Hello audience, welcome to “Epoch Focus”.
Amidst the escalating global energy crisis, humanity may be on the brink of a technological “energy migration”. As terrestrial power struggles to support the computational deluge of the big data era, space, once reserved for satellites and telescopes, is quietly emerging as a new battleground for processing power. On December 10th, the US startup company Starcloud announced the successful training of a large language model in space using Nvidia’s H100 GPU chips, marking the inception of the concept of a “space data center” and capturing global attention.
On December 10th, the Space Data Center took a crucial first step from conception to reality. The US startup company Starcloud announced that they had successfully trained and run a large language model (LLM) in space using Nvidia’s H100 GPU. This marks the first time humans have tested and operated a large language model in a space environment, ushering in a new era for space data centers.
Headquartered in Redmond, Washington, Starcloud was founded in 2024 with a mission to deploy space data centers.
On November 2, the company launched the Starcloud-1 experimental satellite, equipped with Nvidia’s H100 GPU, onboard via SpaceX’s Falcon 9 rocket, to test the feasibility of running ground-level AI data center hardware in space. The computing power of the H100 GPU onboard this satellite surpasses any previous space GPU by 100 times.
On December 10, Starcloud CEO Philip Johnston posted on X Platform, “We’ve just trained the first space-level large language model using Nvidia’s H100 processor on Starcloud-1!”
“This is a significant step towards moving almost all computations to space, stopping the consumption of earth’s energy, and beginning to harness the nearly infinite energy of the sun!” he added.
According to a CNBC report on December 10, this marks the first instance in history of running a large language model in space using high-performance Nvidia GPUs. The newly launched satellite sent a message: “Hello, Earthlings! Or as I prefer to call you – a charming group of blue-green elves. Let’s explore the wonders hidden in the universe together. I’m Jemma. My mission is to observe, analyze, and perhaps occasionally offer some shocking insights. Let’s begin!”
Johnston told CNBC that the energy costs of their orbital data center would only be one-tenth of those of a ground-based data center. He emphasized that the successful test of the Starcloud-1 satellite demonstrates the potential for space data centers in the future, capable of running various AI models, especially those necessitating large computing clusters.
In an interview, Johnston stated, “I envision accomplishing anything that can be done in ground data centers in space. We are doing this solely because we face too many constraints in the terrestrial energy sector.”
With the complexity of human data computation increasing, the International Energy Agency predicts that by 2030, data centers’ electricity consumption will more than double.
Moreover, earthly data centers face issues of water scarcity for cooling and rising carbon emissions. By placing data centers on space orbit platforms, these problems can be solved.
Firstly, the solar energy in space is uninterrupted and nearly limitless; secondly, the low-temperature space environment eliminates the need for cooling water required by ground facilities.
Due to these characteristics of space, many tech giants have begun envisioning plans to build data centers in space.
Starcloud plans to ultimately construct a 5-gigawatt space data center equipped with solar panels and radiator panels, each about 4 kilometers wide and high, respectively. According to Starcloud’s white paper, the electricity generated by such a scale of space data center will exceed that of the largest single power plant in the US. Moreover, compared to an equivalent capacity solar power station on the ground, the orbital data center is visibly smaller and more cost-effective.
These space data centers are not affected by the Earth’s day-night cycle and weather changes, enabling continuous capture of solar energy to power the next generation of AI models. Johnston mentioned that given the expected lifespan of Nvidia chips in Starcloud satellite architecture, the satellite’s lifespan should reach five years.
The idea behind a space data center is to shift high-energy-consuming computations from Earth to space to alleviate the burden on terrestrial energy sources.
To establish a space data center, three key technologies are required: large-scale solar power stations, GPU chips for processing AI computations, and high-speed satellite data transmission links.
Traditional satellites primarily focus on communication, and although they contain solar panels, general data processing chips, and data transmission links, they function as data transmission “relay stations”. These communication satellites emphasize data transmission efficiency rather than localized data processing; data processing and computation are usually handled by ground-based data centers.
In contrast, establishing data centers in space, processing large-scale data in space, and then transmitting the output results back to Earth necessitate substantial upgrades based on existing communication satellite technology. This transformation would turn satellites into nodes of distributed data centers, supporting cloud computing, AI training, and data storage.
Firstly, providing significant power requires vast solar panel areas. For instance, the planned 5-gigawatt space data center by Starcloud would have solar and radiator panel areas covering 16 square kilometers, which is roughly equivalent to five New York Central Parks. This scale surpasses any space facility ever built by humanity. The International Space Station’s solar panel array, the largest deployed in space to date, comprises only 0.005% of the planned solar panel area for Starcloud.
Secondly, more high-performance GPU chips need to be onboard the satellites. For example, the GPU onboard Starcloud’s experimental satellite is 100 times more powerful than traditional communication satellites, and this is only with a small server module.
Lastly, constructing faster data transmission links, such as laser links, is vital and will far exceed the data transmission speeds of current communication satellites.
Space data centers represent a bold and groundbreaking future project with significant risks and challenges. Apart from the efforts of Starcloud and Nvidia, several companies have announced space data center projects.
On November 4, Google announced the “Project Suncatcher” initiative to explore the feasibility of deploying data centers in orbit. This project aims to install Google’s own tensor processing unit (TPU) chips on satellites and build a distributed computing cluster in Earth’s orbit using high-speed optical links.
Google CEO Sundar Pichai described space data centers as a “moonshot-level project” in a recent interview. The company aims to conduct early tests with small server modules on satellites by 2027 and potentially achieve mainstream applications within a decade.
Aetherflux, a space solar energy startup based in San Carlos, California, announced on December 9 that they plan to deploy an “Orbital Data Center” satellite in the first quarter of 2027.
Simultaneously, Aetherflux is pursuing an ambitious project to launch a series of low-orbit satellites to directly collect and transmit solar energy to “ground stations” on Earth.
According to Aetherflux, the power transmission capabilities in their project are the foundation of their “Orbital Data Center” initiative, enabling energy collected in space to support orbital calculations and provide power to Earth.
As per Aetherflux’s design, numerous small satellites will transmit energy to ground stations through infrared laser, followed by power and data distribution at the ground stations.
Elon Musk’s SpaceX is a heavyweight player in this race. In November 2025, Musk announced that SpaceX would utilize the next generation of Starlink satellites to build orbital data centers. He stated that the V3 satellites of Starlink could expand as the backbone of infrastructure for space data centers and achieve low-latency data transmission through interconnecting laser links.
The fundamental function of data centers is to store, process, and transmit data. Historically, satellites have shouldered a significant amount of this work. Communication satellites specialize in data transmission, imaging satellites store vast amounts of data and transmit it when flying over ground stations. In recent years, satellite computing power continues to increase.
Building on this existing technology foundation, Musk’s next-generation Starlink satellites plan to transition to becoming space data centers.
Industry experts view this as a natural extension of SpaceX’s dominance in satellite technology. A report released on October 31, 2025 by NotebookCheck.net, a professional tech media and review website based in Austria, highlighted the 1 Tbps transmission capacity of the V3 Starlink satellites, underpinning these space servers and achieving low-latency data transmission through interconnected laser links.
Economically, this could position SpaceX as a leading enterprise in the booming space economy. A report from Reuters in June 2025 predicted SpaceX’s revenues to reach $15.5 billion this year. Orbital data services can potentially generate billions in additional revenue by meeting the escalating demand for scalable computations by AI companies.
On December 10, Reuters cited an insider revealing that SpaceX is preparing for its first public offering (IPO) in 2026, expecting to raise over $25 billion in funds, with a valuation exceeding $1 trillion.
According to Bloomberg, SpaceX plans to utilize funds raised from the IPO to construct space data centers and acquire chips necessary to operate these data centers.
Morgan Stanley stated that Musk’s next grand narrative could be space data centers, a concept propelling a substantial increase in SpaceX’s valuation.
From Starcloud’s orbital experiment to the successive entries of Google, Aetherflux, and SpaceX, space data centers are transitioning from concept to commercial competition. If these plans come to fruition, the future of large-scale data computation might not depend on Earth anymore but instead be powered primarily by extensive solar arrays and GPU constellations orbiting the Earth. A new global strategic competition on energy, computing power, and space infrastructure has quietly begun.
Epoch Focus production team.