Elon Musk: Expand the scale of Starlink V3 and enter the space computing power market
Recently, Elon Musk stated on X that SpaceX will expand the scale of Starlink V3 satellites and start building data centers in space to address the shortage of computing power resources in the AI era.
What? Is computing power really going to space?
Today, let's talk about the topic of "space computing power". As usual, we'll use AlphaEngine for interpretation.
1
Elon Musk's Bold Statement: Expand V3 Satellite Scale and Enter the Space Computing Power Market
With the significant increase in the demand for computing power in artificial intelligence, people's interest in space computing power (Space-based Data Centers) has also risen sharply.
In May this year, Eric Schmidt, the former CEO of Google, took over as the CEO of Relativity Space to layout in the field of space computing power.
In October, Jeff Bezos, the founder of Amazon, publicly stated that a gigawatt-scale data center will be built in space within the next 10 to 20 years.
Just the day before yesterday, after the technology media Ars reported on the relevant content of "autonomous assembly technology is expected to build large-scale data centers in space", Elon Musk responded on the social platform X that Starlink satellites can be used for this purpose.
He said on the X platform: "It can be achieved by simply expanding the scale of Starlink V3 satellites. These satellites are equipped with high-speed laser links, and SpaceX will promote this work."
Elon Musk's attention to space computing power has significantly increased the attention of this emerging industry.
Currently, SpaceX's Starlink V2 mini-satellites have a maximum downlink capacity of about 100 Gbps, while the capacity of V3 satellites is expected to increase by 10 times to reach 1 Tbps.
SpaceX plans to launch dozens of Starlink V3 satellites each time using the "Starship". These launches may take place as early as the first half of 2026.
2
The Unique Value of Space Computing Power Centers
A space computing power center refers to a modular computing power infrastructure deployed in space orbit. In essence, it is to move the data center from the ground to space.
By carrying high-performance computing payloads, it aims to achieve the core processing mode of "processing data in space", that is, directly processing the massive data generated by platforms such as satellites in orbit, thereby fundamentally breaking through the physical expansion bottleneck faced by ground computing power due to factors such as energy and land.
Facing the severe prediction that the global AIDC power demand will reach as high as 347 GW in 2030, space computing power centers show unique advantages.
In terms of energy consumption, by deploying efficient solar cell arrays, space computing power centers can generate electricity using solar energy. The power generation per unit area is 5 times that of the ground, achieving on-orbit self-sufficiency in energy and completely getting rid of the dependence on the ground power grid.
In terms of heat dissipation, the extremely cold vacuum environment of minus 270 degrees Celsius on the shaded side of space is used for efficient radiation heat dissipation. The heat dissipation efficiency is 3 times that of the ground, and there is no need to consume precious water resources, fundamentally solving the heat dissipation challenge that the ground data center has approached the physical limit.
3
From "Sensing in Space and Computing on the Ground" to "Processing Data in Space"
The space computing power center subverts the traditional data processing process of "sensing in space and computing on the ground" by constructing a new paradigm of "on-orbit processing + on-demand downlink".
In the traditional mode, all the massive raw data collected by satellites need to be transmitted back to the ground. However, limited by the satellite-ground communication bandwidth, the data transmission efficiency is low and the cost is high, resulting in a large amount of data being backlogged or discarded.
The space computing power center completes data cleaning, analysis, and intelligent extraction on orbit, and only transmits the most valuable analysis results and decision-making information back to the ground, realizing "processing data in space".
The first AI satellite planned to be launched by Starcloud is quite representative. It will carry the H100 chip, and its core function is to process the several terabytes of raw data generated by spacecraft and space stations every day.
This satellite can perform real-time analysis on satellite data, specifically covering scenarios such as synthetic aperture radar data analysis and deep space radio signal processing. This design directly bypasses the bottleneck problem of ground data transmission.
Coincidentally, the "Three-Body Computing Constellation" built by Zhijiang Laboratory also focuses on space computing. This constellation consists of 12 computing satellites, which not only realizes the interconnection between satellites in the entire orbit but also has complete on-orbit computing capabilities in space.
The computing power of a single satellite can reach 744 TOPS (i.e., trillions of operations per second), and the inter-satellite laser communication rate can reach up to 100 Gbps, which can efficiently support tasks with extremely high real-time requirements such as disaster monitoring and weather forecasting.
Figure: Launch plan of the Three-Body Computing Constellation, Minsheng Securities, AlphaEngine
4
Space Computing Power Centers vs. Traditional Ground Data Centers
Compared with traditional ground data centers, space computing power centers show subversive advantages in core dimensions such as technical architecture, cost structure, deployment mode, energy efficiency, and scalability.
Figure: Advantages and disadvantages of space computing power centers, AlphaEngine
Especially in terms of cost structure, space computing power centers have significant advantages.
With the goal of operating a 40-megawatt cluster for 10 years, if a traditional data center is used, the 10-year operating cost is about $167 million, of which energy consumption is as high as $140 million and cooling costs are about $7 million.
To achieve the same goal, the total expenditure of using space computing power is expected to be only $8.2 million.
The largest cost is the "one-time launch cost", about $5 million, followed by the cost of solar arrays, about $2 million. The long-term energy is directly provided by solar energy, and the energy cost is almost zero.
Figure: Cost structure of space computing power, StarCloud, AlphaEngine
5
From Fantasy to Industrial Implementation: Five Major Technical Challenges of Space Computing Power
Critics believe that space computing power is a fantasy and has too high a technical threshold. Is this really the case? Let's study the current technical problems of space computing power.
First is the challenge of radiation resistance and hardware reliability.
The extreme radiation environment in space poses a direct threat to computing hardware.
Computing nodes in the Earth's orbit need to deal with radiation problems such as cosmic rays, single-event upsets (SEU), and single-event latch-ups (SEL), which can cause logic errors or permanent damage to chips.
Therefore, the construction of space computing power requires the use of military-grade reinforced electronic equipment or redundant backup systems. For example, Axiom Space tries to use military-grade equipment to deal with the radiation environment, and Lonestar explores placing future lunar data centers in underground lava tubes to prevent radiation.
At the same time, multiple sets of computing module backups need to be designed to form hardware redundancy to deal with the risk of single-point failures.
The second problem is the design of the heat dissipation system.
Although the vacuum environment in space provides efficient radiation heat dissipation conditions, the thermal management of high-power chips (such as GPUs) still faces problems.
In a vacuum environment, heat cannot be dissipated through air convection and needs to rely on heat pipes or fluid loops to conduct heat to the radiation cooling plate and then dissipate heat through infrared radiation.
For example, Starcloud's high-computing-power satellites need to combine a hybrid solution of liquid cooling and large heat dissipation wings.
However, the heat dissipation system of high-power equipment (such as AI chips) requires a larger area of the radiation cooling plate, which increases the satellite's weight and thus drives up the launch cost.
The third technical problem lies in the stability of energy supply.
Although the efficiency of solar energy in space is 2 - 3 times higher than that on the ground, "power supply in the shadow area" is a key issue.
Satellites need to rely on energy storage batteries to maintain operation in the orbital shadow area. The capacity and lifespan of the energy storage system are key limiting factors.
To solve this problem, Starcloud plans to build a solar cell array of 5 km × 4 km. Such a large-scale battery array requires breakthroughs in the on-orbit deployment technology of giant structures.
Figure: Schematic diagram of the sun's permanent illumination orbit (following the terminator line throughout the year)
The fourth technical challenge is the communication bottleneck and autonomous operation and maintenance.
There is a delay in satellite-to-satellite and satellite-to-ground communication. Laser communication is relied on to achieve low-latency interconnection (such as the laser link between Starcloud and Starlink), but it still needs to overcome atmospheric interference and signal attenuation over long distances.
At the same time, the space computing power center operates without human maintenance for a long time, so a lightweight containerized software platform suitable for the space environment needs to be developed to support on-orbit autonomous decision-making and fault repair.
The fifth challenge lies in the launch cost and large-scale deployment.
Although reusable rockets (such as SpaceX's Starlink launch technology) have reduced the single launch cost, a gigawatt-scale data center (such as Starcloud's 5GW project) still requires large-scale networking, and the total cost is still high.
In the long run, the congestion problem in low-earth orbit may affect the heat dissipation efficiency and the selection of deployment locations.
6
Major Participants in the Space Computing Power Field
Currently, the space computing power field is in the early exploration stage, and the main players include startups and technology giants.
Representative startups include Starcloud (formerly Lumen Orbit), Axiom Space, Lonestar, etc.
Among them, Starcloud is a pioneer in space computing power, focusing on the construction of orbital data centers.
The company plans to launch the world's first AI satellite "Cloud-0" carrying the NVIDIA H100 chip, aiming to build a gigawatt-scale orbital data center. The computing performance of its H100 chipset in a zero-gravity environment is expected to be 100 times that of the International Space Station.
In addition to startups, technology giants have also started to layout in the field of space computing power.
NVIDIA cooperates with Starcloud through the Inception project to promote the deployment of orbital data center satellites. In 2025, it plans to launch the first satellite carrying the H100 chip to support high-density computing tasks.
Amazon's Project Kuiper plans to launch a low-earth orbit satellite internet service in Australia in the middle of 2026 to compete with Starlink.
In 2025, the company successfully launched the first batch of 27 satellites using the Atlas V rocket. In the future, it plans to combine the edge computing capabilities of AWS to deploy on-orbit AI data processing nodes.
Microsoft cooperates with SpaceX to launch the Azure Space plan, providing global cloud service access through Starlink satellites. It also plans to test satellites on orbit to deploy new software and hardware for the US government. Currently, the Azure Orbital Cloud Access function has entered the preview stage.
Meta jointly launched the "Space Llama" project with NVIDIA and HP to provide AI scientific research support for the International Space Station, analyzing the needs of astronauts in real-time and optimizing the operation process.
SpaceX is the most potential player in the field of space computing power. Its Starlink constellation has deployed a large number of low-earth orbit satellites and is promoting inter-satellite laser link technology.
Regarding SpaceX, I have conducted in-depth research before. Friends who are interested can read these two articles:
In-depth Decoding of SpaceX: Launch Cost, Starship Progress, Competitive Landscape, etc.
In-depth Decoding of SpaceX (Part 2): Second-Generation Constellation, Starshield, Elon Musk, and the Russia-Ukraine Conflict