Big Tech Competes to Beam Solar Power from Space to Earth... 24-Hour Electricity Race [Into the World of AI]

As the use of artificial intelligence (AI) surges and competition over infrastructure intensifies, the limitations of existing power grids are becoming apparent, leading Big Tech companies to pursue the establishment of space-based data centers.

According to foreign media reports compiled on May 15, Google is in discussions with SpaceX regarding collaboration on space data centers. The talks concern the launch of satellites for Google’s “Suncatcher Project,” which aims to build AI data centers in space.

The Suncatcher Project is a research initiative in which Google’s Tensor Processing Units (TPUs) are installed on solar-powered satellites to create an AI cloud network in orbit. Google plans to launch an initial prototype in partnership with satellite manufacturer Planet Labs around 2027.

Meta has also recently signed a contract with startup Overview Energy to procure electricity from a space-based solar energy infrastructure. Overview Energy is developing infrastructure to collect solar energy in space and transmit it to ground facilities, providing 24-hour uninterrupted power generation. Thanks to this agreement, Meta will be able to utilize up to 1 gigawatt of capacity from Overview’s infrastructure. Anthropic has also shown interest in developing gigawatt-scale, space-based AI data centers by agreeing to use the computing power of SpaceX’s Colossus 1 facility.

Space-based data centers involve launching multiple satellites equipped with computing hardware such as processors and memory into low Earth orbit. Rather than being a single, massive facility, it is a modular system in which satellites are layered to distribute workloads. This approach allows for continuous power supply through solar energy and enables data processing in space, reducing data bottlenecks.

Elon Musk: "AI Can Run Cheapest in Space" vs. Matt Garman: "Not Economically Viable"

There are differing views on space-based data centers. SpaceX is appealing to investors with the argument that, within a few years, space data centers will become the optimal location for AI computing. Tesla CEO Elon Musk has claimed, "The place where AI can run most cheaply will be space." Philip Johnston, CEO of space technology company Starcloud, stated that the company’s AI satellites are already generating revenue by performing inference tasks in orbit. He further predicted that AI inference, from customer service voice agents to ChatGPT queries, will be processed in space.

However, some believe that realizing space-based data centers is a difficult prospect. They argue that, given the cost of manufacturing and launching satellites, terrestrial data centers are more economical. Matt Garman, CEO of Amazon Web Services (AWS), asserted, "We're still quite far off from that goal, as there aren’t enough rockets to launch a million satellites yet," adding, "The cost of sending payloads into space is enormous and not economically viable."

For Space Data Centers to Become Reality... Cost and Space Radiation Are Key Issues

Experts point out that the costs will have to decrease significantly for space-based data centers to become a reality. Space engineer Andrew McCalip estimated the cost of building a 1-gigawatt space data center at $42.4 billion—three times that of a terrestrial data center. McCalip explained, "Efforts are needed across multiple fields, including technology development, capital investment, and the component supply chain."

Google analyzed that the cost of reaching orbit would need to fall to $200 per kilogram. Currently, the launch cost is about $3,400–$3,600 per kilogram, meaning it would have to become roughly 18 times cheaper.

Developing technology to cool the electronic devices in data centers is also cited as a future challenge. Space radiation can cause "bit-flip errors," which damage chips or corrupt data. Effective cooling technology is needed to dissipate the heat generated by AI chips in the vacuum of space.

This content was produced with the assistance of AI translation services.