about 1 month ago
TechCrunch Oct 1, 2026

Google thinks SpaceX’s Starship has to launch 1,800 times before space data centers get off the ground

Google launched its first advanced Tensor Processing Unit (TPU) chip into orbit aboard a SpaceX rocket, marking a significant step in its Project Suncatcher initiative to develop large-scale compute clusters in space. The satellite, built by Planet Labs, will test whether the TPU can operate continuously in the harsh conditions of space, focusing on power supply, cooling, and workload reliability. Future missions aim to deploy more specialized satellites capable of working together via laser communication to handle heavier AI tasks.

The vision behind Suncatcher is ambitious, involving a network of 81 satellites functioning as an orbital data center that can handle complex AI workloads in parallel. Google is taking a long-term approach, preparing for the future needs of AI processing rather than current capabilities. However, expanding space-based data centers depends heavily on reliable, cost-effective launch capabilities that do not yet exist at the required scale, making the project a multi-year moonshot effort.

A key challenge identified by Google is the necessity for SpaceX's Starship rocket to achieve substantial launch frequency and volume to drive costs down sufficiently. Their analysis suggests Starship would need to complete approximately 1,800 launches over the next decade, delivering around 370,000 tons of payload to orbit, to reach the price target of roughly $200 per kilogram by 2035. This projection far exceeds Starship’s current flight rate, highlighting the substantial scale of infrastructure development needed for space data centers.

Despite these hurdles, Google remains optimistic about the resilience of its chips in orbital conditions. The TPU has undergone rigorous radiation testing, confirming a tolerable error rate for inference operations during a satellite’s estimated five-year lifespan. While the chips may be less suited for intensive, prolonged training tasks in space, their performance in real-time AI processing aboard satellites marks a promising advancement toward turning the vision of space-based data centers into reality.

0
0 Read source
Share this post
Facebook Twitter LinkedIn

Discussion

0 comments

No comments yet

Start the discussion with a take, question, or market read.