Compared the Tiansuan profile and accepted mission history with Beijing government cloud-release accounts. Added a separate overlapping service pool and nine scoped relationships. Per-node capacity, power and memory remain unspecified.
Last full profile review: 18 Oct 2026.
Sources collected: 18 Oct 2026.
Source collection and substantive profile review are separate checks.
Open satellite computing and network-security experiments
Launched
Research satellites launched
Next project milestone
Publish current per-node experiment and federated-cloud service availability, verified benchmarks and further phase-two launches
Beijing Haidian Commerce Bureau reports a network of sixteen LEO satellites. This is an attributed program snapshot, not sixteen dedicated compute satellites, an audited live count, or an amount additional to historical launch batches.
More than 40 satellite resourcesBeijing center federated cloud released October 30, 2026Reported specificationattributed government platform accountEvidence & qualifications
Reported coverage is more than 40 satellite resources, including BUPT-2/3 and third-party spacecraft. It is not an independently audited active count, a new launch batch or capacity additional to the 16-satellite Tiansuan snapshot.
University launch account identifies these modules and a small remote-sensing camera. Processor models, memory and computing throughput are not specified. Launch telemetry is not a sustained compute benchmark. The university-authored July 23 launch account on Beijing Daily's author platform confirms the same modules and camera at https://peking.bjd.com.cn/content/s6a622584e4b0e45f3fd517fc.html.
Ground CPU, GPU and NPU resources pooled with satellite resourcesBeijing center federated cloud released October 30, 2026Reported specificationattributed government platform accountEvidence & qualifications
These processor classes describe the ground resource pool. No installed orbital processor model, accelerator count, throughput or arithmetic precision is specified.
University-authored BUPT-4 launch account and supplier solar-wing mission account do not give installed processor ratings, RAM, storage capacity or IT-only electrical power. The separately reported token-per-joule claim lacks an identified configuration and benchmark.
StarVault is scheduled for April 2027. Coral Gables plans a 12-month pilot for recovery data, security keys and historical archives. Published storage-capacity ranges lack a defined spacecraft or network boundary. Freedom’s historical 8 TB payload remains separate.
Last full profile review: 19 Oct 2026.
Sources collected: 19 Oct 2026.
Source collection and substantive profile review are separate checks.
Lunar and orbital storage and processing
Demonstrated
Demonstration program
Next project milestone
First StarVault payload is planned aboard LizzieSat-4 on SpaceX Bandwagon-6, alongside LizzieSat-5, no earlier than April 2027. Timing depends on qualification, integration and launch readiness.
Lunar & orbital storageDetailed configuration not available in reviewed material.
Microchip PolarFire SoC / RISC-V for historical Freedom payloadHistorical Freedom payload flown on IM-2Recorded descriptionExisting catalog prose, not a new measurementEvidence & qualifications
Historical Freedom hardware only. IM-2 ended in March 2025. This is not current StarVault equipment or available service capacity.
Compared November agreement with the October pair and earlier OL-1 history. Planned20 includes the initial pair. Strategic cooperation relationship included in this review. No launch, installed capacity or financing amount established
No completed full profile review is recorded.
Sources collected: Date unavailable.
Source collection and substantive profile review are separate checks.
Mission sensing, autonomy and data fusion
Development
Development agreement
Next project milestone
Launch of the initial Oman satellite pair and implementation of the planned local infrastructure, no date disclosed
Stage detail: Operator-reported onboard AI inference
Compared Google's engineering, launch and inference accounts with project, metrics and three relationships. Added a distinct future clustered-design TPU observation and UC Davis ground-testing role. Corrected the launch caveat and retired stale catalog hardware prose. Prototype chip count, absolute power and new conference cost claims remain unverified
Last full profile review: 20 Oct 2026.
Sources collected: 20 Oct 2026.
Source collection and substantive profile review are separate checks.
TPU hardware and optical interconnect validation
Demonstrated
Operator-reported onboard AI inference
Next project milestone
Continue in-orbit TPU thermal-stress and radiation tests, followed by the separate twin-satellite interconnect experiment planned for 2027.
One prototype launched October 1, 2026. Google's October30 report establishes an operator-reported onboard Gemma4 workload response, not a commercial fleet or measured throughput.
Google identifies Trillium TPUs in the prototype test context. The official pages do not disclose chip count, installed throughput or numerical precision.
Dozens of TPUs per satellite in future clustered designsFuture clustered satellite designs, distinct from the 2026 prototype and the separate 2027 pairReported specificationplanned official architectureEvidence & qualifications
September 24 engineering account describes future designs. No exact chip count, model, arithmetic throughput, memory or electrical load is specified for this configuration. This is not the installed count on the October 1 prototype or a demonstrated cluster
Stage detail: First payload operational data reported
Two STERNA payloads launched in March 2026, with operational data reported from one. Separate 2027 missions are booked. Product test references include thermal-vacuum, radiation, vibration and shock conditions. Official pages give conflicting five- and seven-year product-life targets.
No completed full profile review is recorded.
Sources collected: Date unavailable.
Source collection and substantive profile review are separate checks.
Onboard AI and third-party software hosting
Launched
First payload operational data reported
Next project milestone
Out of Tern is planned for February 2027 on Gracious Gloria, followed by Wingman on D-Orbit ION in October 2027.
Two March 2026 hosted demonstrationsMarch 2026 STERNA demonstrationsRecorded descriptionExisting catalog prose, not a new measurementEvidence & qualifications
One payload returned operational data within two weeks. The second still awaits satellite commissioning. Later 2027 bookings are separate.
STERNA with NVIDIA Jetson Orin NX in Super ModeOut of Tern planned for February 2027Reported specificationofficial project specificationEvidence & qualifications
The operator explicitly maps this hardware and mode to the booked Gracious Gloria demonstration. It does not establish the March 2026 payload mode, numerical performance or sustained flight power. The October28 release describes over50percent more AI performance in Super Mode. Relative design claim only, with workload and arithmetic format unspecified.
Hyperspectral camera, STERNA, Morus based on NVIDIA Jetson Thor, and relay linkWingman planned for October 2027Reported specificationofficial project specificationEvidence & qualifications
The booked D-Orbit ION mission is described as the first complete EDGX payload and first Morus flight. Component models, count, workload throughput and mission power are not disclosed here. Product-wide Morus ratings are not assigned to this mission. The October 28 release identifies Morus as based on Jetson Thor. This is planned hardware, not flown performance.
STERNA built on NVIDIA JetsonMarch 2026 STERNA demonstrationsReported specificationofficial reported specificationEvidence & qualifications
Exact Jetson variant, installed count and achieved AI throughput are not disclosed for the March payloads. Do not assign planned 2027 Orin NX Super Mode to these flights. The Sterna platform is designed to adapt processing to power and thermal conditions, without a disclosed flight operating envelope.
Starcloud →Starcloud · United StatesLatest development 28 Oct 2026 ↗Manual review with evidence gaps 3 Nov 2026
Stage detail: Demonstration program
Compared operator mission history and commercial roadmap with NVIDIA, Quindar, Crusoe, Firefly and the investor portfolio. Starcloud-1 demonstration remains distinct from planned follow-ons. Added qualified technical references. Conflicting follow-on timing, exact mission mapping and power boundaries remain unresolved
Last full profile review: 20 Oct 2026.
Sources collected: 20 Oct 2026.
Source collection and substantive profile review are separate checks.
Orbital AI experiments and planned cloud services
Demonstrated
Demonstration program
Next project milestone
Starcloud-2 commercial follow-on: late-2026/October launch targets in partner and executive announcements, with limited customer access and full operation targeted by 2027.
Operator's current undated product page describes the planned commercial mission. This does not establish installed processor models, GPU counts or arithmetic throughput and is separate from Starcloud-1's H100.
The reviewed material does not establish a current project-specific IT load. Proposed later spacecraft figures are retained separately and are not assigned to Starcloud-1.
Technical specifications retain the source’s original units and evidence status. Hardware counts, compute rates and power are not equivalent; launched hardware does not establish demonstrated performance. Supplier counts include suppliers and hosts only, reflect current filters and cover this selection, not the entire market. “No record” means no disclosure captured; it does not mean the work is unawarded. Latest developments use source dates, not routine review or collection dates.
What changed, the work involved, and what the announcement establishes.
6 recent developments
Project relationships
Beijing platform connects computing on satellites with ground servers
The Beijing Space Computing Innovation Center unveiled a cloud platform on October 30 for managing computing tasks across satellites and ground-based servers. Beijing's Science and Technology Commission says it covers computing resources on more than 40 satellites. The service is designed to let users upload AI models, run them in orbit and receive the results through one interface. It also connects to Baidu Intelligent Cloud's model and dataset services. The satellite count includes existing spacecraft from several operators, including BUPT-2 and BUPT-3.
What it means: A shared service could make it easier for developers to use satellite computers without building separate connections to each operator.
Work: Baige platform model and dataset interfaces interconnected with the Beijing space–ground cloud · Named initial participating company in the federated space–ground cloud · Initial participating company, with Fuyao in-orbit resources intended to support platform construction · Partnership
Coral Gables joins Lonestar for a space-storage pilot
Lonestar and the City of Coral Gables plan a 12-month pilot to add space-based storage to the city’s backup systems. The project covers disaster-recovery information, digital security keys and a heritage archive. The teams are preparing selected city data for Lonestar’s first StarVault launch, scheduled for April 2027.
What it means: The pilot will test whether a separate storage layer in space can help the city restore services after cyberattacks, hurricanes or infrastructure outages.
Work: A 12-month founding-city pilot using StarSafe and StarKey for disaster-recovery information, security keys and a Coral Gables Digital Heritage Archive. · Partnership
STAR.VISION and Oman Lens plan a first batch of 20 AI satellites
STAR.VISION and Oman Lens announced an expanded space-AI agreement covering a planned first batch of 20 satellites, a local satellite assembly, integration and testing center, and joint AI research in Oman. The previously introduced Oman satellite pair is identified as the first two forthcoming spacecraft within this batch. Earlier OL-1 flight heritage remains separate.
What it means: The agreement expands the partners’ satellite roadmap and local engineering capability. The first pair is intended to use model-driven spacecraft control, coordinated observation and onboard data fusion. Launch dates and agreement value were not disclosed.
Work: Expanded space-AI cooperation for a planned first batch of20 Oman Lens satellites, local satellite AIT infrastructure and joint AI research · Development agreement
Beijing center and CAICT announce support platform for space computing
China Academy of Information and Communications Technology (中国信息通信研究院): The Beijing Space Computing Innovation Center and China Academy of Information and Communications Technology announced a joint platform to develop standards, test technology, support new applications and connect companies with potential partners.
What it means: The platform aims to help space-computing developers test their ideas and turn research into practical applications.
Work: The Beijing Space Computing Innovation Center and China Academy of Information and Communications Technology announced a joint platform to develop standards, test technology, support new applications and connect companies with potential partners. · Partnership
Google reports first Gemma 4 response from its orbital TPU prototype
Google reported on October 30 that its Suncatcher prototype ran the Gemma 4 open model onboard a data-center-class TPU and returned its first AI response during ground-station passes including Svalbard. Follow-up prompts covered text generation and payload telemetry. Thermal-stress and radiation-reliability tests continue, ahead of a separate two-satellite interconnect experiment planned for 2027.
What it means: The result advances Suncatcher from launch and commissioning to an operator-reported AI workload demonstration. Numerical throughput, energy use and long-duration reliability are not disclosed.
EDGX reports first orbital data and plans Jetson Thor payload
EDGX reports that one of its two STERNA payloads launched in March 2026 returned operational data within two weeks. The other awaits satellite commissioning. The company has booked two missions for 2027: one using NVIDIA Jetson Orin NX Super Mode and another carrying its Jetson Thor-based Morus computer.
What it means: The flights give EDGX experience operating its computers in space as it develops a service that would let customers upload software and rent computing time in orbit.
