Google and Starcloud Set to Revolutionize AI Data Centers with Space-Based Approach
The race to pioneer AI data centers in space has reached a significant milestone. Google has announced its “Project Suncatcher” initiative, while Starcloud gears up for an imminent satellite launch.
Google’s ambitious plan involves launching two prototype satellites by early 2027. On the other hand, Starcloud is set to make history by deploying the first NVIDIA H100 GPU into orbit this November. This marks a significant leap in GPU compute power in space, being “100x more powerful than any before.”
The space-based approach to AI data centers addresses the escalating energy demands of AI workloads. By harnessing constant solar power and utilizing the vacuum of space for cooling, energy costs could potentially be reduced by 10x compared to Earth-based facilities. Google’s analysis suggests that as launch costs are expected to fall below $200 per kilogram by the mid-2030s, space data centers could become economically comparable to terrestrial ones.
Philip Johnston, CEO of Starcloud, made a bold prediction. He stated, “In 10 years, nearly all new data centers will be built in outer space.” The 60-kilogram Starcloud-1 satellite will test real-world AI inference capabilities, particularly for Earth observation data processing and critical applications like wildfire detection.
Source: https://research.google/blog/exploring-a-space-based-scalable-ai-infrastructure-system-design/