Tech giants eye orbit for AI data centers as energy demand soars

Surging energy demand of artificial intelligence (AI) data centers worldwide prompt tech firms to seriously consider whether it’s feasible to launch data centers into space.

The idea of orbital data centers regained traction after prominent aerospace figures like NASA Administrator Jared Isaacman proposed relocating AI compute infrastructure to Earth’s orbit to meet the sector’s ever-growing power requirements.

Regular data centers on our planet have critical servers, cooling systems, and backup power units required to keep global digital services operational.

There are 12,259 data centers across 179 countries worldwide, according to Data Center Map.

The US boasts the most number of data centers hosting 39% of the world’s total with 4,767 operational hubs, followed by the UK with 568, Germany with 533, France with 393, China with 376, India with 305, Australia with 296, Canada with 292, Italy with 262, and Japan with 261.

Russia houses some 188 data centers, while South Korea has 105, Türkiye operates 81, Israel is home to 67, Saudi Arabia has 61, and Iran houses 19 facilities, while North Korea has a single data center.

Modern AI models’ surging processing needs drastically inflate power consumption, cooling requirements, and physical land use of these facilities.

A single large data campus can generate up to 100m negawatts of wasted heat, which would otherwise be enough to power around 100,000 residential households.

The International Energy Agency (IEA) expects global data center electricity consumption to reach 485 terawatt-hours by 2025 and nearly double by 2030.

Traditional data centers will have made up around 3% of the world’s electricity demand by 2030, while the energy consumption of AI compute hubs is expected to triple during the same period.

Meanwhile, orbital data centers offer a radical alternative, providing uninterrupted and continuous access to solar energy and eliminating the need for land and cooling water.

Engineers will have to overcome massive obstacles to commercialize such a technology, however.

The costs of launching hardware into space, establishing a usable data transmission latency, and the impossibility of routine physical maintenance are among the few hurdles in the way of having widespread orbital data center adoption in the AI sector.

Slava G. Turyshev, a researcher at NASA’s Jet Propulsion Laboratory (JPL), told Anadolu that building a data center in space is not as simple as launching a server rack into orbit.

He stated that the orbit is already crowded with some 16,000 spacecraft orbiting the Earth, while this number is expected to increase exponentially.

Turyshev noted that deploying a one-megawatt node in space would require up to 6,000 square meters (64,583.5 square feet or nearly one and a half acres) of solar panels and would weigh “probably 75 tons.”

The system would also have to boast shielded processors to withstand the cosmic radiation in orbit, while requiring to be made up of sprawling, flexible structures, while accounting for solar arrays and radiators in modular sections to survive potential high-speed collisions.

“You will see a lot of movement in low Earth orbit because they will be reflecting the sun, and will see definitely see the activity in the sky, which okay, we can enjoy that,” he said.

He added that the aerospace industry will start producing tangible results in orbital computing within the next five years, while reaching a reliably functioning commercial satellite network dedicated to orbital data processing could be brought online in as early as seven years or up to 10 years.