Orbital AI Computing: Carbon Tradeoffs Across Satellite Scale
A recent study published on arXiv (2608.14557) enhances the ESpaS framework to analyze the lifecycle carbon intensity of AI computing in Low Earth Orbit (LEO). This research emphasizes the sustainability of AI services in orbit and examines two distinct systems: the lightweight Jetson AGX Orin designed for small satellites and the high-performance DGX H100 supported by large-payload launch vehicles. The findings reveal that emissions from launches serve as a constant carbon overhead, where low-mass systems effectively minimize total emissions, while high-performance systems better distribute this cost, leading to lower carbon intensity. This paper fills a void in earlier models that relied on standard datacenter setups, failing to consider the variations in power, mass, and computing characteristics of contemporary AI hardware, as well as the relationship between launch emissions and system mass. It is classified as a cross-type announcement and can be accessed via the specified arXiv link.
Key facts
- The paper is titled 'Orbital AI Computing: Carbon Tradeoffs Across Satellite Scale'.
- It is available on arXiv with identifier 2608.14557.
- The research extends the ESpaS framework for estimating lifecycle carbon intensity.
- It evaluates two systems: Jetson AGX Orin and DGX H100.
- Launch emissions are identified as a fixed carbon overhead.
- Low-mass systems minimize absolute emissions.
- High-performance systems reduce carbon intensity by amortizing launch costs.
- The study addresses gaps in prior models regarding modern AI hardware and launch emissions.
Entities
Institutions
- arXiv