Global big tech corporations are turning to space to cope with power use that is surging on the back of growing demand for artificial intelligence (AI). The goal is to put data centers in space to receive unlimited solar power and handle massive AI computation. From Elon Musk's space company SpaceX to Google, Nvidia, Amazon, and Meta, they are jumping into related businesses.
Until recently, space data centers were considered a far-off future close to science fiction (SF), but they began drawing attention after Musk said early this year that "solar-powered space data centers will become a reality within two to three years." As AI data centers on Earth—so power-hungry they are called "electricity guzzlers"—are straining the grid, he proposed sending data centers into orbit as an alternative.
According to the industry on the 11th, the space data center Musk envisions is expected to handle massive AI computation demand by consolidating a network of hundreds of low Earth orbit satellites equipped with solar panels and high-performance graphics processing units (GPUs). Each satellite generates power through solar energy and is densely consolidated to one another via laser communications, operating like one gigantic data center.
In theory, because the sun shines 24 hours a day in space, solar generation efficiency is five to eight times higher than on Earth, and the cryogenic environment with an average temperature of minus 270 degrees Celsius can drastically cut server cooling expense.
Musk said, "Because space is not affected by day-night cycles, seasonal changes, or weather, the same solar panels are more than five times as efficient as on the ground, yielding about 30% more power," adding, "If you place the solar panels facing the sun and install a radiator on the opposite side that the sunlight does not reach, you get an efficient cooling system where cooling happens naturally."
SpaceX, which Musk leads, is set for a blockbuster initial public offering (IPO) this year, and it is said to be planning to use part of the funds raised through the listing to develop space data centers. SpaceX already has experience launching thousands of low Earth orbit satellites for its Starlink satellite internet service, giving it an advantage over rivals in building space data centers.
Starcloud, a U.S. startup backed by Nvidia, succeeded in November last year in placing "Starcloud-1," a 60-kilogram satellite the size of a small refrigerator equipped with Nvidia H100 GPUs, into orbit. The company is training and running Google's open AI model "Gemma" in space. On the strength of these results, Starcloud raised an additional $170 million on Mar. 2026, just 17 months after its founding, ascending to unicorn status with a valuation of about $1.1 billion.
Starcloud plans in the long term to build in orbit a 5-gigawatt (GW) data center measuring 4 kilometers in both height and width.
Philip Johnston, Starcloud's chief executive officer (CEO), said, "Even including launch expense, space data centers will cost about one-tenth as much as ground data centers," adding, "We expect that within the next 10 years, most newly built data centers will be constructed in space."
Google is also preparing space data centers through "Project SunCatcher." The project consolidates solar-powered clusters of satellites carrying Google tensor processing units (TPUs) via laser optical communications, with plans to launch two test satellites early next year. Google described it as "a challenge to massively scale machine learning computation in space."
Meta also signed a deal last month to receive up to 1 gigawatt (GW) of space solar energy from space startup Overview Energy to secure power for AI data centers. Overview Energy is developing satellites that collect solar energy in space. The collected energy is converted into near-infrared and transmitted to a large solar power plant on the ground, where the plant converts it back into electricity.
Big tech is trying to launch data centers into space because they have hit limits on the ground with power grids, cooling water, and sites for data centers becoming scarce. AI data centers must run massive servers 24 hours a day, and they also need large volumes of cooling water to chill heat-intensive servers. Cooling currently accounts for about 40% of AI data center operating expense.
The International Energy Agency (IEA) projected that global data center power consumption will rise an average of 15% a year for six years from 2024, reaching around 950 TWh (terawatt-hours) by 2030. That equals the annual power output of 120 large nuclear power plants. The IEA said, "The growth rate of data center power consumption is more than four times faster than that of every other industry."
However, building space data centers is expected to take time to commercialize because there are many technical and regulatory hurdles to clear. Experts cite issues such as high launch expense, the risk of space debris, component damage from space radiation, and the near impossibility of repairs or maintenance. They say space data centers can secure practicality and price competitiveness only after satellite launch costs fall to one-tenth of today's levels and tasks such as developing equipment and semiconductors that withstand space radiation are completed first.
Deutsche Bank said, "Space data centers will not be cost-competitive until at least the early 2030s."