Google thinks SpaceX’s Starship has to launch 1,600 times before space data centers get off the ground

If you rely on hyperscale cloud AI services, their future performance and cost may depend on reliable space-based data processing capabilities.

Google successfully launched its first orbital compute satellite, housing a Google Tensor Processing Unit (TPU), into space aboard a SpaceX rocket. This satellite, part of Google's Project Suncatcher, aims to test the viability of its custom AI chips in the harsh space environment, including power and thermal management challenges. The satellite will perform computations in 15-minute bursts to manage resources. Google is collaborating with Planet Labs on this initiative, with plans for two more purpose-built satellites next year that will attempt laser-based communication. The project is a long-term endeavor focused on building future space infrastructure and AI workloads, envisioning a network of 81 satellites. Google's research also explores the economic feasibility of scaling such data centers, estimating that SpaceX's Starship would need to conduct approximately 1,800 launches over ten years to achieve projected cost reductions for orbital data delivery.

Google's successful launch of a TPU into orbit represents a significant step towards their ambitious goal of establishing large-scale orbital data centers, indicating a commitment to the future of space-based computing. The rigorous testing of their chips in real space conditions, as detailed in their peer-reviewed white paper, suggests a proactive approach to overcoming the unique challenges of radiation and thermal management, potentially setting a new standard for hardware resilience in space.

The projection that SpaceX's Starship must complete around 1,800 launches to make orbital data centers economically viable highlights a critical dependency on next-generation launch capabilities. This ambitious target underscores the current limitations in cost-effective access to space and suggests that the widespread adoption of space-based AI infrastructure is contingent upon dramatic advancements in launch cadence and payload capacity.