Edge AI on Satellites: From D-Orbit's AIX Constellation to STAR.VISION's 1000 TOPS Platform

Edge computing has moved from terrestrial data centers to orbit. Multiple satellite operators now run AI inference directly on spacecraft, processing data in space rather than transmitting raw sensor feeds to ground stations for analysis.

D-Orbit’s AI-eXpress (AIX) constellation and STAR.VISION’s computing platforms represent the operational state of orbital edge AI. These are not laboratory demonstrations or future proposals. The hardware is in space, processing real data.

D-Orbit AI-eXpress Constellation

D-Orbit’s AIX constellation consists of three satellites launched in 2025, with the final unit (AIX-1+) reaching orbit on November 28. The constellation was developed in partnership with Planetek Italia, which D-Orbit fully acquired in 2025, integrating Planetek’s expertise in cloud-based space applications and AI-powered data processing.

The AIX satellites perform autonomous decision-making in orbit. Instead of downlinking every image to ground stations for analysis, the satellites carry machine learning models that identify specific features (ships, vehicles, changes in vegetation, infrastructure development) and transmit only relevant processed data.

This approach addresses bandwidth constraints. A typical Earth observation satellite generates terabytes of imagery per day. Downlinking all of this data requires extensive ground station networks and wide communication windows. By processing data on-orbit and transmitting only analysis results, AIX reduces bandwidth requirements by orders of magnitude.

The constellation utilizes blockchain technology for data integrity verification. Processed results are cryptographically signed on-orbit, creating tamper-evident records of satellite observations before transmission to ground users.

Technical Architecture

The AIX satellites integrate:

  • Imaging sensors (optical and potentially SAR on future variants)
  • AI accelerator hardware for neural network inference
  • Edge computing processors running machine learning models
  • Distributed operating system enabling satellites to coordinate tasks
  • Optical and RF communication systems for ground and inter-satellite links

The distributed operating system allows the constellation to function as a coordinated network rather than independent units. Satellites can delegate tasks based on current position, sensor availability, and computational load.

STAR.VISION AI Computing Platforms

STAR.VISION manufactures AI computing units designed specifically for satellite integration. Their “STRING” platform delivers over 300 TOPS (tera-operations per second) of INT8 computing power, while the “Light String” variant provides 150 TOPS in a smaller form factor.

These units function as the satellite’s “brain,” enabling real-time data processing in space. Applications include:

  • Forest fire detection and tracking
  • Flood risk monitoring
  • Urban development assessment
  • Typhoon path prediction
  • Agricultural land use analysis
  • Emergency response coordination

STAR.VISION’s platform integrates AI models from Zhejiang Lab, creating a space computing network where interconnected satellites can dynamically handle various tasks. The company offers a “Spaceborne AI Algorithm Rideshare Program,” providing other space technology companies on-demand access to satellite AI computing capabilities.

Operational Deployments

STAR.VISION has delivered hardware to multiple customers:

  • Oman’s first artificial satellite, equipped with onboard AI processing
  • Rwanda Space Agency, developing AI algorithms for real-time land usage assessment tested on the WJ-1A satellite
  • G60 Star Chain, providing intelligent power controllers for China’s satellite internet constellation

The platforms utilize AI and 3D analysis to create a “digital Earth” ecosystem, providing end-to-end solutions from data acquisition to application deployment.

D-Orbit SkyServe STORM Platform

Beyond the AIX constellation, D-Orbit operates the SkyServe STORM platform aboard its ION Satellite Carriers. This system enables edge computing through software-defined satellite infrastructure.

STORM (Space-based Tactical Operations and Resource Management) allows third-party developers to deploy geospatial applications directly on D-Orbit’s satellites. Organizations can upload AI models, run them against live Earth observation data feeds, and receive processed results without building their own satellite hardware.

This “infrastructure as a service” model for orbital edge computing parallels terrestrial cloud platforms (AWS, Google Cloud, Azure) but with the satellite acting as the compute node. Geospatial analytics companies can deploy machine learning models for specific applications (maritime monitoring, agricultural assessment, disaster response) without investing in dedicated satellite systems.

The more performant ION variant expected in 2025 will expand available computing resources, enabling more complex AI workloads in orbit.

Technology Readiness Level

Orbital edge AI is at TRL 5-7 (operational demonstrations in relevant environment). Unlike speculative orbital computing proposals, these systems are flying hardware processing real data:

  • D-Orbit AIX constellation: TRL 7 (operational system)
  • STAR.VISION computing units: TRL 6-7 (operational on multiple satellites)
  • SkyServe STORM: TRL 6 (demonstrated in operational environment)

The technology has moved beyond laboratory validation. The engineering challenges now focus on scaling, reliability over multi-year missions, and economic viability rather than fundamental technical feasibility.

Applications and Use Cases

Earth Observation

Satellites equipped with edge AI can identify specific features in imagery without human intervention:

  • Ship detection for maritime domain awareness
  • Building footprint extraction for urban planning
  • Crop health assessment for agricultural monitoring
  • Infrastructure damage assessment for disaster response

Processing occurs on-orbit, reducing latency between image capture and actionable intelligence delivery to end users.

Autonomous Operations

Edge computing enables satellites to make operational decisions without ground control:

  • Retasking sensors based on detected events (forest fire spotted, adjust imaging schedule)
  • Optimizing downlink priorities (transmit high-priority detections first)
  • Power management (adjust processing load based on battery state)
  • Collision avoidance (process tracking data and adjust orbit autonomously)

This autonomy becomes critical for large constellations where ground operators cannot manually manage thousands of satellites.

Bandwidth Reduction

Transmitting raw sensor data from orbit consumes limited RF spectrum. Edge computing reduces bandwidth requirements:

  • A 12-bit multispectral image might be 500 MB
  • Processed feature detection results might be 5 KB
  • Bandwidth reduction: 99.999%

This allows satellite operators to extract more value from existing communication infrastructure without expanding ground station networks.

Comparison to Orbital Data Center Proposals

Edge AI satellites differ from larger orbital data center concepts (Google Suncatcher, SpaceX FCC filing, ESA ASCEND):

Edge AI (Current Reality):

  • Smallsat-scale computing (1-100 TOPS per satellite)
  • Inference-only workloads (running pre-trained models)
  • Processing satellite’s own sensor data
  • Power budgets: 100-500 watts
  • Operational today (TRL 6-7)

Orbital Data Centers (Future Proposals):

  • Constellation-scale computing (1,000+ TOPS distributed across satellites)
  • Training workloads (updating model weights based on new data)
  • Processing arbitrary computational workloads, not just satellite data
  • Power budgets: 1-10+ kilowatts per satellite
  • Mostly at TRL 3-4 (concept/prototype phase)

Edge AI satellites prove that space-based computing is viable for specific applications. Whether this scales to general-purpose orbital data centers remains unproven.

Economics

Orbital edge AI makes economic sense when:

  • Bandwidth savings exceed the cost of additional satellite hardware
  • Processing latency matters (real-time analysis vs. hours-later ground processing)
  • Data sovereignty requires on-orbit processing rather than ground station downlinks in foreign countries
  • Application requires continuous monitoring that would overwhelm ground station capacity

Current implementations target niche applications where these conditions hold. Broader adoption depends on continued reductions in satellite computing hardware costs and improvements in power efficiency.

Path Forward

The AIX constellation and STAR.VISION platforms demonstrate operational viability for orbital edge computing. Future developments will likely focus on:

  • Increased computing power per satellite (1,000+ TOPS per unit)
  • On-orbit model training (not just inference)
  • Inter-satellite data sharing for distributed workloads
  • Standardized APIs for third-party AI model deployment
  • Integration with 5G/6G networks for direct satellite-to-device services

These systems validate the core ArkSpace thesis: computing is moving to orbit. The question is no longer whether space-based computing is possible, but how rapidly it scales and which applications migrate to orbital infrastructure first.

Official Sources