Google prepares to launch its first experimental satellite with AI hardware

Google is sending AI chips into space aboard a small test satellite launching early next month. Ars Technica reported that the mission, part of Google's Project Suncatcher initiative, aims to explore whether orbital servers can reduce the massive energy demands of land-based data centers by tapping into continuous solar power.

The spacecraft, constructed by Planet Labs and roughly the size of a domestic refrigerator, houses four standard Tensor Processing Units. These are the same processors Google uses in ground facilities rather than expensive, specialized space-hardened hardware. The mission will evaluate if standard commercial chips can endure intense launch vibrations alongside cosmic radiation without corrupting calculations.

Operating in orbit presents severe thermal hurdles because space lacks air to carry heat away. Google built a specialized radiator system, but the processors can still only run machine learning tasks for about fifteen minutes before needing to cool down. Additionally, as Ars Technica noted from reporting by The New York Times, the satellite's solar panels generate only about one kilowatt of electricity, which is roughly enough to power a basic household microwave.

While orbital computation sounds like an elegant answer to power grid constraints, actual implementation remains a distant prospect. It is still unclear whether high-speed laser communications and complex thermal controls can ever become cost effective, or if moving server farms off Earth simply trades land and water shortages for orbital debris risks and extreme launch expenses.