SpaceX and Google race to build AI data centers in space as scientists question the physics

SpaceX Files for Million-Satellite Constellation as Space Data Centers Race Accelerates

SpaceX filed with the Federal Communications Commission on January 30 for permission to launch up to 1 million satellites into low Earth orbit, each carrying computing hardware that would collectively form what the company described as a constellation with "unprecedented computing capacity to power advanced artificial intelligence models" 3

3

. The satellites would operate at altitudes between 500 and 2,000 kilometers, in orbits designed to maximize time in sunlight, and route traffic through SpaceX's existing Starlink network 3

3

. Elon Musk announced in March that SpaceX, which recently merged with his AI company xAI, would pursue placing data centers in space to bypass power constraints on Earth 4

4

. SpaceX filed for an IPO that could raise up to $75 billion, potentially making it the largest IPO in history, with proceeds intended to bankroll the orbital computing effort 2

2

.

Blue Origin followed seven weeks later with its own application for Project Sunrise, proposing 51,600 satellites in sun-synchronous orbits between 500 and 1,800 kilometers, complemented by the TeraWave constellation of 5,408 satellites providing ultra-high-speed optical backhaul 3

3

. Google announced Project Suncatcher in partnership with satellite imagery firm Planet, planning to launch two pilot satellites in early 2027 to test hardware in Earth's orbit 5

5

. CEO Sundar Pichai stated that "a decade or so away we'll be viewing it as a more normal way to build data centers" 5

5

.

Energy Consumption of AI Drives Search for Alternative Infrastructure

The commercial logic for extraterrestrial data centers rests on a genuine problem with AI's escalating electricity demands. Global data center electricity consumption reached roughly 415 terawatt-hours in 2024, and the International Energy Agency projects it could exceed 1,000 TWh by 2026, with accelerated AI servers driving 30 percent annual growth 3

3

. In Virginia alone, data centers consume 26 percent of total electricity supply, while Ireland's share could reach 32 percent by year's end 3

3

. Google alone has more than doubled its electricity consumption on data center use in the past five years, using 30.8 million megawatt-hours of electricity last year compared to 14.4 million in 2020 5

5

.

Proponents believe that in constantly illuminated sun-synchronous orbits, space-borne data centers would have uninterrupted access to solar power, while excess heat would be easily expelled into the cold vacuum of space 1

1

. Starcloud CEO Philip Johnston predicted extraterrestrial data centers will produce 10 times lower carbon emissions than their earthbound counterparts 5

5

. The startup raised $170 million at a $1.1 billion valuation in March, becoming the fastest unicorn in Y Combinator history just 17 months after completing the program 3

3

. Starcloud launched its first satellite carrying an Nvidia H100 GPU in November 2025, marking the first orbital test of an advanced AI chip 1

1

.

Thermal Management and Radiation Present Fundamental Physics Challenges

Scientists argue the physics makes orbital computing spectacularly difficult at any meaningful scale. The most fundamental challenge is cooling. In space, there is no air to carry heat away from processors, only radiative cooling, which requires vast surface areas 3

3

. Dissipating just one megawatt of thermal energy while keeping electronics at a stable 20 degrees Celsius demands approximately 1,200 square meters of radiator, roughly four tennis courts 3

3

. "Thermal management and cooling in space is generally a huge problem," said Lilly Eichinger, CEO of the Austrian space tech startup Satellives 1

1

.

Radiation presents the second structural problem. Low Earth orbit exposes unshielded chips to cosmic rays and trapped particles that induce bit flips and permanent circuit damage 3

3

. Radiation hardening adds 30 to 50 percent to hardware costs and reduces performance by 20 to 30 percent, while the alternative of triple modular redundancy means launching three copies of every chip 3

3

. Latency is the third constraint, as a million satellites spread across orbital shells cannot achieve the tight coupling required for frontier model training, where inter-node communication latencies must remain in the microsecond range 3

3

.

Microsoft's Undersea Data Center Failure Offers Cautionary Tale

Microsoft's "Project Natick" lowered a shipping-container-sized data center onto the seabed off Scotland in 2015, aiming to cut energy use through natural seawater cooling and tapping offshore wind and tidal power 2

2

. While the project successfully met all its technical targets, underwater data centers were abandoned more than two years ago due to a lack of client demand and unviable economics 2

2

. Customers were not interested in scaling them, instead expanding conventional land-based facilities that allowed cheaper, faster upgrades as AI development accelerated 2

2

. The sealed, "locked-for-life" design has limited flexibility, since AI chips are rapidly improving every year, while a satellite or undersea data center might be replaced only every five to seven years 2

2

.

Launch Costs and Economic Viability Remain Major Hurdles

Analysts at MoffettNathanson estimated that Musk's plan to put a million AI satellites in space would run into the trillions of dollars 2

2

. For sustainable data infrastructure to become commercially viable in orbit, launch costs would need to fall from today's low thousands of dollars per kilogram to the low hundreds of dollars per kilogram 2

2

. Musk's case hinges on Starship, SpaceX's next-generation rocket, which is designed to be fully reusable and carry far larger payloads, but Starship is years behind schedule and has suffered explosive setbacks in some of its 11 suborbital test flights since 2023 2

2

. MoffettNathanson estimates achieving Musk's goal would require 3,000 Starship launches a year, or eight per day 2

2

. IEEE Spectrum estimated that a one-gigawatt orbital data center would cost upwards of $50 billion 3

3

. Meanwhile, earthbound data centers are expected to require more than $5 trillion in capital expenditures by 2030, with hyperscalers including Alphabet, Amazon, Oracle, Meta and Microsoft issuing $121 billion in new debt through bonds in 2025 alone 5

5

.