Towards a future space-based, highly scalable AI infrastructure system design
2025/11/22 by Blaise Agüera y Arcas, Arcas, Blaise Agüera y, Travis Roland Beals +16 · 10 voices · 3 citations
Engineering · #Satellite Communication Systems #Spacecraft Dynamics and Control #Spacecraft Design and Technology
paper · pdf · doi:10.48550/arxiv.2511.19468
Abstract
If AI is a foundational general-purpose technology, we should anticipate that demand for AI compute -- and energy -- will continue to grow. The Sun is by far the largest energy source in our solar system, and thus it warrants consideration how future AI infrastructure could most efficiently tap into that power. This work explores a scalable compute system for machine learning in space, using fleets of satellites equipped with solar arrays, inter-satellite links using free-space optics, and Google tensor processing unit (TPU) accelerator chips. To facilitate high-bandwidth, low-latency inter-satellite communication, the satellites would be flown in close proximity. We illustrate the basic approach to formation flight via an 81-satellite cluster of 1 km radius, and describe an approach for using high-precision ML-based models to control large-scale constellations. Trillium TPUs are radiation tested. They survive a total ionizing dose equivalent to a 5 year mission life without permanent failures, and are characterized for bit-flip errors. Launch costs are a critical part of overall system cost; a learning curve analysis suggests launch to low-Earth orbit (LEO) may reach \lesssim$200/kg by the mid-2030s.
Citations
Cited by
Discussions
- 2/2) These are often engineers with years of experience, and they've considered all these issues. For example, heat would need to be dissipated by radiator surfaces. See the preprint from the Suncatch [bsky, 9 points, 0 comments]
- i also had this thought from the first time i heard of these "plans", and then google put out a white paper (on arxiv no less!) and i figured "hmm surely this must address their plans for heat managem [bsky, 8 points, 1 comments]
- "AI appears to be a foundational general-purpose technology — akin to electricity or the steam engine — we should anticipate that its use will continue to broaden across all aspects of human endeavor. [bsky, 8 points, 6 comments]
- As far as i can tell this is the most serious publication on the GPUs in space [bsky, 7 points, 4 comments]
- Only makes sense if launch costs fall precipitously- now, the cost to bring to LEO for just the panels far, far exceeds the value of the energy produced. Google had a stab at the economics here: arxiv [bsky, 4 points, 1 comments]
- Google’s project goes into more detail on the challenges but also refers to heat dissipation as an optimization arxiv.org/abs/2511.19468 [bsky, 1 points, 0 comments]
- The radiation bit, at least, has been tested by Google and at the very least their TPUs (purpose built GPUs) handle much higher ionizing radiation doses than they would experience in 5 years of LEO re [bsky, 1 points, 1 comments]
- It’s interesting how little time google spends on heat dissipation. Basically just “yeah we have to do it” [bsky, 0 points, 1 comments]
- Ah yeah, there it is lol: arxiv.org/abs/2511.19468 [bsky, 0 points, 0 comments]
- #MLSky #DataCenters #GoogkeAI This work explores a scalable compute system for machine learning in space, using fleets of satellites equipped with solar arrays, inter-satellite links using free-space [bsky, 0 points, 0 comments]
Related