Computing Companies

EDGX is a Ghent, Belgium-based startup, founded in 2023 by CEO Nick Destrycker and CTO/COO Wouter Benoot, that builds radiation-hardened, AI-capable onboard computers for satellites. Its flagship Sterna computer, built around Nvidia Jetson hardware, and its Morus computer and SpaceFeather software are designed to let satellite operators process imagery and sensor data in orbit and downlink actionable insights rather than raw data, targeting Earth observation, satellite communications and in-orbit servicing customers across commercial, governmental and defence markets. The company says it went from concept to its first in-orbit hardware within about two years of founding. EDGX closed a €2.3 million seed round in August 2025, backed by the Belgian early-stage funds imec.istart and the Flanders Future Tech Fund, alongside institutional backers including VITO, Belspo, the European Defence Fund, Ariane Group, the European Commission and EESA. The company has also reported a €1.1 million multi-unit commercial contract with an undisclosed satellite operator and additional undisclosed government and institutional contracts, and has completed radiation-resilience testing of its hardware. EDGX's computers have already flown: an Early Bird demonstration mission and a Leaving the Nest mission both launched in February 2026 on a SpaceX Falcon 9, and the company describes a further Earth Spotting mission flying in December 2026. Separately, EDGX hardware was carried as part of the SEOPS-integrated Transporter-16 rideshare mission in March 2026, giving the company actual on-orbit flight heritage for its edge-computing products rather than hardware still in development.

Ramon.Space

Tel Aviv, Israel commercial

Ramon.Space is an Israeli space computing company founded in 2004 and headquartered in Tel Aviv, developing radiation-hardened, high-performance computing systems that enable artificial intelligence inference, image processing, signal processing, and data compression onboard satellites and spacecraft — bridging the performance gap between commercial processors and the limited compute of traditional radiation-hardened (rad-hard) space chips. Conventional space-grade processors achieve radiation tolerance through silicon-on-insulator (SOI) semiconductor processes, triple modular redundancy (TMR), and conservative clock speeds — but deliver only a small fraction of contemporary commercial processor performance. This gap has become a critical limitation as satellite Earth observation sensors generate terabytes of raw imagery per day that cannot practically be downlinked to ground without onboard processing and compression, and as AI-driven remote sensing analytics require edge inference capability that traditional space computers cannot support. Ramon.Space's RD1000 and RD2000 processor families use RISC-V based architectures combined with dedicated machine learning accelerator cores, implemented in radiation-hardened process technology and tested to withstand total ionizing dose (TID) environments characteristic of LEO, MEO, GEO, and deep space radiation environments. The chips enable onboard AI applications including change detection, object recognition, automatic target recognition, and hyperspectral data processing without requiring massive downlink bandwidth. The company's computing systems are deployed in commercial and Israeli government spacecraft. Ramon.Space competes with Microchip Technology's RTAX FPGA series, Cobham's LEON processors, and emerging radiation-tolerant solutions from BAE Systems and Xiphos Technologies for onboard computing contracts.

Spiral Blue

Sydney, Australia commercial

Spiral Blue is a Sydney, Australia-based space technology company founded in 2018 that develops edge computing platforms for Earth observation satellites — enabling AI and machine learning algorithms to process satellite imagery directly in orbit rather than transmitting raw data to ground stations, dramatically reducing downlink bandwidth requirements and enabling real-time delivery of actionable insights for time-critical applications. The fundamental problem Spiral Blue addresses is data bottleneck: modern EO sensors can capture multiple gigabytes of raw imagery per orbital pass, but contact time with ground stations is limited to 5-15 minutes per pass, and downlink rates are constrained by atmospheric transmission, regulatory spectrum allocations, and ground antenna aperture. A 50-megapixel camera at 8-bit color generates ~150MB per frame; at multiple frames per second, a single orbit can generate hundreds of gigabytes of raw data that cannot feasibly be downlinked without onboard selection and processing. Spiral Blue's Space Edge One (SE1) computing module runs neural network inference on the satellite itself — performing cloud detection, object classification, change detection, or feature extraction — and downlinks only the compressed results (e.g., coordinates and confidence scores of detected ships, coordinates of wildfire hotspots, field-level biomass indices) rather than full raw imagery frames. The Space Edge One platform uses radiation-screened ARM-based system-on-chip processors (Rockchip-class or similar) with passive thermal management optimized for LEO orbital temperature cycling. It supports TensorFlow Lite and custom model inference frameworks, with a model deployment pipeline allowing satellite operators to upload new algorithms via over-the-air software update — making the platform reprogrammable for different mission requirements after launch. Key application verticals include: maritime domain awareness (detecting vessels, identifying dark ships by cross-referencing AIS absence with optical detections), wildfire early warning (detecting thermal anomalies in near-real-time before fires grow to catastrophic scale), agricultural monitoring (field-level change detection and crop stress indexing), and flood mapping (rapid extent mapping during disaster response when every hour matters). Spiral Blue operates within Australia's growing space ecosystem — supported by the Australian Space Agency, SmartSat CRC (Cooperative Research Centre), CSIRO's data61 analytics programs, and the University of Sydney's aerospace engineering department. The platform has undergone orbital flight validation. Competitors include Unibap (Sweden, iX5-100 space AI module), KP Labs (Poland, onboard processing), and Aethero (US, formerly OrbitsEdge) in the emerging edge-AI-for-satellites market.

TakeMe2Space is an Indian space-technology startup founded in 2024 (following pre-seed development work from 2023) by Ronak Kumar Samantray, founder and CEO, and based in Hyderabad, Telangana. The company is developing in-orbit computing and data-processing infrastructure, aiming to build what it describes as India's first orbital data centre by deploying AI-inferencing hardware on small satellites in low Earth orbit; its OrbitLab platform has offered customers the ability to upload models and run compute jobs on an in-orbit satellite for a usage-based fee. TakeMe2Space has also developed hardware flight heritage: its PowerBank-50 satellite battery pack flew as a payload on Skyroot Aerospace's Vikram-1 orbital launch vehicle mission in July 2026, which the company and Indian trade press described as the first India-made satellite battery to achieve flight-proven ('space-proven') status. The company has separately partnered with QOSMIC to develop an indigenous laser-based optical inter-satellite data-relay network, and with HYLENR to test low-energy nuclear reaction (LENR) powered compute modules in space. TakeMe2Space raised roughly ₹5.5 crore (about $630,000) in a pre-seed round led by Seafund in early 2025, followed by a $5 million seed round in January 2026 led by Chiratae Ventures with participation from Artha Venture Fund, SeaFund and Unicorn India Ventures, intended to expand its satellite constellation to around six spacecraft. An earlier TakeMe2Space satellite mission was lost in a January 2026 ISRO PSLV-C62 launch failure; the company subsequently moved a planned mission, MOI-1a, to a SpaceX Falcon 9 launch. TakeMe2Space is active and continues to expand its in-orbit compute and satellite hardware business.

Exo-Space built FeatherEdge, an edge computing payload that mounts directly onto a satellite and runs machine vision on imagery before it is downlinked. Rather than sending every frame to the ground, the platform detects and classifies objects and structures in the field of view and transmits only the relevant results, cutting downlink cost and shortening the time between an event occurring and an operator learning about it. The platform was designed to integrate with satellites across size classes, addressing a bottleneck that grows with sensor resolution: Earth observation constellations now generate far more data than their radio links can return to the ground. Exo-Space was acquired by Sidus Space in August 2023, which integrated FeatherEdge into its own satellite constellation to provide real-time Earth observation processing on orbit.

Orbital (legally Orbital Compute, Inc.) is a Los Angeles-based startup founded in early 2026 by CEO Euwyn Poon to build a constellation of GPU-equipped satellites in low Earth orbit for running Nvidia AI compute workloads in space. Its pitch is that space offers continuous, low-competition solar power and can use the vacuum as a passive heat sink for cooling, advantages the company argues let it scale AI infrastructure faster and more cheaply than terrestrial data centers once satellites are in mass production. The founding team is drawn from Amazon's Leo satellite-internet program, Northrop Grumman, Lockheed Martin, Firefly Aerospace and Millennium Space Systems. Orbital raised an oversubscribed $5 million pre-seed round backed by a16z's Speedrun program and the NVIDIA Inception program, with additional participation from Basis Set, Human Element, Wayfinder, Anti Fund and Antler. In April 2026 the company signed Reflex Aerospace, the German satellite manufacturer, as the exclusive satellite-platform provider for its first dedicated spacecraft, Orbital-1, targeted for a 2028 launch. Orbital's publicly stated roadmap calls for a smaller Pathfinder hosted-payload demonstration of AI inference compute in low Earth orbit in 2027, flying on a SpaceX Falcon 9, ahead of the multi-GPU-node Orbital-1 spacecraft. As of its FCC filings in mid-2026, Orbital had outlined ambitions to eventually deploy up to 100,000 satellites, a plan that drew scrutiny from astronomers and space-debris researchers over its potential effects on the night sky and orbital congestion. The company had not flown any hardware as of its most recent public disclosures.

OrbitsEdge builds SatFrame, a satellite chassis that supplies the thermal control, power, communications and radiation hardening needed to run ordinary terrestrial server hardware in low Earth orbit. The premise is that adapting the environment around commercial off-the-shelf computers is cheaper than developing space-qualified processors, which typically lag terrestrial parts by generations in performance. The company was founded in 2018 and is based in Cocoa Beach, Florida. It signed an original equipment manufacturer agreement with Hewlett Packard Enterprise to host HPE Edgeline converged edge systems inside SatFrame, giving commercial operators datacentre-grade compute on orbit for processing sensor data before downlink. OrbitsEdge has partnered with Above Orbital to fly its Edge1 radiation-shielded computational platform as a payload on the Prometheus Spark mission, scheduled no earlier than the first quarter of 2026, to demonstrate real-time sensor data processing in orbit.

Satlyt is developing decentralised space-based data centres — moving processing onto satellites in low Earth orbit so that data is reduced or analysed before it is sent down, rather than downlinking everything for processing on the ground. The company frames this as cutting both transmission latency and the power and fuel budget that downlink drives. Its software targets optimisation of satellite operations, allocation of on-board computing resources, and management of data traffic between spacecraft, ground stations and terrestrial data centres. The commercial argument is that downlink bandwidth, not sensor capability, increasingly limits what large Earth observation constellations can deliver. Satlyt was founded in 2024 by Rama Afullo and remains a small team. It has signed an agreement with Star Catcher, which is developing a space-based power grid, and was one of ten companies selected to pitch at Via Satellite's Startup Space 2026 competition.

Aethero

McLean, United States commercial

Aethero, formerly operating as OrbitsEdge, is a McLean, Virginia-based space computing company founded in 2019 that develops radiation-hardened edge computing modules for satellites and spacecraft — enabling onboard AI inference, data processing, and autonomous decision-making directly in orbit rather than transmitting raw data to ground stations for analysis. The company's flagship SkyMicro platform integrates commercial-off-the-shelf (COTS) system-on-chip processors — including NVIDIA Jetson-class GPUs and ARM-based SoCs — within radiation shielding enclosures and power management architectures designed to survive the high-energy particle environments of LEO and MEO orbits. By processing imagery, signals, or sensor data onboard, operators can selectively downlink only actionable results rather than full raw data streams, drastically reducing the contact time and bandwidth required from ground station networks — a critical constraint as constellation sizes grow to hundreds of satellites. Aethero's target applications span defense and intelligence (onboard change detection, target recognition, SAR processing for NRO/Space Force customers), commercial EO (cloud masking, ship/aircraft detection at the edge), and scientific missions requiring low-latency autonomous response. The U.S. Space Force's proliferated LEO architectures and the NRO's commercial imagery programs both create pull for vendors able to deliver space-qualified AI hardware outside traditional defense prime supply chains. The company emerged from the Y Combinator W21 cohort and has secured contracts with DoD customers. Competitors include Unibap (Sweden, iX5-100/iX10 space AI modules), SpaceCubics (Japan, radiation-tolerant FPGA+CPU modules), D-Orbit's ION carrier computing services, and Kubos/Xplore for on-orbit processing software. The trend toward larger constellations and spectrum-constrained downlink budgets makes edge computing an increasingly strategic layer of the satellite value stack.