth orbit satellite internet through microwave or laser links, the data exchange speed would be comparable to ground fiber optic networks. However, to achieve global data coverage, the constellation of computing satellites should be a uniform constellation, unlikely all on the same orbital plane.

SpaceX’s application did not disclose many technical details, such as satellite size and mass. The proposed constellation will operate in orbits 500–2000 kilometers above the ground, with an inclination of 30 degrees. Each orbital layer could have a height span of up to 50 kilometers, with laser communication between satellites.

With 1 million satellites filling the sky, power generation efficiency will inevitably be a concern. Yu Guang pointed out that the most suitable orbit for computing satellites—dawn-dusk orbit—is a special type of sun-synchronous orbit, typically at a height of 700–800 kilometers from Earth. Satellites in dawn-dusk orbit have orbital planes that roughly coincide with the dawn-dusk plane of Earth, allowing them to be almost constantly illuminated by the Sun. Many weather and scientific satellites are already in this orbit. In other orbits, satellites will enter Earth’s shadow for periods of time.

Clearly, dawn-dusk orbit has only one orbital plane, which cannot meet the deployment needs of all computing satellites. “The worst case is that satellites are shadowed by Earth for 40% of their operational time, unable to generate solar power, and can only rely on batteries, which greatly increases deployment costs. Even if SpaceX’s reusable Starship rockets achieve effective payload launches this year, the cost of solar power in space remains significantly higher than on the ground,” said Yu Guang.

Additionally, referencing the International Space Station, which has the world’s largest solar array with an area of over 3,000 square meters—about the size of two football fields—and a maximum power output of about 160 kilowatts. However, Yu Guang believes that to reach Musk’s planned gigawatt-level power, the total area of solar panels would need to be measured in square kilometers, which presents enormous material and engineering challenges.

“The numbers don’t add up”

Undoubtedly, both domestic and international companies are accelerating the engineering implementation of space computing.

In November last year, the U.S. startup Starcloud launched the Starcloud-1 satellite, which successfully operated Nvidia H100 chips, completing the first on-orbit training and inference of large models. Almost simultaneously, Google announced the “Captor Plan,” aiming to launch solar-powered satellites equipped with self-developed large model training chips by 2027. In the aforementioned application, Musk proposed a five-year goal to achieve the “lowest cost of space AI training,” planning to deploy 100 gigawatts of computing power annually.

Domestic efforts are also underway. In May last year, with the launch of the first 12 computing satellites, the “Trisolary Computing Constellation,” led by Zhejiang Zhijiang Laboratory, officially entered the networking stage. The project aims to be completed by 2030, with 1,000 satellites covering the globe, forming China’s first fully interconnected space computing constellation. It is understood that the “Trisolary Computing Constellation” will be widely used in urban governance, emergency rescue, environmental monitoring, and other fields, with in-orbit computing reducing data processing time to seconds.

In comparison, for the more long-term application of “ground-to-space day-to-day calculation,” some industry insiders remain cautious because most of human society’s data still originates from the ground. “Ground-to-space day-to-day calculation” would mean introducing larger and denser in-orbit devices. Zhang Shan, president of Beijing Xingchen Future Space Technology Research Institute, stated at the aforementioned space computing seminar that space computing still faces challenges in safety, including radiation protection and hardware reliability, space debris, and autonomous collision avoidance. These depend on breakthroughs in key technologies such as large-scale attitude and orbit control and radiation-hardened chips.

Yu Guang pointed out that large-scale in-orbit devices are greatly affected by cosmic radiation. After chips are launched into space, high-energy particles penetrating the chips can cause data errors, increasing the probability of errors in large models themselves, and making devices more prone to damage. If errors occur during training, the entire process may need to be restarted. Even for inference, a commercial large model that frequently malfunctions cannot guarantee user experience.

Adding protective and error-correcting devices would introduce additional costs. A space science popularization expert told China News Weekly that Musk’s ambitions for space computing largely stem from the capacity provided by the “Starship.” SpaceX plans to achieve full reusability of the “Starship” this year, drastically reducing space entry costs by 99%, with the cost per pound of payload dropping below $100. SpaceX has applied for licenses for 42,000 Starlink satellites, with over 9,600 already deployed and still expanding. Domestic reusable launch vehicle technology is still in validation, with limited satellite payloads per launch, so the numbers don’t quite add up.

Thermal management is another major issue. Yu Guang pointed out that on the ground, heat can be dissipated through water and air via convection. In space, radiation is the only form of heat dissipation, meaning computing satellites may need very large radiators and constantly adjust their attitude to both collect solar energy and radiate heat away from the Sun-facing side.

Clearly, Musk is aware of these issues. According to multiple media reports, Musk’s team recently visited several Chinese photovoltaic companies to inspect equipment, silicon wafers, battery modules, and the entire industry chain. This visit was not a spur-of-the-moment decision. The aforementioned space science expert said Musk needs to find low-cost, radiation-resistant, high-efficiency photovoltaic components to support his space computing strategy.

Yu Guang emphasized that despite many difficulties and controversies, space-based computing remains the future trend. The FCC’s attitude toward SpaceX’s application will continue to be a key concern in the industry. Meanwhile, Musk himself seems confident, ending his X post with the phrase “Ad Astra,” which originates from a Latin motto “Per aspera, ad astra,” meaning “Through hardships, to the stars.”

This article is from China News Weekly.

Risk Warning and Disclaimer

Market risks exist, and investments should be cautious. This article does not constitute personal investment advice and does not consider individual users’ specific investment goals, financial situations, or needs. Users should consider whether any opinions, viewpoints, or conclusions in this article are suitable for their particular circumstances. Invest accordingly at your own risk.