Launch 1 million satellites, what does Musk want to do?

Elon Musk has started the "left hand to right hand" again.

On February 3, Elon Musk, founder of the American aerospace company SpaceX, announced on social media platform X that his artificial intelligence company xAI would be acquired by SpaceX to create "the world's most ambitious innovation engine." Recently, a SpaceX shareholder responded to the media, stating that the merger has been approved by the board of directors.

Clearly, this is not a simple business merger. Just on January 30, SpaceX submitted an application to the Federal Communications Commission (FCC) to launch "a satellite constellation with unprecedented computing power" to support the on-orbit application of AI large models. This constellation could contain up to 1 million satellites, forming a global orbital data center network. On February 4, local time, the FCC announced that it had officially accepted the application and began soliciting public opinions.

What Musk really wants to lay out is a brand new growth point: space computing. Space computing refers to deploying computing resources in space, using satellites and other infrastructure to achieve data processing and intelligent decision-making. Musk has publicly stated that the global ground AI infrastructure is heating up, and the Earth will eventually be unable to bear the computing power required by large models, making it inevitable to move computing power to space. So, can the space that he sees as "always sunny" and capable of providing unlimited energy really become the future breeding ground for large models?

Power Generation Efficiency Remains a Challenge

This is not the first time Musk has made a big splash with asset restructuring. In March last year, xAI, which had just been established for two years, acquired the social media platform X, forming a closed loop between social media and model computing power.

The merger of xAI and SpaceX means that the former will have stable launch capabilities, while the latter can secure long-term satellite launch orders, thus forming a complete closed loop of "rocket launch + space computing + AI large models." According to documents released by SpaceX, the constellation Musk is applying for will directly capture solar energy and use an energy conversion system to power the data centers in orbit, thereby supporting the "power-hungry" large models.

This concept has many attractions. According to data published by Barclays Bank's research team in 2025, the total capacity of large data center projects currently planned in the United States exceeds 45 gigawatts, and by 2030, it will exceed 200 gigawatts, accounting for 40% of the total electricity generation in the United States. High-altitude 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 solar energy.

"The currently recognized application of space computing is 'sky computing in the sky.'" Domestic aerospace expert Yu Guang told China News Weekly that, for example, in the case of remote sensing satellites, the traditional working mode is to transmit remote sensing data, such as photos taken by satellites, back to the ground via radio, and then process them into data products for utilization, which is "sky computing on the ground." The training and inference of large models are still conducted on the ground, meaning that a large amount of computing power is still deployed in core hubs such as intelligent computing centers. If large models can go to space, this data can be processed and analyzed directly in space, with the ground only receiving the analyzed results, thus freeing up ground equipment space The prospects of this application have become a consensus both 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 enhance the timeliness of space information services such as remote sensing, communication, and navigation, but also help solve multiple constraints faced by ultra-large-scale computing clusters, such as energy supply.

The core advantage of "Sky Computing" is efficiency. Yu Guang believes that large models need to exchange massive amounts of data with databases or other large models. If they can connect with low-orbit space-based internet satellites and communicate via microwave or laser links, their data exchange speed would be comparable to that of terrestrial fiber optic networks. However, to achieve global data coverage, the computing constellation should be a uniform constellation, which is unlikely to all be on the same orbital plane.

SpaceX's application did not disclose many technical details, such as satellite size and weight. Its proposed constellation will operate within an orbital range of 500 to 2000 kilometers above the Earth, with an orbital inclination of 30 degrees, and the height span of each orbital layer can reach 50 kilometers, with satellites communicating via laser.

With one million satellites filling the sky, the issue of power generation efficiency is inevitable. Yu Guang pointed out that the theoretically most suitable orbit for computing satellites—the dawn-dusk orbit—is a special type of sun-synchronous orbit, typically at an altitude of 700 to 800 kilometers above the Earth. The orbital plane of dawn-dusk satellites roughly coincides with the Earth's dawn-dusk plane, allowing the satellites to receive sunlight almost continuously. There are already many meteorological and research satellites in this orbit. In other orbits, satellites will spend some time in the Earth's shadow zone.

Clearly, the dawn-dusk orbit has only one orbital plane, which cannot accommodate all computing satellites. "In the worst-case scenario, satellites are obscured by the Earth for 40% of their operational time, unable to generate power through solar energy and relying solely on supporting batteries, which will significantly increase deployment costs. Even if SpaceX's reusable rocket 'Starship' successfully delivers payloads to space this year, the cost of generating power in space remains significantly higher than on the ground," Yu Guang said.

Additionally, referring to the International Space Station, which has the largest solar panels in the world, with a total area of over 3,000 square meters, equivalent to two football fields, and a maximum power output of about 160 kilowatts. However, Yu Guang believes that based on the power generation capacity per square meter of solar panels, to achieve the gigawatt-level power planned by Musk, the total area of solar panels would need to be measured in square kilometers, which poses a significant challenge for materials and engineering design.

"The Numbers Don't Add Up"

Undoubtedly, companies both domestically and internationally are accelerating the engineering implementation of space computing.

In November last year, the Starcloud-1 satellite launched by the U.S. startup Starcloud successfully operated the NVIDIA H100 chip, completing the on-orbit training and inference tasks of large models for the first time. Almost simultaneously, Google announced the launch of the "Sun Catcher Project," planning to launch solar-powered satellites equipped with self-developed large model training chips by 2027. In the aforementioned application, Musk proposed the vision of achieving "the lowest cost for space AI training" within five years and planned to deploy 100 gigawatts of computing power annually There are corresponding explorations in China as well. In May last year, with the launch of the first batch of 12 computing satellites into orbit, the "Three-Body Computing Constellation," led by the Zhejiang Zhijiang Laboratory, officially entered the networking stage. This project plans to be completed by 2030, consisting of 1,000 satellites that will cover a global network, making it the first fully interconnected space computing constellation in China. It is understood that the "Three-Body Computing Constellation" will be widely used in urban governance, emergency rescue, environmental monitoring, and other fields, with on-orbit computing capable of reducing data processing time to seconds.

In contrast, for the more long-term application of "Earth Data Space Computing," some industry insiders remain cautious, as the vast majority of data in human society still comes from the ground. "Earth Data Space Computing" means introducing a larger scale and denser on-orbit devices. Zhang Shancong, director of the Beijing Xingchen Future Space Technology Research Institute, stated at the aforementioned space computing power seminar that space computing power still faces challenges in terms of safety, including radiation protection and hardware reliability, as well as space debris and autonomous collision avoidance. This relies on breakthroughs in key technologies, such as ultra-large-scale attitude and orbit collaborative control and radiation-resistant chip reinforcement.

Yu Guang pointed out that large-scale on-orbit devices are greatly affected by cosmic radiation. After chips are sent into space, high-energy particles penetrate the chips, causing data errors, which increases the probability of errors in the large model itself and makes the devices more prone to damage. If an error occurs during training, the entire training process may need to be restarted. Even if used only for inference, a commercial large model that frequently makes errors cannot guarantee user experience.

If protective and error-correcting devices are added, it will introduce additional costs. A space science popularization expert told China News Weekly that Musk's ambition to lay out space computing power largely stems from the capacity support provided by "Starship." SpaceX plans to achieve full reusability of "Starship" this year, drastically reducing the cost of entering space by 99%, with the cost per pound of payload dropping below $100. SpaceX has applied for licenses for 42,000 satellites for the "Starlink" project, currently deploying over 9,600 and still expanding. Domestic reusable launch vehicle technology is still in the verification stage, with a limited number of satellites carried per launch, making the economics unfeasible.

Heat dissipation is also a major issue. Yu Guang pointed out that on the ground, mediums such as water and air can be used to dissipate heat through convection. However, in space, there is only radiation as a form of heat dissipation, which means that computing satellites may require very large heat dissipation panels while constantly adjusting their attitude to both collect solar energy and dissipate heat in the direction away from the sun.

Clearly, Musk is also aware of these issues. According to multiple media reports, Musk's team recently visited several photovoltaic companies in China to investigate the entire industrial chain, including equipment, silicon wafers, and battery components. This visit is not a spur-of-the-moment decision. The aforementioned space science popularization expert stated that Musk needs to find low-cost, radiation-resistant, and high-efficiency photovoltaic components to support his space computing power strategy.

Yu Guang pointed out that despite many difficulties and controversies, sending computing power into space is still a future trend. The FCC's attitude towards SpaceX's current application will continue to stir nerves in the industry. Musk himself seems very confident, as he quoted "Ad Astra" at the end of his tweet on X, which comes from the Latin proverb "Per aspera, ad astra," meaning "Through this difficult journey, to reach the stars." Risk Warning and Disclaimer

The market has risks, and investment should be cautious. This article does not constitute personal investment advice and does not take into account the specific investment goals, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Investment based on this is at one's own risk