How would AI data centers in space even work? A former NASA robotics chief explains

Follow ZDNET: Add us as a preferred source on Google.ZDNET's key takeawaysA robust space-based future lies ahead for the IT industry. It's still too expensive to launch data centers, but costs are dropping.Robots and predictive analytics will play a crucial role in maintenance.In the very near future, some or all of your workloads and data may be processed in space-based data centers. But will you notice? Probably not, a former NASA leader has predicted.A robust space-based future lies ahead for the IT industry, said Dr. Robert Ambrose, former chief of NASA's Software, Robotics and Simulation Division, and currently chairman of robotics and AI at consultancy Alliant. Moving data processing into orbit to handle the insatiable demands of AI and the digital economy may have compelling benefits. Earth-based data centers have met resistance from community and environmental groups who cite impacts on energy and water, with the facilities even creating some forms of pollution. There is also uncertainty as to how existing resources will meet the rising energy demands of AI processing. Also: The cloud vs. clouded leopard: America's data center backlash on display at Nashville ZooAmbrose said space-based data centers could offer freedom from the energy grid. They would consume energy via attached solar arrays, rather than from land-based power grids. "You collect electricity in space, and you eject heat in space. The only thing that comes to Earth is data," he said. While full-fledged space data centers still may be a few years out, space-based data processing is already happening in various forms today. "Much of our data already is at some point doing a transit through space," Ambrose said. "For example, if you're taking a picture of a forest fire from a spacecraft, and doing analysis or enhancements on it, your data is already in space."Overcoming challengesWhile space-based data centers offer an interesting alternative to land-based data centers, they also have their own share of issues. "The challenge now becomes managing the logistics of an orbital data ecosystem," according to the authors of a report out of retail estate specialist JLL. Also: Nvidia wants to own your AI data center from end to endLaunch economics, operational reliability, and debris management pose challenges. Currently, it costs SpaceX about $2,900 per kilogram to launch a payload, and that number needs to drop below $500 to make space data centers economically viable. "Launch costs drive all kinds of aspects of the space economy," Ambrose said, predicting that the new heavy-lift rockets being developed by private space companies, such as SpaceX and Blue Origin, may help drive down the cost of launching per kilogram of material in space.With commercial applications and services, a user may not be aware where their requests are being processed in an orbital data center -- and that may not even matter. "Did my AI product that I just got come from something in space, or a data center in North America?" Ambrose said. "It won't speak with a space accent. You'll have no idea."Larger companies are likely to maintain their own dedicated space-based resources to ensure resilience: "Space will become a really good place to have backups and alternate capabilities," said Ambrose. "An earthquake, hurricane, or forest fire is not going to take out a data center in space." Also: I quit ChatGPT for a free, private, and local AI called Ollama - here's whyLatency is not an issue either, as most data centers will reside in geostationary orbit about 300 miles from the ground, with data transmissions traveling to terrestrial antennae at the speed of light. However, if data centers make it to the moon, you can expect a 1.6-second lag back and forth with transmissions, Ambrose added. Will space-based data center performance slow down due to wear and tear on orbital data centers, starting with damage from launch vibrations? Launches can be extremely shaky events, and NASA and the space industry already have processes to ensure all payloads remain stable, Ambrose explained. At NASA, for instance, "every piece of flight hardware that I've ever been responsible for that's been launched into space has had to go through vibration tests. The space industry has created a very rigorous way to test for vibration to survive launch."Predictive maintenanceOnce in orbit, it's impractical and expensive to get humans up to the stations to make any fixes, of course. Therefore, remote management and maintenance of these data centers will require robotics and analytics. NASA's maintenance of the International Space Station provides a model for orbital data centers. "We knew when things would break," said Ambrose. "We used robots and some people, with the spare part sitting right next to the part that was being used. Like a critical ammonia pump that was part of the radiator, for example. So if there's ever a problem, we could move all the hoses over to the new pump, turn it on, and check it out."The key to continuous data center performance will be to automatically and robotically perform routine preventive maintenance versus emergency maintenance, Ambrose emphasized. With the space station pump, for example, "we used analytics to estimate when we thought the old pump would break, and scheduled moving over to the new pump before it broke."Also: How ChatGPT's new Lockdown mode protects you from data theft (and what else it does)Space-based data centers will have scheduled maintenance based on analytics of the expected life of components. With this predictive maintenance, "it's better to change out critical elements before they break on a nice low-drama schedule," Ambrose said. Such predictive maintenance is unlikely to impact data center performance. "If you have massive banks of computers that form the computational backbone, taking down small parts of it at a time to do maintenance is probably trivial. At any given time you're only taking down one out of a t...

Indexed and credited by AIPROPX. Originating outlet: ZDNet. Open at source →