Session
Constellations
Location
Salt Palace Convention Center, Salt Lake City, UT
Abstract
Over the past two decades, small satellite architectures have undergone a fundamental transformation — from platform-centric spacecraft, to globally distributed data-centric constellations. This paper traces that evolution and argues that a third shift is now underway: cognition-centric space architectures enabled by artificial intelligence (AI) as infrastructure.
We begin by revisiting one of the earliest disruptive space architecture concepts: the fractionated satellite. Developed at the Defense Advanced Research Projects Agency (DARPA) in the mid-2000s, the fractionation concept decomposes a monolithic spacecraft into distributed elements. While the original vision emphasized physical subsystem decomposition, the System F6 program ultimately revealed that the greatest value of distribution lay not in hardware fractionation, but in software-centric service virtualization — particularly communications, compute, and mission applications. Although F6 never launched, it reshaped how a generation of researchers and engineers approached distributed spacecraft systems.
These ideas later scaled to the constellation level through Proliferated Low-Earth Orbit (P-LEO) architectures, first explored under DARPA's Blackjack program and subsequently operationalized by the Space Development Agency. P-LEO systems introduced global, resilient, constellation-level data services driven by physics-based realities — sensor geometry, latency, inverse-square effects, and the speed of light. This shift caused the space data value chain to expand dramatically, enabling persistent sensing, lateral data movement in orbit, and rapid delivery to users.
We argue that the next architectural transition builds directly on this foundation. As AI evolves into persistent, distributed infrastructure — analogous to electricity and the internet — space systems will increasingly host AI-enabled services that provide proximity inference close to where data is generated. In this emerging Space Cognitive Network (SCN), on-orbit compute and storage exist not to replicate terrestrial cloud data centers, but to support low-latency sense-making, multi-sensor fusion, and real-time orientation.
In this model, AI functions as a network of cognitive exoskeletons that augment human understanding rather than simply automate decisions. Space systems move beyond data delivery toward continuous support for human-centered, query-driven workflows across commercial and national security domains. The paper concludes by examining the implications of cognition-centric space architectures for future small satellite constellations, on-orbit services, system design, commercial applications, and the implications for national security.
Document Type
Event
The Space Cognitive Network: From Fractionated Spacecraft to Collaborative AI-Enabled Space Infrastructure
Salt Palace Convention Center, Salt Lake City, UT
Over the past two decades, small satellite architectures have undergone a fundamental transformation — from platform-centric spacecraft, to globally distributed data-centric constellations. This paper traces that evolution and argues that a third shift is now underway: cognition-centric space architectures enabled by artificial intelligence (AI) as infrastructure.
We begin by revisiting one of the earliest disruptive space architecture concepts: the fractionated satellite. Developed at the Defense Advanced Research Projects Agency (DARPA) in the mid-2000s, the fractionation concept decomposes a monolithic spacecraft into distributed elements. While the original vision emphasized physical subsystem decomposition, the System F6 program ultimately revealed that the greatest value of distribution lay not in hardware fractionation, but in software-centric service virtualization — particularly communications, compute, and mission applications. Although F6 never launched, it reshaped how a generation of researchers and engineers approached distributed spacecraft systems.
These ideas later scaled to the constellation level through Proliferated Low-Earth Orbit (P-LEO) architectures, first explored under DARPA's Blackjack program and subsequently operationalized by the Space Development Agency. P-LEO systems introduced global, resilient, constellation-level data services driven by physics-based realities — sensor geometry, latency, inverse-square effects, and the speed of light. This shift caused the space data value chain to expand dramatically, enabling persistent sensing, lateral data movement in orbit, and rapid delivery to users.
We argue that the next architectural transition builds directly on this foundation. As AI evolves into persistent, distributed infrastructure — analogous to electricity and the internet — space systems will increasingly host AI-enabled services that provide proximity inference close to where data is generated. In this emerging Space Cognitive Network (SCN), on-orbit compute and storage exist not to replicate terrestrial cloud data centers, but to support low-latency sense-making, multi-sensor fusion, and real-time orientation.
In this model, AI functions as a network of cognitive exoskeletons that augment human understanding rather than simply automate decisions. Space systems move beyond data delivery toward continuous support for human-centered, query-driven workflows across commercial and national security domains. The paper concludes by examining the implications of cognition-centric space architectures for future small satellite constellations, on-orbit services, system design, commercial applications, and the implications for national security.
