{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreieul3fy75vro2xeuzriqe74wisq5f4wbe3zrcsfc3d5w4mgxpjhva",
    "uri": "at://did:plc:qzjwstutqk2cy7df7jbzd2hx/app.bsky.feed.post/3mh6uikypvra2"
  },
  "coverImage": {
    "$type": "blob",
    "ref": {
      "$link": "bafkreifhvgmiwbcopgxktijbvtafvv4vyyfsx6trr5gzoq53sshupvnsh4"
    },
    "mimeType": "image/jpeg",
    "size": 379904
  },
  "path": "/article/4144330/whos-in-the-data-center-space-race.html",
  "publishedAt": "2026-03-16T14:32:02.000Z",
  "site": "https://www.networkworld.com",
  "tags": [
    "Artificial Intelligence, Data Center",
    "how much money",
    "Frederic Grandmont",
    "SpaceX filed a proposal",
    "Project Suncatcher",
    "Travis Beals",
    "launch an Nvidia H100 GPU into space",
    "press release",
    "launched a data center mission to the moon",
    "plans to put a data center in space by 2030",
    "data center on the ISS",
    "orbital data center satellite",
    "Star Catcher ran a ground-based test",
    "has already logged more than 1,000 days of operation",
    "China announced",
    "Star Computing project",
    "predicted last fall",
    "echoed Marc Benioff",
    "Onik Quddus",
    "sent its YAM-9 commercial satellite into orbit",
    "multi-sensor satellite constellation",
    "Tom Coughlin",
    "According to Gartner",
    "told The Indian Express",
    "Holger Mueller",
    "Richard Bonner"
  ],
  "textContent": "Tired of paying high-sky electric bills to power your server racks? Tired of paying even more to cool them down? Tired of vibrations and high temperatures messing up your superconducting quantum computers? Tired of your unreliable connections? Get ready for data centers… in SPAAAAAAAACE.\n\nIt’s easy to imagine the pitch for space-based data centers. But clearly there are obstacles to overcome, starting with how much money it takes to send anything up there. Plus, if you need to do any kind of maintenance or upgrade, how are you going to get to it? Not to mention the fact that your racks can barely survive a strong kick, much less being strapped to a rocket that literally moves by setting off a bunch of controlled explosions. And then there’s all the space radiation.\n\nBut wait. It’s not as crazy as it sounds. For one, think of the environmental benefits.\n\n“Data centers on Earth consume massive amounts of energy,” says Frederic Grandmont, technology and business development manager for space and defense systems at engineering company ABB Canada. “In a world where carbon footprints are under scrutiny, a self-cooling, solar-powered server farm seems to be an ideal end-state,” he adds.\n\nIn fact, we already have a ton of computing power up there already. There are computers on the International Space Station, in every ship we send up, and in pretty much every satellite whizzing around the Earth. We’ve already figured out the biggest problems—how to keep them working, how to power them, and how to cool them down.\n\nAnd laser-based communications in space are point-to-point since there’s no atmosphere or much of anything to get in the way, and they happen at the speed of light. In fact, even Earth-to-satellite communications are now getting pretty reliable and delivering low latency, atmosphere and all.\n\n“Data volumes and latency impediments are on the verge of being solved,” says Grandmont.\n\n“Mega-constellations of satellites with direct optical downlinks to Earth are essentially a fiber-optic connection without the cable,” he says. The data is routed at high speeds within nodes of the constellations and then sent to receiving systems with direct views to cloud-free ground stations on Earth.\n\nThe next challenges are to scale up, to get the launch costs down, and to figure out the whole maintenance-in-space problem.\n\n## Who’s in the race?\n\nAll the big players have thrown their hats in the ring.\n\nIn January, SpaceX filed a proposal with the FCC for an orbital data center constellation with up to one million satellites in low Earth orbit. SpaceX, which already has about 10,000 Starlink satellites in orbit, merged with xAI in February and said its new constellation will use solar to power space-based AI data centers. To fully protect its frequencies from interference, SpaceX will also have to file with ITU, the International Telecommunication Union.\n\nSpaceX did not explain when exactly its 1 million satellites will be going up. But it’s not the only company in the race.\n\nIn November, Google announced Project Suncatcher, a plan to put Google’s TPUs—their answer to Nvidia GPUs—in solar-powered satellite constellations with free-space optical links, so that they can run AI workloads in space.\n\nAccording to Google, a solar panel in orbit is eight times more productive than one on Earth and produces power continuously, reducing the need for batteries. “This approach would have tremendous potential for scale, and also minimizes impact on terrestrial resources,” wrote Travis Beals, the company’s senior director for paradigms of intelligence.\n\nGoogle has already tested and validated communications, figured out how to keep all the parts of the data center orbiting in formation, tested the hardiness of their chips against radiation—they passed—and figured out when launch costs would be low enough to make the whole project economically feasible. According to their calculations, they’ll need launch costs to be less than $200 per kilogram, and they expect that milestone to be reached in the mid-2030s. Their next step is to launch two prototype satellites by early 2027.\n\nIn mid-November, Nvidia teamed up with satellite company Starcloud to launch an Nvidia H100 GPU into space for the first time. The sixty-kilogram satellite is about the size of a small refrigerator, Nvidia said, and will have more than a hundred times more GPU compute than any previous space operation.\n\nAccording to Nvidia, space-based data centers will eventually cost ten times less than traditional ones, including the launch costs.\n\nOne company that’s been in the headlines recently for promising that this cost reduction will come sooner rather than later is SpaceX. “My estimate is that within 2 to 3 years, the lowest cost way to generate AI compute will be in space,” SpaceX CEO Elon Musk said in a press release in February.\n\nIn 2025, about 3,000 tons of payload was launched into space, most of it being Starlink satellites carried by SpaceX’s Falcon rocket. But this year, the company’s newest vehicle, Starship, will come online, he said. Starship’s 12th launch is scheduled for April. Starship is designed for full reusability and can carry significantly more cargo than the Falcon, reducing costs. So far, it’s had six successful fights and five explosions.\n\nThere are also a number of space startups in the game. Lonestar is a data storage and edge processing services startup that plans to serve data centers on Earth, in space, and even on the moon. In August, it launched a data center mission to the moon and did a successful test of their edge processing capabilities.\n\nIn Europe, a consortium that includes Ariane Group, Airbus, Thales Alenia Space, and HPE plans to put a data center in space by 2030.\n\nIn September, another startup, Axiom Space announced that it will put a data center on the ISS, with the first nodes to launch soon by the end of this year. The goal is to have three interconnected data center nodes in space by 2027 and to provide computing services to any satellites and spacecraft with compatible optical communication terminals.\n\nAetherflux, another space infrastructure startup, plans to put its first orbital data center satellite in orbit by the first quarter of 2027.\n\nIn other space infrastructure news, there’s a startup called Star Catcher that is building a space-based power grid. It has signed deals with Loft Orbital, Astro Digital, Starcloud, Satlyt, Mission Space, and other space computing players.\n\nIn November, Star Catcher ran a ground-based test, setting a new world record for wireless optical power transmission, and plans to have its first in-orbit demonstration next year. The idea is that Star Catcher collects solar power, concentrates it, and beams it to satellites that can’t collect enough on their own for all the on-board stuff they want to do.\n\nThe satellites collect that power using the solar panels they already have in place—no modifications required.\n\nMeanwhile, some of the biggest space data projects are already underway, in China. The Aurora 1000 space computer from Comospace has already logged more than 1,000 days of operation on a Jilin-1 satellite, and the next-gen Aurora 5000, which will have a GPU, is scheduled for trials in 2026.\n\nThen there’s the Three Body 12-satellite computing constellation carrying an 8-billion parameter AI model, which was launched last May. In February, China announced that, after nine months of in-orbit testing, the constellation successfully demonstrated key capabilities in space networking, computing, and model deployment. Called the Star Computing project, it will consist of 400 AI training satellites and 2,400 inference computing satellites, capable of tens of thousands of petabytes of inference and millions of petabytes of training power by 2030. By 2035, the network is expected to have enough computing power to support hundreds of millions of AI agents.\n\n## Extreme edge computing\n\nWe seem to be on the cusp of a major shift in how we use space for our computing needs.\n\nJeff Bezos predicted last fall that gigawatt-scale data centers will be built in space within the next 10 to 20 years and would outperform those based on Earth.\n\n“The lowest cost place for data centers is space,” echoed Marc Benioff at a conference in November.\n\nAccording to Onik Quddus, vice president with Booz Allen’s national security business, the break-even point will come sometime between 2030 and 2035. “At that point, the cost per kilogram will be low enough that deploying hardware in orbit becomes comparable to deploying a terrestrial data center,” he says.\n\nBut if you don’t want to wait years for the ability to send your processing to space, there are already companies renting edge-compute capabilities to run AI models in orbit.\n\n“For most technology managers, the first practical use cases won’t be shipping entire cloud workloads to orbit,” says Quddus. “It will be the emergence of edge processing on satellites.”\n\nSatellites can run compact AI models, including small language models, he says. “Which means we can interpret imagery, detect anomalies, compress insights, and respond autonomously.” And do that all in space, without waiting for data to be sent back to Earth, he adds.\n\nTake, for example, Loft Orbital, which sent its YAM-9 commercial satellite into orbit and successfully deployed it in November. It has four nodes and can handle multiple AI applications running simultaneously.\n\nWhy bother when it’s so much cheaper to run AI models here at home? Loft’s system includes access to sensors and data streams from the other Loft satellites already up in low Earth orbit.\n\nAnd, this year, Loft is partnering with Helsing, a European defense technology and AI company, to launch an AI-powered multi-sensor satellite constellation into orbit, designed for border surveillance, troop movement tracking, and infrastructure protection.\n\n“I believe we will see the early implementation of data center communication, storage, and processing in space—low Earth orbit or lunar—within the next three years,” says Tom Coughlin, IEEE fellow and president of Coughlin Associates, a consulting firm. And he predicts significant growth in this area, as well as general outer space commercialization, within the next 10 years.\n\nAnd these data centers won’t just be for serving Earth-based customers, he says, especially as human activities in space increase. “The need for local data center capabilities will also rise,” he says, “To mitigate latency issues associated with long-range communication.”\n\nBut not everyone is that optimistic. According to Gartner, space-based data centers won’t be useful for decades, so companies should focus on expanding capacity down here on Earth.\n\n“I honestly think the idea with the current landscape of putting data centers in space is ridiculous,” OpenAI CEO Sam Altman told The Indian Express in February.\n\nCurrent satellite computing can’t easily scale to data centers, agrees Holger Mueller, an analyst at Constellation Research. “Weight is still the restriction,” he says. “It’s the equivalent of you buying a tablet or small laptop to travel across Latin America versus putting in a data center in the Amazon. Different power requirements, investment, totally different setup.”\n\nThen there are issues like damaged solar panels from meteorite storms and satellite debris, he adds. “You would have to pay for operational redundancy, which is further investment.”\n\n“Data centers will be built where they are affordable,” he says. “I don’t see space happening soon. Remember the Microsoft submerged one? Crickets…”\n\nBut he agrees that solar power is nice, though the sun is only visible from one side of the planet at any given time. And space is cold, he says.\n\n## Cooling down in outer space\n\nIn fact, space is very cold. Close to absolute zero cold. But vacuum is also a great insulator, and there’s no air to move the heat around.\n\n“You can’t convect heat away,” says Richard Bonner, CTO at Accelsius, a liquid cooling company. Bonner has worked on NASA research projects about the challenge of cooling in space and is very familiar with the problem.\n\nA small proportion of the heat might be turned back into useful electricity, but that’s not really a solution, he says, because computer chips don’t get quite that hot.\n\nInstead, heat is radiated. When an object warms up, it generates electromagnetic radiation. This is how we’re able to see warm bodies at night with infrared glasses.\n\n“So, the only way to let that heat go is by radiation,” Bonnor says. “And it requires surface area, so you need these big panels to radiate the heat.” For data centers, these could be football-field-sized panels, he says.\n\nIt might seem that the backs of the solar panels would be a good place to radiate excess heat, but solar panels generate their own heat that needs to be radiated away.\n\nThe technology is there, Bonner says, and is already in use, though on a much smaller scale than what a full-sized data center would need. “Solar panels, radiator panels—not only do they exist, but they’ve existed for many, many decades,” he says.\n\nAnother kind of computing that benefits greatly from being surrounded by the coldness of space is quantum computing. Space could be uniquely beneficial here, Bonner says. “Quantum computers don’t have to dissipate a lot of heat, but they do require very, very cold temperatures. The other nice thing about space is that there’s no vibration.”\n\nBut that doesn’t mean that quantum computers are going to be launched into space tomorrow. “Given how sensitive these instruments are, there’s a lot that has to happen to allow them to survive launch,” he says.\n\nUnless they’re built up in space to start with. But that’s a subject for a different article, maybe a decade from now.",
  "title": "Who’s in the data-center space race?"
}