How Hard Is It to Build Orbital Data Centers, Actually? Ars Runs the Numbers on SpaceX’s Trillion-Dollar Constellation
Featured image: SpaceX AI1 satellite concept art with massive solar arrays and radiator panels; credit: SpaceX
SpaceX has pinned the bulk of its future value on orbital data centers. Not rockets. Not spacecraft. The company envisions launching and maintaining a constellation of 1 million satellites capable of generating 120 gigawatts of power to drive tens of millions, and potentially up to 100 million, frontier-class GPUs.
But how hard is it, actually?
Ars Technica’s senior space editor Eric Berger, in the second installment of a three-part series published July 15, ran the numbers. The short answer: a lot has to go right. The price tag ranges from $1.45 trillion in the best case to $9.8 trillion in the worst: that is before radiation damage, radiator efficiency, and satellite-to-satellite latency are fully tested at scale.
“This subject has sparked a broad debate about the near-term viability of this technology, both in terms of feasibility and whether it’s all hype now that SpaceX is a publicly traded company,” Berger wrote.
The AI1 Satellite
In June, Elon Musk and Ian Dahl, SpaceX’s director of satellite engineering, revealed details of the company’s first orbital data center design, called the AI1 satellite, in a promotional video. Each satellite would carry solar panels spanning about 600 square meters (1.5 times the size of a basketball court), generating 150 kilowatts of peak power and 120 kilowatts for computing.
“There’s not some magic that’s necessary that doesn’t exist,” Musk said in the video. “A lot of this is technology we’ve already made for Starlink V3 satellites. Basically, we don’t think this is a super hard problem.”
Iridium Communications chief executive Matt Desch, a veteran satellite industry leader, was more measured when asked about the concept during an earnings call earlier this year.
“It’s a hot, hot area right now of discussion, mainly because of Starlink’s announcement and some others,” Desch said. “It looks like a problem that can be solved in space… (But) there’s massive technical challenges to overcome.”
The Math: 10 to 42 Launches a Day
Ars constructed three scenarios based on Starship payload capacity, AI1 satellite mass, and launch costs.
| Scenario | Starship Payload | Sat Mass | Cost/Launch | Sats per Launch | Total Launches | Launches/Year |
|—|—|—|—|—|—|—|
| Optimistic | 200 t | 3.5 t | $20M | 57 | 17,500 | 3,500 |
| Neutral | 150 t | 5.5 t | $50M | 27 | 37,000 | 7,400 |
| Pessimistic | 100 t | 7.5 t | $100M | 13 | 77,000 | 15,300 |
Even in the optimistic scenario, that is 10 launches every single day. The pessimistic case demands 42 launches daily. By comparison, the entire world managed 329 orbital launch attempts last year, of which SpaceX conducted 170.
The cost of producing a million satellites adds up as well. Quilty Space estimates Starlink V3 satellites cost around $1 million each; AI1 satellites will be more expensive due to larger solar panels and high-end GPUs. Factoring in $100 billion for ground systems, the all-in estimates:
- Optimistic: $1.45 trillion ($350B launch, $1M/satellite)
- Neutral: $3.45 trillion ($1.85T launch, $1.5M/satellite)
- Pessimistic: $9.8 trillion ($7.7T launch, $2M/satellite)
Radiation: Manageable, But Unproven at Scale
SpaceX has learned from operating thousands of Starlink satellites for five years or more that many computing components are fairly radiation-tolerant. Sam Waldman, a physicist who worked at SpaceX on Starlink avionics, said power supplies are more vulnerable, but known mitigation techniques exist.
Musk said SpaceX plans to initially use Nvidia Rubin chips before developing its own. The startup Starcloud already tested an Nvidia H100 GPU in orbit and found it performed well with modest shielding. “The lifetime will be the same as on the ground, and there’s an argument to be made that it could be even longer,” said Starcloud CEO Philip Johnston.
Google’s experiments with its V6e Trillium TPU compute tray on orbit found that ionizing radiation can cause device failures over time, but that chips should operate reliably in space for about five years. That aligns with SpaceX’s planned five-to-seven-year satellite lifespan.
Heat: The Biggest Engineering Challenge
Shedding waste heat in a vacuum is arguably the most difficult problem. On Earth, cooling relies on convection; in space, only thermal radiation works. The International Space Station’s six ammonia-cooled radiators weigh a combined 6 metric tons and dissipate just 70 kilowatts of heat.
Starcloud is betting on a solution. Two-thirds of its engineering team is focused on developing a low-cost, low-mass deployable radiator. Its upcoming Starcloud-2 mission, a 450-kilogram satellite with 8 kilowatts of power generation, is due to launch in October and will demonstrate whether the approach scales.
“The ISS radiators are expensive and heavy,” Johnston said. “We’re focused on making them cheap and light.”
Latency: It Depends
Satellite-to-satellite latency could cripple large-scale AI training that depends on high-speed GPU interconnects, where rack-to-rack latency on Earth is measured in microseconds. But other workloads, particularly inference, are far less sensitive to latency and may adapt well to a distributed constellation.
The key question is how “shardable” a workload is and whether the shards fit within a single satellite’s capability. Berger concluded: “It’s a speed bump that might mean certain kinds of workloads aren’t a good fit, but it’s not a showstopper that sinks the entire idea.”
The Bottom Line
“There are no real fundamental barriers to building data centers in space, just some very serious technical problems to solve,” Berger wrote. “You need unprecedented heavy lift: reusable and rapid launch. You need the ability to manufacture the largest satellites humans have ever built and to build 100 times more of them than humans ever have for a single constellation. You have to hope that radiation’s impacts on chips are manageable and that radiational cooling scales.”
As 1ban.news covered in our earlier reporting, the orbital data center race has attracted a wave of startups and filings, from Orbital’s FCC plans for 100,000 satellites to environmental challenges that could slow deployment. The Ars analysis adds a dose of hard numbers to the conversation: the physics is well understood, but the economics demand a revolution in launch cadence and satellite manufacturing that has never been achieved.
“You also need a few trillion bucks,” Berger noted.