TLDR
- BNP Paribas says launch costs must fall below $300/kg before orbital data centers are economically viable — current costs are $1,500–$3,600/kg
- A 1-GW space data center would cost over $100B today vs. $35B–$50B for a ground-based one
- BNP expects Google, Amazon, and Meta to be among the first to run early-stage tests if costs come down
- Elon Musk says space will be the “most economically compelling” place for AI within 30–36 months
- Nvidia-backed Starcloud already launched the first Nvidia H100 GPU into space in November 2025
Space-based data centers are moving from science fiction to boardroom discussions. Investment bank BNP Paribas has published an analysis on the idea, and while it sees long-term potential, it says the economics don’t work yet.
At current launch costs — between $1,500 and $3,600 per kilogram — building a 1-gigawatt orbital data center would cost more than $100 billion. A comparable ground-based facility costs between $35 billion and $50 billion to build.
BNP analyst Nick Jones said the bank does not view orbital data centers as a “viable near- to medium-term solution.” He cited high launch costs, expensive space-grade hardware, and the challenges of cooling systems and electrical infrastructure in orbit.
For the economics to make sense, BNP says launch costs need to drop below $300 per kilogram. That would require a major reduction from today’s prices.
If costs do fall to that level, BNP expects Google, Amazon, and Meta to be among the first companies to run early alpha tests on orbital data centers. The bank did not give a specific timeline for when that might happen.
Why Space? The Energy Problem on Earth
The push toward space is partly driven by the energy demands of artificial intelligence. Data centers on Earth are consuming more power than ever. As of 2023, US data centers accounted for about 4.4% of total US electricity consumption, according to the Department of Energy.
McKinsey estimates that by 2030, data centers globally will require $6.7 trillion in investment to keep up with demand. Capital spending from big tech firms is expected to hit $600 billion in 2026, with Amazon alone planning to spend $200 billion.
Elon Musk has made space-based data centers a core part of SpaceX’s strategy. He has said that within 30 to 36 months, space will be the “most economically compelling place” to put AI infrastructure. SpaceX’s stated goal is to launch a constellation of one million satellites that operate as orbital data centers, each generating around 100 kilowatts of compute power per ton.
Musk’s argument is not primarily about saving money on energy costs. It’s about energy availability. He pointed out that electrical output outside of China has been largely flat, raising questions about where new ground-based data centers will source their power.
SpaceX Is Already Hiring
SpaceX has moved beyond just talking about the plan. Michael Nicolls, vice president of Starlink Engineering, posted on X that the company is hiring for “many critical engineering roles” related to space-based data centers, including a Space Lasers Engineer role based in Redmond, Washington.
The company announced it would acquire Musk’s AI company xAI, citing space-based AI as a long-term priority.
Early Tests Are Already Happening
In November 2025, Nvidia-backed startup Starcloud launched the first Nvidia H100 GPU into space on a SpaceX Falcon 9 rocket. The Starcloud-1 satellite weighed about 60 kilograms — roughly the size of a small refrigerator.
Starcloud’s long-term goal is a 5-gigawatt orbital data center about 4 kilometers wide, fitted with large solar and cooling panels.
BNP noted that over the long term, advances in ground-link technology, cooling systems, and solar power could bring the operating costs of orbital data centers closer to those of ground-based facilities.





