When Artemis II beamed back rare images of the far side of the Moon, it wasn’t just a historic moment for science. It was a wake-up call for how data is handled in space.
Behind the mission’s livestreamed spectacle was an invisible backbone as torrents of telemetry on life support, navigation and space weather were processed not only on Earth but increasingly in orbit.
A new era of space computing is taking shape, and its implications stretch far beyond a single mission.
Space computing is the deployment of high-performance processors and AI data centres directly in Earth’s orbit, enabling satellite operations by processing vast data volumes on-site rather than beaming raw feeds back to ground stations.
It is an emerging field that addresses critical bottlenecks in bandwidth, latency and energy, and vital for the next era of space exploration and global connectivity.
Radiation-hardened GPUs, CPUs and specialised AI accelerators are integrated into satellites and orbital platforms.
Traditional satellites capture imagery, signals intelligence or environmental data, and relay it unprocessed to Earth for analysis in costly ground-based data centres.
However, space computing enables real-time edge processing where satellites analyse petabytes of data in orbit and transmits only actionable insights such as disaster alerts or crop yield predictions.
This shift leverages abundant solar power, near-absolute zero cooling from space vacuums, and distributed constellations for uninterrupted global coverage. Hardware must endure extreme radiation, temperature swings from -150°C to +150°C and micro-vibrations, demanding space-grade silicon with error-correcting architectures and redundant cores.
It’s like having a space-based internet with AI models independently training and making inferences using vast data streams from huge satellite networks such as Starlink or OneWeb.
Why It matters
The importance of space computing cannot be overstated amid the AI boom and satellite proliferation. More than 10,000 global satellites circling the Earth are generating zettabytes annually — far beyond terrestrial networks’ capacity. Downlink costs exceed US$1,000 per GB for high-resolution imagery, and latency delays impact critical uses such as wildfire detection and maritime tracking.
Orbital processing cuts these costs by 90 percent, accelerates decisions to milliseconds, and enhances security by keeping sensitive data off-planet.
For defence, space computing powers autonomous drones and hypersonic threat detection. In climate science, it refines models for sea-level rise while commercially, it fuels precision agriculture and 5G augmentation.
As AI models grow hungrier, space’s abundant energy and cooling offer sustainable scaling, averting Earth’s power grid crises.
Orbital leap
Leading technology players see this future and are racing towards it.
At GTC in March, NVIDIA launched the Vera Rubin Space-1 chip and module ecosystem. Named after the astronomer who mapped galaxies, this radiation-tolerant GPU delivers 25 times the AI performance of the H100, with 141GB HBM3e memory and NVLink for multi-satellite clustering. Optimised for orbital data centres, it handles exaflops-scale inference on hyperspectral imagery, enabling applications from urban planning to biodiversity monitoring.
Axiom Space into integrating the chip into its infrastructure for next-generation orbital missions and its successor to the International Space Station (ISS) for protein folding simulations.
Planet Labs is partnering with NVIDIA to integrate advanced AI into its imaging satellites for daily Earth scans, aiming to reduce processing time from days to minutes. Starcloud is envisioning deploying swarms of 1,000 plus Rubin-equipped satellites as a planetary brain.
AMD has also entered the space computing race. In November 2025, it unveiled expanded Versal Adaptive SoCs that includes the space-grade AI Edge Gen 2 series with lidless packaging for 15-year lifespans in geostationary orbits. These chips pack 10x scalar compute, integrated RF transceivers and Zynq UltraScale+ R5 for real-time autonomy to target Class Y (deep-space) reliability.
Versal’s programmable fabric allows post-launch reconfiguration, which is vital as missions evolve from Low Earth Orbit (LEO) imaging to lunar relays. A single SoC consolidates sensor fusion, AI inference and data compression to reduce satellite mass by 40 percent.
Beyond NVIDIA and AMD, a constellation of innovators is accelerating adoption. Lonestar Data Holdings plans to set up nuclear-powered data centres on the Moon by 2027 for Bitcoin mining and secure cloud. Airbus Defence is testing photonic chips for light-speed orbital links. China’s space agency is fostering domestic ecosystems with gallium-nitride processors.
Startups such as Axelspace and Iceye deploy radar-AI hybrids for all-weather maritime surveillance, while Microsoft Azure Orbital beams quantum-secure keys via laser comms.
Singapore in the space
Singapore is also carving a niche in space computing despite lacking a space agency until recently. In February 2026, Nanyang Technological University (NTU) spearheaded three Space Technology Development Programme (STDP) projects, including the TELO-S nanSat with onboard edge AI for hyperspectral analysis, which can process 1TB per day to downlink only anomalies, cutting bandwidth by 95 percent.
These projects are among the first efforts funded by the Space Access Programme (SAP), a component of STDP run by Singapore’s Office for Space Technology & Industry (OSTIn), with planned satellite launches each year in 2026, 2027 and 2028.
Established on April 1, 2026, the National Space Agency of Singapore (NSAS) has unveiled initiatives for in-space manufacturing and AI autonomy, partnering with ST Engineering for hybrid LEO-MEO (Medium Earth Orbit) constellations. This positions Singapore as a gateway for Asia-Pacific space tech.
The road ahead
Space computing is moving from concept to infrastructure. Its future points to widespread orbital AI data centres by 2028-2030, driven by unlimited solar power, edge processing for satellite data, and AI training unconstrained by Earth’s grids.
Like NVIDIA Founder and CEO Jensen Huang said at GTC, “AI belongs in space — the final frontier for computation.”
The race to build that frontier has already begun.
BY EDWARD LIM
Edward Lim is Editor of Entelechy Asia and Managing Consultant of CIZA Concept. This story is produced through a content partnership between Entelechy Asia and News on Tech, bringing together shared perspectives on technology and innovation.
Entelechy Asia 