Link to the code: Zae-Project / arkspace-core

Software-Defined Satellites: How Cloud Architecture Is Transforming the Orbital Stack

The satellite industry spent decades building hardware that could only do one thing. A communications satellite launched in 2010 was designed, frequency-allocated, and beam-pointed before it left the factory. Once in orbit, the coverage map was fixed. Operators paid for capacity that did not match where demand actually appeared.

Software-defined satellites break that constraint. The same physical transponder can serve a different region, different frequency band, or different customer mix without touching the spacecraft. That flexibility is driving a forecast that would have seemed implausible five years ago: ABI Research projects software-defined satellite deployments will grow from 234 in 2024 to over 10,000 by 2031, moving from 3 percent to 26 percent of all active satellites in orbit.

What Software-Defined Actually Means

The term gets applied loosely. At minimum, a software-defined satellite has a digital payload processor that can be reconfigured via software upload from the ground. The meaningful capabilities include:

Electronically steerable beams. Rather than fixed reflector geometry, an active phased array or digital beamforming system shapes and points beams on demand. SES’s O3b mPOWER system generates over 5,000 fully steerable spot beams per satellite. Each beam can be shifted and scaled in real time to serve specific users or regions. That is not a marketing claim; it is the difference between statically allocating capacity to an ocean shipping route and dynamically following ships as they move.

Frequency agility. A reconfigurable payload can shift operating frequencies under software control. Eutelsat Quantum, launched in 2021, is the first fully reconfigurable commercial satellite: coverage maps, frequency bands, power levels, and orbital position can all be changed post-launch via software command. A satellite like this is more like a VM than a dedicated server.

Dynamic power allocation. Fixed transponders divide transmit power statically across beams. Digital payload systems can reallocate power to where link margin demands it, accommodating rain fade events or demand spikes without physical hardware changes.

Cloud-native ground processing. Software-defined satellites are paired with ground systems that expose the same abstraction layers as cloud infrastructure. ABI Research notes that cloud-supported satellites will grow from 736 in 2025 to approximately 11,000 by 2031, a parallel curve to the software-defined fleet itself.

The 234 to 10,000 Trajectory

The growth projection reflects three converging forces.

3GPP NTN standardization. The 3rd Generation Partnership Project’s non-terrestrial network standards (NR-NTN in Release 17, enhanced in Release 18 and 19) define how mobile devices connect directly to satellites using standard cellular protocols. A satellite that can support NTN must implement the 3GPP air interface, which requires a flexible software-defined payload capable of emulating a terrestrial base station. Every telco-integrated LEO satellite deployed for direct-to-device connectivity is effectively a software-defined satellite by definition.

LEO constellation scale. Starlink, Kuiper, OneWeb, and Logos Space are all deploying LEO constellations at mass-production volumes. At that scale, the economics of reconfigurable hardware improve substantially over fixed-function designs. A software-defined satellite that can be repurposed for different markets over its operational life is worth more than one locked to a single application from launch.

Vertical integration in satellite manufacturing. SpaceX manufactures Starlink satellites internally, enabling rapid iteration of the payload software stack. The same is true of Amazon’s Project Kuiper and, to a lesser extent, OneWeb. This supply chain structure makes it practical to develop and update software-defined features iteratively, rather than locking specifications three years before launch.

ABI Research estimates the addressable market for applications built on this infrastructure, including connected vehicles, IoT, and direct-to-cellular services, at US$22 billion by 2032.

Operational Examples in 2026

SES O3b mPOWER is the most technically advanced software-defined constellation currently in service. Boeing’s integrated payload array technology supports Adaptive Resource Control (ARC) software that synchronizes space and ground resources, routing traffic, allocating power, and steering beams in real time. The March 2026 activation of the latest O3b mPOWER satellite pair brings the constellation to full initial operating capacity.

Eutelsat Quantum operates in GEO and demonstrates the extreme end of post-launch flexibility. Eutelsat can redirect the satellite to cover a different region, respond to regulatory changes in frequency allocation, or repurpose capacity from a declining market to a growing one. It is the closest existing analog to a software-defined orbital computing node.

The edge AI systems deployed by D-Orbit and STAR.VISION represent a further step: not just reconfigurable communications payloads, but satellites with on-board AI inference stacks that can process Earth observation data without downlinking raw imagery. Software-defined communications and software-defined compute are converging on the same platform.

Cloudification: The Satellite as a VM

The deeper architectural shift is not just that satellites can be reconfigured. It is that the operational model is beginning to mirror cloud infrastructure.

In a cloud data center, compute and network resources are abstracted from physical hardware. Workloads run as virtual machines or containers, allocated to physical servers based on demand. Ground network engineers manage global routing tables updated in milliseconds. The same abstraction layer logic is appearing in satellite ground systems.

What this enables for orbital computing proposals like Starcloud and Google Project Suncatcher is a path to operating satellite-hosted compute as a service rather than as dedicated hardware. A software-defined satellite carrying an AI accelerator could, in principle, be tasked with different inference workloads on different orbital passes, allocated to different customers, and billed accordingly. That is the same model that made cloud computing economically dominant on the ground.

LLM-based satellite management systems at Thales Alenia Space and ESA are already using AI to manage component data and configuration. That capability feeds directly into the software-defined paradigm: if the ground-side management layer uses AI to optimize configurations, the satellite-side payload needs the flexibility to accept those configurations.

NTN Integration: Telco Satellites Are Software-Defined by Design

The most direct near-term driver for the 10,000 software-defined satellite forecast is not the orbital computing market. It is the telco integration market.

Direct-to-cellular services from Starlink (partnered with T-Mobile), AST SpaceMobile, and others require satellites to implement 3GPP NR protocols directly. These protocols were designed for flexible spectrum use, including dynamic bandwidth allocation and multi-user MIMO techniques that only make sense in a software-defined payload context.

Europe’s ASCEND orbital data center program explicitly considers software-defined infrastructure as a design requirement. A satellite that can switch between providing broadband connectivity and providing compute resources for edge processing represents a dual-use case that improves mission economics.

The autonomy capabilities described in Autonomy Ascending: How Spacecraft Are Learning to Fly Themselves are also relevant to software-defined operations. A satellite capable of autonomous maneuver planning and anomaly response is a satellite running a significant on-board software stack. Managing that stack remotely, updating it safely, and verifying its behavior is the software-defined operations problem applied to flight software rather than communications payloads.

Path Forward

The 3 percent to 26 percent shift forecast by ABI Research has a specific meaning: by 2031, one in four active satellites in orbit will be reprogrammable in meaningful ways. The other three will not. That remaining 74 percent represents billions of dollars in stranded capability, locked into frequency plans and coverage geometries that may no longer match where demand is.

The economic pressure this creates is a forcing function. Operators who replace aging fixed-function satellites with software-defined successors gain optionality. Those who do not find their capacity increasingly misaligned with market demand.

For orbital computing, the implication is more fundamental. The same reconfigurability that lets a communications satellite serve a different region is what would let an orbital compute node serve a different customer, run a different AI model, or participate in a distributed inference job for a ground-based system. The infrastructure is converging, even if the applications are not there yet.

Official Sources