Welcome to Issue 22, where eight Star Compute satellites reach orbit and Orbitworks reports first light
In-space compute returned to orbit this week as hardware, not slideware: eight orbital data-center test satellites and first light from an AI-native spacecraft both reached orbit.
On the Space4AI side, a Chinese commercial launch carried eight orbital data-center test satellites with liquid cooling and optical crosslinks, the thesis flying at scale. Orbitworks reported first light for the opening satellite of its AI-native constellation, after the October 1 Transporter-18 launch. A seed round for onboard-satellite runtime software rounds out the week, funding the layer that makes that compute usable.
On the AI4Space side, NASA's Perseverance rover took the off-world driving record and credited its onboard navigation for the pace. A reentry-vehicle startup pivoted to spacecraft autonomy outright, winning NASA and Space Force contracts for a system meant to run operations without ground control.
The compute arriving in orbit and the autonomy learning to use it without the ground are two halves of the same shift.
Specifics below.
Did you know?
We regularly update the Bits & Orbits Insights
Check out the growing set of independent dashboards, trackers, and maps on the
AI × Space frontier:
Space4AI
LandSpace Zhuque-2E launches 8 ADAspace 'Star Compute' orbital data center satellites with 10 kW liquid cooling and optical crosslinks
On October 14, 2026, Chinese commercial launch provider LandSpace flew a Zhuque-2E carrier rocket from the Jiuquan Satellite Launch Center, deploying 10 satellites into sun-synchronous orbit. The vehicle lifted off at 7:40 p.m. Beijing Time from Launch Area 96A, placing eight orbital data center test spacecraft built by ADAspace Technology, plus two secondary communications satellites, into their planned trajectories.
The eight ADAspace spacecraft are the second technology validation group for the company's proposed 2,800-satellite "Star Compute" network. Each carries onboard processing clusters engineered for in-orbit AI model training and algorithm execution, cooled by 10,000-watt liquid-cooling loops paired with deployable radiative panels to dissipate computational heat in vacuum. The satellites also carry high-throughput inter-satellite laser transceivers for mesh-network coordination across the cluster, and integrated low-light sensors for autonomous night-imaging analysis.
The mission fits a Chinese push to move computation into orbit, running Earth-observation processing and AI inference natively before transmitting synthesized results to the ground, which addresses terrestrial power constraints and downlink bottlenecks. The Zhuque-2E is a two-stage medium-lift rocket fueled by liquid oxygen and liquid methane, the same methalox lineage LandSpace established with the Zhuque-2, the first orbital methalox launch. The flight also carried LandSpace's Guangchuan-03 and Guangchuan-04 satellites, orbital testbeds for commercial LEO laser communications.
Over the coming months, ADAspace controllers will evaluate the satellites' thermal dissipation under continuous computational load and validate inter-satellite laser routing, ahead of larger operational deployment tranches planned through 2027.
Orbitworks and Loft Orbital report first light, with dual NVIDIA Jetson AGX Orin modules onboard
On October 12, 2026, Orbitworks and Loft Orbital announced that the first satellite of their AI-native constellation is operating on orbit, after launching October 1, 2026 aboard SpaceX's Transporter-18 rideshare. We covered the programme announcement in Issue 17.
The spacecraft carries two NVIDIA Jetson AGX Orin modules to run AI models directly onboard. Onboard georeferencing corrects each image against known reference points on the ground, so every detection leaves the satellite pinned to its precise position on the map. Data stored onboard is encrypted with post-quantum protection, using ML-KEM-1024, the highest security level in the new NIST standard.
Models already loaded include Mistral's Ministral 3, in its 3B and 8B versions, which summarize acquisition data and answer operator queries in orbit, running alongside nearly twenty other models for ship detection, cloud masking, wildfire, burn scar and flood mapping, image super-resolution, and scene description from NASA's Jet Propulsion Laboratory, Google, IBM, Meta, Microsoft, and Hugging Face. For the next generation of satellites, Loft Orbital is evaluating the NVIDIA Space-1 Vera Rubin compute module as well as the NVIDIA IGX Thor for real-time AI reasoning.
Satlyt closes $8M seed to scale onboard-satellite AI runtime and cloud virtualization
On October 13, 2026, software venture Satlyt announced an $8 million seed round to deploy artificial intelligence and edge-computing software directly onto orbiting satellites. The round was led by Houston-based Non Sibi Ventures, with participation from TLcom Capital, Antler, Slauson & Co., Launch Africa Ventures, and Enza Capital. The company was founded by former Google Cloud and SpaceX product manager Rama Afullo and maintains dual headquarters in Sunnyvale, California, and Nairobi, Kenya.
Rather than building dedicated spacecraft, Satlyt runs a hardware-agnostic software layer that executes AI workloads across third-party satellites, converting disparate orbital hardware into coordinated virtual space data centers in a model that mirrors enterprise cloud virtualization. Its runtime executes machine-learning inference at the orbital edge, processing imagery, telemetry, and system logs locally to ease ground-station downlink constraints. In earlier flight validation, Satlyt ran Google DeepMind's Gemma open-weights model aboard a Momentus Vigoride orbital service vehicle, compressing onboard error logs by more than 60% before transmission to the ground.
Satlyt's latest runtime flew on the same SpaceX Falcon 9 rideshare as TakeMe2Space's MOI-1A. We covered MOI-1A's launch on Transporter-18 in Issue 20. The current mission hosts Satlyt workloads including research funded through a NASA Glenn STTR project to test space cloud-computing protocols, edge-vision tracking software for orbital-surveillance firm Stellerian, and application hosting for TakeMe2Space. Over the coming year, Satlyt plans to demonstrate distributed computing tasks synchronized across multiple satellites, targeting integration aboard 20% of active orbital spacecraft by 2030.
AI4Space
NASA Perseverance breaks off-world driving record, credits AutoNav onboard autonomy
On October 11, 2026, NASA's Perseverance rover became the farthest-driven vehicle on another world, pushing its odometer to 28.10 miles (45.22 km) and eclipsing the 28.06-mile (45.16-km) record held by the Opportunity rover, which operated from 2004 to 2018. A 299.6-foot (91.31-meter) drive on the mission's 2,006th Martian sol carried it past the mark. Perseverance covered in five and a half years a distance that took Opportunity nearly 15.
NASA credits the pace to AutoNav, the rover's onboard autonomous navigation system. While rolling, the rover images the terrain, builds a 3D hazard map, and plots its own course between waypoints that rover planners on Earth select. It does this at or near top speed, covering as much as 1,351 feet (412 meters) in a single sol. Perseverance has driven more than 90% of its total distance under AutoNav, compared with about 5% for the first Martian AutoNav system on Opportunity and its twin Spirit; where the earlier Sojourner rover assessed the path ahead using 20 measurement points, Perseverance processes 600,000 with every look.
The record came as Perseverance drove toward an area called "Idubi" through the "Lac de Charmes" region off the western rim of Jezero Crater. In the crater-rim campaign it is examining impact-generated rocks roughly 4 billion years old. Steven Lee, project manager for the Curiosity and Perseverance rovers at NASA's Jet Propulsion Laboratory, said drives like these are now routine, and that the original mission called for no more than 12.4 miles of driving. Mark Maimone, deputy team chief for robotic operations, said newer software teaches the rover to steer less and roll more, and to check its own position against orbital maps mid-drive, with camera glare near the Sun the next obstacle to solve.
NASA approved the mission's first extension in October, carrying exploration of the crater rim through roughly the end of 2028. Having collected 29 rock and regolith samples across the crater floor, delta, and rim, the rover will continue south toward Rainbow Falls, a 328-foot-wide block of ancient crust uplifted by the impact that formed Jezero.
In Orbit Aerospace pivots to spacecraft autonomy, wins government contracts for onboard AI
On October 14, 2026, In Orbit Aerospace disclosed that it has pivoted from building reentry vehicles to spacecraft autonomy software, with NASA and the Space Force funding the shift. The Torrance, California-based startup, founded in 2020 as an in-space logistics company, has won new government contracts for an onboard AI system it says can run a spacecraft's operations without waiting on ground control.
CEO and cofounder Ryan Elliott told Payload the team arrived slightly too late to the crowded reentry market, prompting the shift to a different reentry problem: plasma around a returning vehicle can cut off GPS and comms, leaving ground operators unable to see or react to an anomaly like one side overheating mid-descent. The company's answer is INTELLACT, a software-only system that manages spacecraft health, executes maneuvers, and responds to anomalies onboard using a vehicle's existing hardware, while feeding ground operators context and recommended next steps rather than raw telemetry. CTO and cofounder Ishaan Patel framed the aim as moving humans from in-the-loop control to on-the-loop oversight, with the ground operator as architect of the mission, citing scaling pressure: a satellite can no longer be monitored by one ground operator, or even several.
INTELLACT is slated to debut in Q1 2027 on the AFRL/NASA HyperSTEP mission, running in shadow mode. The latest awards include a NASA Phase I SBIR for spacecraft swarms and a SpaceWERX Phase I STTR focused on LEO constellations, the latter coming through the Space Force's Adaptive and Intelligent Space (AIS) Challenge, which funds early research into satellites that can operate autonomously in degraded or contested environments. In Orbit's proposal was one of roughly 29 selected from more than 300 submissions, and both awards build on a $1.8M Phase II STTR that funded INTELLACT's development for reentry and hypersonic vehicles. The company had raised about $2M in venture funding before the pivot and could raise again within six months; it is also building StratoDrop, an autonomous paraglider that deploys from high-altitude balloons to deliver cargo.
Satellite autonomy is becoming a crowded lane. Shield AI and Sedaro tested Shield AI's Hivemind AI pilot on a NOVI satellite in August, and Y Combinator-backed Orca Aerospace says it is building AI that controls satellites through deterministic tooling.
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: