This image provided by NASA-TV shows the SpaceX Dragon commercial cargo ship taken from Canadarm 2’s video camera as Dragon approaches the International Space Station on Friday, May 25, 2012. The foreground is part of Canadarm 2. Expedition 31 flight engineers Don Pettit and Andre Kuipers will use their Canadarm 2 robotic arm Friday morning to grapple with a supply ship anchored on the Earth side of the station’s Harmony node. Dragon is scheduled to stay at anchor for about a week.
no credits
Orbital data centers will help build AI at scale, but many major advances are needed before they can become a reality.
SpaceX CEO Elon Musk is aiming for a launch in late 2027, but other industry officials say a major launch is likely in the 2030s.
“This is what the next decade is going to be about,” said Evelyn Chow, a portfolio manager at Neuberger. “We need to see massive satellite launches and the accompanying increased connectivity over the next four to five years before we start looking at the true scale of data centers in orbit,” he told Squawk Box Asia on Wednesday.
Blaine Curcio, founder of Orbital Gateway Consulting, said the 2030s timeline is a “fair valuation,” but warned that SpaceX has proven investors wrong before. “If you had asked any satellite industry expert in the late 2010s whether SpaceX could launch 10,000 satellites by 2025, they would have said, ‘No chance,’ including myself,” he said.
Here are some of the issues that need to be overcome:
cooling
“Cooling is a big issue,” Chow said. While terrestrial data centers can rely on liquid cooling to manage heat, cooling works very differently in the vacuum of space. “It also needs to be resistant to radiation.”
quickly become obsolete
Curcio believes there are hurdles related to the speed of technological progress.
“GPUs are evolving so rapidly that if you spend a huge amount of money launching a state-of-the-art data center into space, it could become obsolete after a few years,” he said.
Large data transfer
Transcelestial, a company specializing in laser-based communications, is working to solve the critical challenge of bringing large amounts of data back to Earth.
“Anyone can build a data center, but if they can’t communicate with these AI systems, those data centers are useless,” CEO and co-founder Rohit Jha said.
Powering hyperscale
Jha also said it could take another five to seven years for orbital data centers to reach hyperscale, and that level would likely require nuclear power.
