NVIDIA Announces 'Space Computing'—AI Is Finally Heading to Orbit
On March 16, 2026, NVIDIA held its developer conference, GTC 2026, in San Jose. The announcements made there were impossible to ignore, even from the perspective of the space business. The company introduced a new concept called 'Space Computing.' In short, it is a technology foundation designed to enable AI processing—which has traditionally been handled by data centers on the ground—to be performed in orbit.
The era of processing data where it is 'born'
Satellites generate vast amounts of data every day. Earth observation, weather forecasting, communications relay—information acquired by these satellites has traditionally been sent to ground stations for processing. However, this architecture has its limits. Communication bandwidth to the ground is finite, and time lags in data delivery are unavoidable.
The paradigm shift of 'processing data where it is born' is now becoming a reality.
NVIDIA announced a space-accelerated computing foundation consisting of three platforms: the 'NVIDIA Space-1 Vera Rubin Module,' 'NVIDIA IGX Thor,' and 'NVIDIA Jetson Orin.' Each aims to provide data-center-class AI processing power while meeting the size, weight, and power constraints (SWaP constraints) essential for satellite deployment.
Of particular note is the performance of the flagship model, the 'Vera Rubin Space Module.' This module is said to deliver up to 25 times the AI computing performance of NVIDIA's H100 GPU for orbital inference workloads. It is designed with the premise of running large language models and advanced foundation models directly in space, and its architecture, which tightly couples CPU and GPU with a high-bandwidth interconnect, can process large-scale data streams generated by space-borne sensors in real time.
Six space companies have already adopted it
In its announcement, NVIDIA emphasized not just a product roadmap, but the fact that 'partners are already in action.' Six companies—Aetherflux, Axiom Space, Kepler Communications, Planet Labs PBC, Sophia Space, and Starcloud—are utilizing NVIDIA's platform for their next-generation missions.
Mina Mitry, CEO of Kepler Communications, stated, 'By integrating the NVIDIA Jetson Orin directly into our satellites, we can intelligently manage and route data across the entire constellation. This evolves the entire network into a smarter and more efficient platform.'
Planet Labs continues to operate by imaging the entire Earth every day, and the accumulated image data is enormous. Will Marshall, co-founder and CEO of the company, said, 'By integrating NVIDIA's accelerated platform from space to ground, we are significantly enhancing our ability to index the entire Earth. We are leveraging the NVIDIA CorrDiff AI model to convert raw pixel data into actionable insights in near real-time.'
CEO Jensen Huang said in his keynote address, 'Space computing, the final frontier, has finally arrived. By extending AI processing across space and ground systems, we enable real-time sensing, decision-making, and autonomy.'
Starcloud—The current state of space data centers
Among the six partners announced, there is one company I have covered on this blog before. Yes, it is Starcloud.
As I introduced in my previous article on orbital data centers,
Starcloud is a startup based in Redmond, Washington, that is attempting to build a purpose-built data center in space. In space, solar energy is available 24 hours a day, and the natural vacuum of space can be utilized for cooling. The orbital environment is free from the power and cooling cost issues that plague ground-based data centers.
Philip Johnston, CEO of Starcloud, said, 'Through our collaboration with NVIDIA, we can bring true hyperscale-class AI computing to orbit. By processing data at its source, we reduce our reliance on downlinks and enable the execution of training and inference workloads in space for the first time.'
'Processing data at its source'—this is the core of Starcloud's philosophy and perfectly aligns with the vision NVIDIA presented in this announcement. The significance of these two teaming up is by no means small.
The strategy for ground integration is also not to be missed
What is interesting about this announcement is that it strengthens not only the space side but also the ground-side processing. The 'NVIDIA RTX PRO 6000 Blackwell Server Edition GPU' is specialized for ground processing for geospatial intelligence and is said to achieve up to 100 times the speed compared to traditional CPU-based batch processing systems.
Not all data acquired by satellites can be processed in orbit. By realizing a division of labor where real-time decisions are made in orbit and detailed long-term analysis is performed on the ground, a seamless AI infrastructure from space to ground is completed.
This concept represents a fundamental transformation in data utilization within the space business. A world is approaching where these use cases—such as wildfire detection, flood damage assessment, agricultural yield forecasting, and climate change tracking—using satellite data will function in real-time without latency.
A New Phase: The 'Cloudification of Space'
Looking at it from a broader perspective, NVIDIA's announcement this time can be read as the full-scale beginning of the 'cloudification of space'.
Just as cloud computing on Earth transformed the way industries operate, the distributed deployment of computing resources in orbit should give rise to services and decision-making mechanisms that were previously impossible. The space business is shifting its center of gravity from the world of hardware and rockets to the world of data and software.
The IGX Thor, Jetson Orin, and RTX PRO 6000 Blackwell Server Edition GPU are already available, while the Space-1 Vera Rubin Module is scheduled for release at a later date.
What kind of companies will utilize this platform, and what kind of services will emerge? It looks like we are in for a year where we cannot take our eyes off the front lines of space and AI.
Reference Articles
○ NVIDIA Launches Space Computing, Rocketing AI Into Orbit (March 16, 2026)
○ Nvidia announces Vera Rubin Space Module — up to 25x the AI compute of H100 for orbital data centers (March 16, 2026)
○ Nvidia unveils Vera Rubin, Groq 3 at GTC 2026 (March 2026)
○ The Future AI Infrastructure Realized by 'Starcloud,' a Data Center Floating in Space (November 1, 2025)