Orbital computing is moving from onboard processing toward commercial infrastructure — but the paying-customer evidence is still thin

§What changed

Radiation-tolerant flight processors have improved: NASA's High Performance Spaceflight Computing (HPSC) program, developed with Microchip Technology, targets a next-generation flight computer for missions through 2040 and is documented in NASA project pages and technical white papers. Commercial launch cadence has made hosted-payload experiments routine — Starcloud-1 flew on a SpaceX rideshare into a ~325 km orbit, and Kepler's first optical relay tranche launched on a Falcon 9 on January 11, 2026. Optical intersatellite links are shifting from demonstrations toward commercial services, with Kepler explicitly commissioning an optical relay network in 2026 that Axiom Space has publicly contracted to host compute payloads on. Company-reported milestones now include: Lonestar's IM-1 Independence Payload data-storage test in February 2024; Lonestar's IM-2 Freedom cislunar operation in early March 2025; Axiom's April 2025 announcement of ODC nodes on Kepler satellites; Axiom and Red Hat's September 2025 AxDCU-1 prototype on the ISS; Starcloud-1 with an NVIDIA H100 launched in November 2025 and reported to have run Google's Gemma model in December 2025; and Google's November 4, 2025 announcement of Project Suncatcher as a research direction rather than a product.

§Why it matters

If a repeatable commercial service does emerge on orbit, it changes three things at once. It reduces the need to downlink every raw observation from Earth-observation constellations by processing data before it leaves orbit, which matters for real-time defense, disaster response, and satellite-image analytics. It creates an alternative venue for AI workloads whose bottleneck is not silicon but power and cooling, which is the stated motivation for both Starcloud and Google's Suncatcher. And it opens a data-sovereignty story — computing performed outside any single national jurisdiction — that has already been used in Axiom Space's public positioning for national-security customers. All three benefits remain largely proposed, not demonstrated at scale.

§What most people may be missing

Coverage often collapses very different things into the phrase "orbital data center." A single ISS-based edge server, a lunar-transit storage test, a rideshare satellite running one GPU, an announced hosted-compute payload, and a research paper about future TPU constellations are not the same product. As of mid-2026, most public claims of "commercial" activity describe either (a) a provider-to-provider arrangement — Axiom buying compute hosting from Kepler, for example — or (b) demonstrations funded by the provider itself. That is not the same as an unaffiliated customer paying for a delivered orbital compute or storage workload. The gap between what has flown and what has been sold as a service is the story most missing from public coverage.

§What to watch next

• Whether Axiom Space or Kepler publicly announces a named external customer whose paid workload actually ran on the orbital nodes after commissioning completes.
• Whether Starcloud transitions from internal Gemma/nano-Banana demonstrations to a disclosed paying tenant.
• Whether Lonestar completes a persistent lunar-surface storage deployment following the IM-2 landing anomaly.
• Whether NASA HPSC deliveries and follow-on missions accelerate radiation-tolerant compute available to commercial providers.
• Whether ESA/Thales Alenia move ASCEND from feasibility toward a funded flight demonstrator.
• Whether independent regulatory filings (FCC, ITU) show sustained downlink of processed customer data rather than telemetry.

§Skeptical view

The skeptical case is straightforward and mostly untested. Radiation, thermal management, power supply, and hardware maintenance in orbit remain genuinely hard; ground-based hyperscale sites can be upgraded in place while an orbital node has whatever silicon it launched with. Terrestrial data centers still have overwhelming cost advantages at the workload level, and moving bulk data up to orbit is usually a worse trade than moving processed insights down. Several currently-cited "commercial" milestones dissolve on close reading: they are provider-funded demonstrations, hosted-payload experiments, or intercompany arrangements between the orbital-compute provider and its own launch or relay partner. Others sit in cislunar transit rather than persistent orbital operation. Some, like Google's Project Suncatcher, are explicitly research programs rather than product announcements. There is also a plausible reading in which "orbital data center" is currently more marketing frame than deployed infrastructure, and edge processing on individual satellites is being conflated with data-center-class services. Serious independent verification of an external customer's paid, delivered workload on orbit remains scarce.

§Key facts

• Starcloud-1 launched in November 2025 with the first NVIDIA H100 GPU in orbit and separated into an approximately 325 km orbit.
• In December 2025, Starcloud reported running Google Gemma large language model on Starcloud-1.
• On April 7, 2025, Kepler Communications announced that Axiom Space had purchased the first on-orbit computing payloads on Kepler optical data relay network.
• On January 11, 2026, Kepler Communications launched the first tranche of its optical relay satellites from Vandenberg Space Force Base on a SpaceX Falcon 9; the satellites are now in commissioning.
• Axiom Space and Red Hat AxDCU-1 orbital data center prototype reached the International Space Station in September 2025 as an edge-computing testbed.
• Lonestar Data Holdings and Space Florida described the Independence Payload on Intuitive Machines IM-1 mission in February 2024 as a successful test of the lunar data-storage concept.
• Lonestar reported that its Freedom data center payload met technical and commercial milestones in cislunar space en route to the Moon in March 2025; the IM-2 Athena lander itself did not achieve nominal lunar-surface operations.
• On November 4, 2025, Google announced Project Suncatcher, a research moonshot exploring solar-powered satellite constellations equipped with TPUs and free-space optical links.
• NASA High Performance Spaceflight Computing project is developing a next-generation flight computing system intended to serve NASA missions through 2040 and beyond.
• The Thales Alenia Space-led ASCEND feasibility study, funded by the European Commission under Horizon Europe, published results in June 2024 concluding that orbital data centers appear technically, economically, and environmentally feasible at study level.

§Evidence and sources

Citations link to the primary sources used to compile this signal.