Welcome to Issue 26, where Adaspace lists in Hong Kong to sell computing time from its satellites
This week's developments were about money and coordination for space AI: a stock listing, an ESA industry forum, a workshop report and a fleet-management software launch.
On the Space4AI side, Chengdu-based Adaspace began trading in Hong Kong on November 13 after raising $95.6 million. The company plans to use the listing to fund the engineering needed to sell computing time on its satellites. On November 12 and 13, ESA held a two-day forum on space-based data centres at its technical centre in the Netherlands with industry, operators, investors and research groups.
On the AI4Space side, Seattle-based Constellation Space began rolling out the Constellation Platform on November 12. The software uses machine learning to run satellite fleets and the ground networks that serve them. On November 9, an ESA research lab published five takeaways from its first workshop on agentic AI for Earth observation. According to the report, models and agents built for Earth observation can outperform general-purpose solutions on specialised tasks. The workshop was held in Berlin in October and drew more than 500 attendees.
ESA appears on both sides this week, as host of the data-centre forum and as co-organiser of the Earth observation workshop.
Specifics below.

Did you know?
Bits & Orbits Insights
Public dashboards on in-space compute, space autonomy, and the SpaceX lock-up
calendar:
Space4AI
Adaspace begins trading in Hong Kong and plans to sell in-orbit computing time
Chengdu-based Adaspace began trading in Hong Kong on November 13, 2026, after raising HK$750.01 million ($95.6 million). It sold the shares at HK$24.90, the bottom of its marketed range, and the shares opened at HK$24.44, 1.8 percent below the offer price. The company plans to use the listing to fund the engineering needed to sell computing time on its satellites.
Adaspace was founded in 2018 to put AI on satellites, and it launched its first AI application satellite that December. It launched 12 computing satellites into a single orbital plane in May 2025 and eight more in October 2026. The computing satellites needed more electrical power, more heat dissipation and a flat design that stacks several units for launch. Between the two launches, Adaspace reported that it had deployed Alibaba's Qwen3 model on a satellite already in orbit. At an algorithms competition in Guangzhou, Jiazhihuihang, part of PCI Technology, ran a traffic-analysis model on an Adaspace satellite, which processed imagery in orbit before sending the analysis to the ground.
Satellites and related services made up 57.7% of Adaspace's revenue in the first half of 2026, up from 0.6% in 2023. By October 20, the company had orders for 114 AI satellites not yet delivered, with a combined contract value of RMB3.48 billion ($521 million). Under some contracts, Adaspace books part of the revenue only after a satellite is launched, tested in orbit and handed over.
Under its Star-Compute plan, Adaspace intends to link satellites so they share computing capacity, storage and software, with 100 satellites launched by the end of 2028. The spectrum and orbital resources behind the plan cover 2,800 satellites. Adaspace plans to put about 36% of IPO proceeds into research and development and 24% into a satellite design, development and testing centre in Shenzhen.
ESA holds Strategy Forum with industry on space-based data centres at ESTEC
On 12-13 November 2026, ESA held the Strategy Forum on Space-Based Data Centres at ESTEC in Noordwijk, the Netherlands. The two-day in-person workshop brought together industry, operators, digital-technology providers, research organisations, investors and institutions.
The forum is part of ESA's Strategic Initiative on Space-Based Data Centres (SBDC). ESA set up the initiative to develop a European vision, strategy and roadmap for SBDC, which it has named a new strategic domain. ESA links the effort to rising global demand for data processing and storage. It also links it to European priorities on digital sovereignty, infrastructure resilience, sustainability and competitiveness.
Participants were to assess a set of scenarios for how future space-based data infrastructure ecosystems could operate and which services they could enable. The scenarios drew on a Request for Information that ran on ESA's Open Space Innovation Platform (OSIP) from 12 August to 25 September 2026. RFI contributors got priority if registrations exceeded capacity. The preliminary programme opened day one with a strategic exploration session, followed by two World Café working sessions. On 5 November, ESA Technology said more than 100 selected contributors would attend, chosen from more than 200 applications across the space and digital ecosystems.
AI4Space
Constellation Space launches ML-based platform for managing satellite fleets and ground networks

Illustration of an operations room running the Constellation Platform. Credit: Constellation Space
On November 12, 2026, Seattle-based Constellation Space began rolling out the Constellation Platform, its flagship software for satellite fleets and the ground networks that serve them. The platform uses machine learning to streamline fleet operations and optimize satellite data transfer. It is built for the growing number of constellations in low Earth orbit.
The company's machine-learning model is pre-trained on data from a global network of sensors. After deployment, the model adapts to each fleet's own hardware and ground stations. CEO Kamran Majid said the company builds its own models for weather, attenuation, signal jamming and demand. He said the platform combines those forecasts to show operators their future ground contact plans and how to optimize their fleets. Constellation Space says operators can increase the data downlinked per pass by up to 36%.
Constellation Space was founded last year. Majid previously worked as a NASA engineering intern and as a software engineer on SpaceX's Starlink network. More than 20 academic institutions already use the platform.
ESA Φ-lab publishes takeaways from first Agentic AI for Earth Observation workshop

Participants of the Agentic AI for Earth Observation workshop in Berlin. Credit: BIFOLD/Michael Setzpfandt
On 9 November 2026, ESA Φ-lab published five takeaways from the first Agentic AI for Earth Observation workshop. The workshop ran in Berlin from 19 to 21 October. ESA Φ-lab and the Berlin Institute for the Foundations of Learning and Data (BIFOLD) organised it. More than 500 people attended, including researchers, practitioners, end users and decision-makers from AI, geospatial analytics, climate and weather science, remote sensing and space mission operations.
The report says agentic AI for Earth observation is already in practical use. Working prototypes covered utility and infrastructure monitoring, semantic object discovery without labels, crop and vegetation inspection, satellite image discovery through natural language, and image retrieval and similarity search. These systems combine Earth observation data with geospatial foundation models, large language models, AI agents and specialised tools. According to the report, models and agents built for Earth observation can outperform general-purpose solutions on specialised tasks.
Participants said agents should take decisions only where autonomous action adds value, and humans should keep oversight. The report says adding more agents and tool calls does not necessarily improve results. As systems move to multiple cooperating agents, those agents need to be discoverable, interoperable and reusable. They also need safeguards for identity, data provenance, permissions and accountability. Workflows should be reproducible under the FAIR (Findable, Accessible, Interoperable and Reusable) framework. For Europe, the report also calls for identifying dependencies on external services, tools and agents to protect technological sovereignty. Claudio Iacopino, Head of the ESA Φ-lab Explore Office, said the autonomy of these systems must come with human oversight and scientific validation. Begüm Demir, BIFOLD Deputy Director and Professor at the Technical University of Berlin, said the workshop aimed to build the community's skills in designing efficient and trustworthy Earth observation agents.
The report names the next step as shared frameworks that move agentic Earth observation from proofs-of-concept into real-world operations. That work will require community benchmarks to test whether agents are trustworthy, scientifically sound and ready for use. It will also require closer links between data, models, tools, computing resources, and space- and ground-based assets.
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:
