Google's New AI Plan: Project Suncatcher

By Samaa B. '30

Back in November of last year, Google unveiled one of its most ambitious solutions to the AI infrastructure problem: Project Suncatcher.

Google’s plan is to eventually put artificial intelligence data centers into space and harness the sun’s energy to power them. If that sounds prone to failure, it’s because it has been.

This first launch, set to take place October 1st, marks a test to see whether the Tensor Processing Units (or TPUs), which Google uses to process the massive amounts of data needed for AI development, can survive the harsh environments of space. The satellite (which Google has named “MVP”) will orbit Earth for six years before falling back to Earth due to the planet's gravity and burning up in the atmosphere. It has also already passed preliminary tests at Google’s lab in San Francisco, indicating that there may be promising signs of success in the future.

It is, however, important to note that there are many drawbacks to this solution that critics often emphasize. One key vulnerability is that the MVP can only power its four processing units at a single kilowatt (about the energy it takes to power a clothing iron) for roughly 15 minutes before it overheats. At this point, the goal is to see if the chips can survive the harsh conditions, not how much they can process.

Another major potential flaw is the cost of this ambitious plan. While Google has not disclosed its full expenses, the cost of even sending a rocket into space is enormous, let alone the cost of building, developing, and researching the satellite. Many other AI companies, including OpenAI, have pledged to work on building space data centers, yet the financial challenges of developing and launching such technologies remain a major obstacle. Whether space-based data centers will ultimately be financially feasible remains an open question.