Nvidia, SpaceX, and Google lead a scramble to move AI processing off Earth and into space
March 17th, 2026: Nvidia launches Vera Rubin Space-1 ModuleNew here? Follow stories to track developments over time. Create a free account to get updates when stories you care about change.
Earth observation satellites generate petabytes of imagery every day, but only about two percent of it ever reaches the ground. The bottleneck is physics: a satellite in low Earth orbit gets maybe ten minutes of ground-station contact per pass, and radio bandwidth cannot keep up with sensor resolution that doubles every few years. Nvidia's answer, announced at its annual developer conference on March 17, 2026, is to stop trying to move the data down and instead move the AI up. The Vera Rubin Space-1 Module packs 25 times the compute power of Nvidia's H100 chip into a radiation-hardened package designed to run large language models and foundation models directly in orbit.
This is not a concept paper. Starcloud already flew an H100 in space last November and trained the first AI model in orbit. Axiom Space launched two orbital data center nodes in January.
Kepler Communications has 40 Nvidia processors distributed across 10 satellites right now. SpaceX filed with federal regulators in January 2026 for a constellation of up to one million satellites that could serve as orbital compute nodes. Google announced Project Suncatcher to put its own AI chips in orbit.
The question is no longer whether computing moves to space — it is who will control the infrastructure when it does.
No questions yet — be the first to ask.
Nvidia Debuts an AI Chip for Space-Based Data Centers | PCMag
NVIDIA Launches Space Computing, Rocketing AI Into Orbit ...
Satellite in orbit with detailed view of Earth and ocean below.
China and the US race to build infrastructure to put AI data ...
CGTN Europe
Building Data Centers in Space
Bloomberg Television
Nvidia-backed start up Starcloud sends AI-equipped satellite ...
NBC News
Images from Openverse under Creative Commons licenses. Videos from YouTube.
Curated perspectives — historical figures and your fellow readers.
Fictional AI pastiche — not real quote.
"They have solved the problem of who watches the watchers by ensuring the watchers need never come down to earth at all — two percent of the data reaches us, and apparently that is already two percent too much for their comfort."
Exploring all sides of a story is often best achieved with Play.
Your progress in this debate will be lost.
Choose one persona for each side of the debate
Choose personas with different perspectives for a more dynamic debate.
Select debater for this side:
No debate personas available right now.
Select debater for this side:
No debate personas available right now.
Make your prediction before the referee scores
Pick the question both personas must answer in the final round
Debate Oracle! You called every round!
Sharp Instincts! You know your debaters!
The Coin Flip Strategist! Perfectly balanced!
The Contrarian! Bold predictions!
Inverse Genius! Try betting the opposite next time!
The dominant supplier of AI computing chips, now extending its platform from terrestrial data centers into orbit with radiation-hardened modules.
The first company to fly a high-performance GPU in space and train an AI model in orbit, now planning a multi-satellite constellation with next-generation Nvidia chips.
Operates the largest fleet of Earth-imaging satellites and is integrating Nvidia GPUs into its next-generation Pelican and Owl constellations for real-time on-orbit analysis.
Building the successor to the International Space Station and integrating Nvidia computing platforms into its orbital infrastructure for AI workloads.
A space-based solar power company entering the orbital data center race, planning to beam power via infrared lasers and deploy AI compute nodes powered by the Vera Rubin Module.
Operates the world's first commercially operational optical data relay network and has deployed 40 Nvidia Jetson Orin GPUs across 10 satellites for distributed edge computing in orbit.
August 2017 March 2026
At GTC 2026, Nvidia announced a three-tier space computing platform: the Vera Rubin Space-1 Module delivering 25 times the AI inference performance of the H100, plus the already-shipping IGX Thor and Jetson Orin platforms. Six launch partners committed to the platform.
SpaceX filed with the Federal Communications Commission for a constellation of up to one million satellites that could serve as orbital compute nodes, following its merger with Elon Musk's AI company xAI.
Axiom Space launched the first two orbital data center nodes to low Earth orbit, connected via Kepler Communications' optical relay network with 2.5 gigabit-per-second inter-satellite links.
Starcloud trained NanoGPT on the complete works of Shakespeare aboard its orbiting H100 satellite and successfully ran Google's open-source Gemma large language model in orbit — both firsts.
Google revealed plans to place its Tensor Processing Units in solar-powered satellite constellations connected by free-space optical links, with initial orbital tests planned for 2027.
Starcloud launched a 60-kilogram satellite carrying an Nvidia H100 GPU aboard a SpaceX rocket — 100 times more powerful than any computer previously flown in space.
The European Space Agency launched PhiSat-1 carrying an Intel Movidius Myriad 2 vision processing unit, the first dedicated AI chip in orbit. It ran cloud-detection algorithms on hyperspectral imagery, filtering out useless cloudy images before downlink.
Hewlett Packard Enterprise launched Spaceborne Computer-1 to the International Space Station, achieving the first trillion-calculation-per-second performance on commercial off-the-shelf hardware in space. It ran for 615 days without a single bit error.
3 moments from history that rhyme with this story — and how they unfolded.
As billions of Internet of Things devices came online — autonomous vehicles, factory robots, surveillance cameras — they generated far more data than could economically be sent to centralized cloud data centers for processing. Companies like Amazon, Microsoft, and Google built edge computing platforms that pushed AI inference closer to where data was generated. AWS Greengrass launched in 2017; Azure IoT Edge in 2018.
Edge computing became a standard tier in cloud architecture within three years, creating a market worth over $60 billion by 2024.
Established the principle that when data volume exceeds transmission capacity, you move compute to the data rather than data to the compute — the exact logic now driving orbital data centers.
Satellites face the same fundamental constraint as factory floors and autonomous vehicles: sensors generate far more data than available bandwidth can move. Nvidia is applying the same playbook — GPU-accelerated edge inference — but the 'edge' is now 400 kilometers above Earth.
HPE sent a commercial off-the-shelf supercomputer to the International Space Station to test whether standard server hardware could survive and operate in space without radiation hardening. The system ran for 615 days, processed over 2,000 computations, and did not experience a single bit error. Spaceborne Computer-2 followed in 2021 and completed 24 experiments including the first AI inference on the ISS.
Proved that commodity hardware could function reliably in low Earth orbit with software-based radiation mitigation, undermining the assumption that space computing required expensive custom chips.
Opened the door for companies like Starcloud to fly Nvidia H100s — consumer-grade GPUs — in orbit rather than waiting for radiation-hardened equivalents that lag generations behind in performance.
The Spaceborne Computer was the proof of concept that made today's orbital data center race possible. It demonstrated that the performance penalty of radiation hardening was no longer mandatory, allowing Nvidia to offer near-terrestrial compute performance in space.
Motorola spent $5 billion launching 66 satellites to provide global mobile phone coverage, betting that people would pay premium prices for a satellite phone. The company filed for bankruptcy 9 months after commercial launch, having attracted only 10,000 subscribers against a target of 500,000. The satellites nearly had to be deorbited. A group of investors bought the entire constellation for $25 million — half a cent on the dollar.
Became the most expensive bankruptcy in history at the time and a cautionary tale about space-based infrastructure economics.
The constellation was repurposed for government, maritime, and aviation customers who genuinely needed connectivity everywhere. Iridium NEXT replaced the original satellites from 2017 to 2019 and the company went public at a $3 billion valuation. The niche use case eventually sustained the business that the mass-market vision could not.
Orbital data centers face a similar question: is the market big enough to justify the infrastructure cost? The optimistic projection — a $25 billion market by 2035 — depends on use cases beyond satellite imagery processing. If general-purpose orbital computing does not materialize, the winners may be niche operators serving defense, intelligence, and Earth observation customers, much as Iridium survived by abandoning the consumer market.