Weekly Brief.

One new project was announced. Cowboy Space unveiled Big Iron, a rocket whose upper stage is intended to remain in orbit as a megawatt-class AI data center after launch; its first flight is targeted for 2028.

Three supplier engagements were disclosed. SEALSQ signed a letter of intent to acquire WISeSat capacity and operations for its proposed security cloud; closing still depends on definitive agreements. EDGX identified NVIDIA Jetson hardware in its STERNA computers, and Starcloud partnered with Quindar to deploy mission-management software on future orbital infrastructure.

Google said on October 30 that its prototype powered up its Tensor Processing Unit (TPU), ran early machine-learning tests and returned a response generated by the Gemma 4 model during a ground-station pass. The output itself was simple; the useful result is that a data-center-class accelerator, model, spacecraft power system and communications path worked together in orbit.

This advances last week’s watch point: Suncatcher now has a completed workload, not just a healthy spacecraft. Google’s next tests will examine thermal stress and radiation exposure, the conditions that determine whether this can become reliable infrastructure rather than a one-off demonstration.

The week’s other announcements sketch the systems around the processor. EDGX reported that one of two STERNA payloads launched in March returned operational data within two weeks, while the second still awaits host commissioning. Quindar’s planned role for Starcloud focuses on telemetry, scheduling and autonomous spacecraft operations. SEALSQ’s letter of intent would add capacity access, ground systems and mission control to its proposed service.

Those layers matter because useful orbital computing depends on more than chips. Power, heat rejection, communications and fleet operations have to function as one service. The evidence is uneven: EDGX’s result is retrospective and does not include a published AI benchmark; Quindar has not yet been deployed on Starcloud; and the SEALSQ transaction is not closed.

Suncatcher’s returned answer is modest but observable. It tells readers more about near-term capability than Big Iron’s far larger proposed architecture, because it closes a real loop from uplink to onboard computation to downlink. The market is separating into two paths: small systems processing data where it is collected, and much larger linked data centers. This week strengthened the first path; the second remains mostly design, partnership and schedule.

Watch for Suncatcher’s thermal and radiation results, then its planned 2027 two-satellite optical-link test; those would show whether compute can remain dependable and work across spacecraft. Also watch for a published STERNA workload result, an on-orbit Quindar deployment, definitive SEALSQ–WISeSat agreements and Cowboy’s promised engine and vehicle tests. Each would convert a different piece of the surrounding stack from intent into measured performance or binding commitment.

Two new projects were announced. HPE unveiled Spaceborne Computer-4, a fanless CPU/GPU module intended to run AI and high-performance-computing tests from Astrolab’s FLIP rover on a late-2026 lunar mission. NOAA and Cowboy Space disclosed Reason-2, a 2027 spacecraft that would run NOAA processing software on NVIDIA GPUs and return selected results by optical link.

Five newly disclosed engagements were verified: the NOAA–Cowboy and Sophia Space–MIT Lincoln Laboratory research agreements; the Redwire–EnduroSat memorandum; and Astrolab’s hosting and Deloitte’s workload participation for HPE. These are two CRADAs, one preliminary agreement and two project relationships—not five procurement awards. HPE described KIOXIA’s storage role as continuing, so it is not counted as new this week.

Spaceborne Computer-4 is more concrete than a generic lunar-computing proposal. HPE names LINPACK and HPCG benchmarks, LiDAR processing, a NASA federated-learning experiment and Deloitte cyber analytics. The computer was integrated with FLIP in February, but launch, landing and lunar operation remain ahead. HPE does not disclose compute capacity, commercial terms or a completed test.

NOAA’s two-year CRADA requires Cowboy to fund, build, launch and operate Reason-2, while NOAA supplies historical sensor data and an open model; test output will not enter operational forecasting. Sophia’s CRADA will evaluate weather workloads in terrestrial and potential space configurations. Redwire and EnduroSat, by contrast, only intend to pursue opportunities that include orbital data centers. None of the three announcements establishes paid orbital-compute demand.

Planet said on October 20 that the Project Suncatcher prototype launched October 1 is healthy and progressing through calibration. That is useful spacecraft-status evidence, not a TPU result. Planet reported no completed AI workload, thermal measurement or radiation finding.

SignalSpore now publishes 73 project profiles and 472 supplier records, including 61 contract awards. Five profiles were added this week, but Loft’s YAM platform, Nebula Nimbus and Korea’s K-Moonshot are historical backfill. Reason-2 and Spaceborne Computer-4 are the two current announcements. Most of the 26 additional supplier records also fill older gaps; they are not 26 new weekly deals.

Reflex Aerospace said on October 19 that a planned NVIDIA minority investment would support Bavarian production and onboard-processing development. The statement confirms intent, not completion. It gives no amount, closing date or allocation to a named orbital-data-center project. A closing notice, disclosed financing terms or a project-specific award would materially strengthen the signal.

Weather processing, lunar autonomy and cyber analytics are clearer jobs than undifferentiated “compute in space.” This week’s evidence is strongest on technical scope and weakest on procurement, capacity and results. Watch for the HPE mission’s launch and lunar benchmarks, Reason-2 hardware and schedule, funded work under the Redwire–EnduroSat MOU, Suncatcher TPU results and evidence that NVIDIA’s proposed Reflex investment has closed.

LandSpace reported a successful October 14 launch of ten satellites. Eight ADA Space Star Era spacecraft will test high-performance computing payloads, high-speed laser links and large radiative-cooling wings. Two Lanta Intelligence Guangchuan satellites will test onboard processing, laser communications, integrated routing and new solar-wing systems.

That combination matters because computing, networking, power and heat rejection have to work together. The launch establishes delivery to orbit, not successful technology validation. Public evidence does not yet tie these spacecraft to a verified operating constellation or measured computing capacity.

Unibap said on October 13 that it signed its first contract to supply satellite edge-computing platforms for a space-based signals-intelligence mission. The customer is an unnamed European space-defence company, the end user is confidential and the initial delivery covers only a handful of computers.

This is firmer commercial evidence than a partnership announcement. It still leaves the contract value, mission, hardware configuration and delivery schedule undisclosed. Unibap’s suggestion that a larger order could follow is a supplier expectation, not an agreed expansion.

Loft Orbital’s October 12 release says the first spacecraft in the Orbitworks–Loft 50-satellite program launched on October 1 with two NVIDIA Jetson AGX Orin modules. It names Mistral models and nearly twenty other onboard models for tasks such as ship detection, wildfire mapping and scene description.

The event predates the previous cutoff, but the detailed company disclosure does not. No commissioning result, processing benchmark or paying workload is reported. Loft is evaluating NVIDIA Space-1 and IGX Thor for later satellites; evaluation is not a supplier selection.

SignalSpore now has 68 published project profiles and 446 supplier records, including 60 classified contract awards. No new project profile was created between this week’s cutoffs. Research instead filled older supplier and evidence gaps, so the larger totals describe better coverage more than a sudden wave of deals.

The distinction is especially important when confidential or historical contracts enter the record: verification this week does not make the underlying award a new event.

On October 15, Elon Musk wrote on X that the “Star Mind” computer SpaceX is co-engineering with NVIDIA is performing better than expected and will reach volume production next year. The new signal is his confidence, not a revised public schedule.

This remains an executive forecast. The post gives no completed test, unit count, factory output or procurement value. A joint technical release, flight hardware and dated manufacturing evidence would strengthen it; schedule slippage or unresolved deployment approvals would cut the other way. It is not included in SignalSpore’s project or supplier totals.

This week points toward practical onboard work—sensor processing, signals intelligence and rapid alerts—rather than a single generic “data center in space” market. Hardware and small contracts are appearing before the economics of large orbital compute fleets.

Watch for commissioning results from the October 14 launch, named delivery milestones behind Unibap’s contract, customer workloads on the Orbitworks satellite and evidence that Starmind has moved beyond a production forecast.