Welcome to Issue 25, where Orbital selects Orca to run autonomy on its data-center satellites
Satellite software that makes decisions in orbit without waiting for ground commands drew a commercial deal and an in-orbit demonstration this week.
On the Space4AI side, Orca Aerospace announced on November 5 a letter of intent worth up to $20 million to run its onboard autonomy software on the planned data-center satellite fleet of Los Angeles-based Orbital. Payload reported the deal as an exclusive.
On the AI4Space side, Ubotica Technologies reported that onboard AI on an Open Cosmos satellite detected a vessel in the Pacific Ocean in early October. The satellite then scheduled and ran a follow-up observation with no tasking commands from ground stations. Separately, ESA announced that terrAIntel, a tool that answers natural-language questions about Earth observation data, is now fully operational. GeoVille and Cortecs developed it with ESA co-funding.
The Orca deal and the Ubotica trial both concern autonomy software that runs on board the satellite.
Specifics below.

Did you know?
Bits & Orbits Insights
Public dashboards on in-space compute, space autonomy, and the SpaceX lock-up
calendar:
Space4AI
Orbital signs deal worth up to $20M with Orca Aerospace for onboard autonomy software on its planned data-center satellite fleet

Rendering of an Orbital data-center satellite. Credit: Orbital
On November 5, 2026, Orca Aerospace announced a letter of intent worth up to $20 million to run its onboard autonomy software on the planned data-center satellite fleet of Los Angeles-based Orbital. Payload reported the deal as an exclusive.
Under the letter of intent, Orbital will pay Orca an integration fee plus a license fee for each satellite that runs the software. Orca will integrate once with the flight software from Reflex Aerospace, the bus partner Orbital named in Issue 18. The software is slotted to fly on Orbital-1 in 2028, and the companies may test parts of it earlier on Orbital's July 2027 Pathfinder. Orbital plans to grow its fleet from 10 satellites in 2029 to 100 in 2030 and 1,000 in 2031.
Orca's software links each satellite's flight software to Orbital's ground control and handles maneuvers, tasks, and fault responses on its own, calling on operators only when needed. It starts in shadow mode, sending telemetry down for human review, then gains autonomy in layers. A small onboard AI model suggests maneuvers or plan changes, and a rules-based program turns those choices into commands. A safety kernel then checks each command against the mission rules and vehicle limits Orbital sets before the command is sent.
Jacob Ososke, CEO and cofounder of Orca Aerospace, said onboard autonomy lets Orbital handle collision avoidance without a large operator team, and that the software has to run onboard for the fleet to scale safely. Euwyn Poon, founder and CEO of Orbital, compared the software to Tesla's full self-driving mode and said it replaces a team of human operators. The companies hope the deal leads to a longer partnership in which Orbital's data centers provide the compute for Orca's platform, so other spacecraft could join by connecting to Orbital satellites. Orca plans to keep its coordination protocols open to share data with other operators.
AI4Space
Ubotica demonstrates autonomous onboard detection and follow-up tasking of a maritime vessel on CogniSAT-6

Artist's impression of CubeSats in orbit. Credit: Ubotica
On November 4, 2026, Ubotica Technologies announced that it had demonstrated autonomous onboard observation scheduling on an operational satellite in low Earth orbit. During an early-October trial on Open Cosmos' CogniSAT-6, onboard AI detected a vessel in the Pacific Ocean. The satellite then scheduled and executed a follow-up observation with no tasking commands from ground stations.
CogniSAT-6 is a 6U CubeSat designed, built, and operated by Open Cosmos. It launched in March 2024 on SpaceX's Transporter-10 rideshare mission and operates within the OpenConstellation shared infrastructure. It carries Ubotica's CogniSAT-XE2 edge processing platform. In the trial, onboard computer vision scanned an initial image of the Pacific and identified a cruise ship. The flight computer then calculated the spacecraft's next revisit opportunity over the target coordinates to check whether the vessel had stayed in place.
Two days later, CogniSAT-6 imaged the same location with no ground command and confirmed that the ship had left. In conventional Earth observation, raw imagery is downlinked for processing and human analysis, and new tasking is then uplinked. That cycle adds 30 minutes to a full day of latency. The trial advances Ubotica's Live Maritime Intelligence platform, which targets shadow fleets, unflagged vessels, and risks near subsea cables and offshore energy infrastructure. We covered the platform's funding round in Issue 6.
David Rijlaarsdam, Ubotica's Director of Space System Engineering, said autonomous scheduling lets satellites interpret what they see and decide on their next action. He said the capability gains weight across a constellation, where one satellite that spots a vessel can cue another to time its own follow-up image with no ground station or human involved. Ubotica plans to add the scheduling algorithms to multi-satellite constellations for tip-and-cue workflows in civil, naval, and environmental monitoring.
ESA InCubed-funded terrAIntel natural-language Earth observation AI tool becomes fully operational

The terrAIntel user interface. Credit: GeoVille
On 3 November 2026, ESA Φ-lab announced that terrAIntel is now a fully operational platform. The tool answers natural-language questions about Earth observation data. Austrian company GeoVille and Cortecs developed it with co-funding from ESA's InCubed programme. ESA describes it as one of the first ChatGPT-style tools for Earth observation that the agency has co-funded.
Users type questions in their own words about places, events or environmental changes. They also set the location and time period to explore. The platform turns each request into the matching geospatial analysis. It returns maps, visualisations, statistics and data provenance information. Gabriele Meoni, ESA Technical Officer for terrAIntel, said AI automates many of the processing steps that usually require specialist knowledge.
GeoVille first built terrAIntel for data journalists and digital investigative journalists. It now also supports public health officers and emergency response teams. They can map flood-affected areas, track land-cover change, find settlements exposed to hazards and monitor wildfires. Humanitarian teams can use it to locate remote villages and estimate their populations when planning vaccination campaigns. GeoVille Project Manager Donvan Grobler and Cortecs Managing Director Markus Tretzmüller presented the platform at AI WEEK. GeoVille also showed it at ESA Industry Space Days.
José Manuel Delgado Blasco, InCubed Officer for terrAIntel, said it is the first activity of its kind completed with InCubed support. GeoVille plans to add analytical capabilities and datasets and to improve the user experience. It is also doing business development to build a commercially sustainable service.
Till next time,
Meta-beat Column of this week
Read also about the AI Pipeline that sits at the core, producing this Newsletter, including its ups and downs of this week:
