Google Launches Satellite Prototype to Test AI Data Centers in Space

Google has launched a refrigerator-sized prototype satellite into space to test running artificial intelligence chips in orbit and harnessing solar energy.

The Test

The spacecraft carries four specialty chips called Tensor Processing Units, or TPUs, designed for machine learning. These chips will run a version of Google's Gemma AI model for 15 minutes at a time to answer simple queries. The short run time is due to heat management limits. The mission will check how the chips handle space radiation, thermal extremes, and the physical stress of spaceflight. A SpaceX rocket carried the satellite into orbit, while the aerospace company Planet helped develop it and will get it running. Google plans to test the mission for a year, and hopes to launch two more satellites next year to test laser connections.

Growing Demand

Demand for AI services currently exceeds supply. At an annual developer summit in May, CEO Sundar Pichai stated that capital expenditures this year will reach $180 billion to $190 billion as the company builds computing infrastructure. Meanwhile, public opposition to power-hungry data centers on Earth is increasing. Placing data centers in orbit offers virtually unlimited solar energy and avoids terrestrial protesters.

Challenges Ahead

Significant challenges remain for space-based computing. Carnegie Mellon University electrical and computer engineering professor Brandon Lucia noted that heat is a major problem because the vacuum of space prevents wind or water cooling. Google is working on pipes and radiators to wick heat away, though heavy radiators increase launch costs. Maintenance is another hurdle since technicians cannot easily fix malfunctioning hardware hundreds of miles above Earth. Rocket fuels and atmospheric pollution during re-entry also raise environmental concerns, while high launch costs and slower laser communication pose economic questions. Project Suncatcher lead Travis Beals admitted that space data centers will likely take longer than five years to become economically viable.

Context

AI data centers require massive amounts of electricity and cooling capacity, leading tech companies to explore alternative energy sources and locations. While space offers constant solar power without land use conflicts, overcoming the extreme physical conditions and high launch costs remains a major engineering hurdle for the industry.