NVIDIA Heads to Space—Announcing the "Vera Rubin Space-1" Orbital AI Module at GTC 2026
Process data generated in space, in space. NVIDIA has given shape to this vision through concrete hardware and partnerships. Let's take a look at the "Vera Rubin Space-1" orbital AI module announced at GTC 2026, along with the six partner companies involved.
1. Vera Rubin Space-1—Bringing 25x the AI Computing Power of the H100 to Orbit
On March 17, 2026, GTC 2026 was held in San Jose. One of the highlights announced during Jensen Huang's keynote was the "Vera Rubin Space-1" module, an AI computing unit for orbital data centers.
It is an optimization of NVIDIA's next-generation "Vera Rubin" architecture for the space environment, and is said to achieve up to 25 times the AI computing performance of the H100. Featuring a CPU-GPU integrated architecture with a high-bandwidth interconnect, the key point is that it is designed not just as a space version of a GPU, but as an AI computing infrastructure that is self-contained in orbit.
Also announced were the IGX Thor and Jetson Orin. These are platforms for environments with strict constraints on Size, Weight, and Power (SWaP), covering edge AI processing within satellites and stations. With Vera Rubin Space-1 targeting full-scale orbital data centers and IGX Thor and Jetson Orin handling the edge, the composition is one of capturing the entire AI computing layer in space with a lineup of large, medium, and small solutions.
2. "Process where the data is generated"—Why NVIDIA is heading to space
Jensen Huang stated it clearly in his keynote: "AI processing must happen where the data is generated"—AI processing should be done where the data is born.
This is not an abstract vision, but stems from a practical awareness of the problem. The amount of data generated by Earth observation satellites and ISS experimental equipment is growing enormously every year, but there is not enough bandwidth or budget to downlink all that data to the ground. If pre-processing and real-time inference can be done in orbit, the data sent to the ground can be dramatically reduced, and immediate decisions can be made.
For NVIDIA, space is likely positioned as the next AI computing market following data centers, automobiles, and robotics. The platformer that seized GPU hegemony on the ground is extending its territory into orbit. Huang's words, "bringing intelligence to places it has never reached before," are a vision and, at the same time, a declaration of business strategy.
I have an article I wrote previously about what kind of services are about to be born in orbit.
3. Six partners—Capturing the orbital AI ecosystem all at once
There are six partners mentioned this time. When you line them up, their respective roles are clearly divided.
The data creators—Planet Labs PBC is a major data source operating over 200 Earth observation satellites, generating vast amounts of image data every day. "AI pre-processing in orbit" is the use case that hits home the most.
The place creators—Axiom Space is developing commercial modules for the ISS and is one of the leading candidates for a private space station (an overview of the private station competition is here). They are a player that will have a physical "place" in orbit. Starcloud is a company that makes orbital data centers its business, placing satellite data centers equipped with NVIDIA GPUs at the core of its business model.
The connectors—Kepler Communications is building an inter-satellite data relay network. If you are going to process data in orbit, you need infrastructure to pass data between satellites.
The power suppliers—Aetherflux is working on space-based solar power. Running AI chips in orbit requires stable power, so they are positioned as a partner in the power layer.
On the software side—Sophia Space is working on AI solution development for space environments and is responsible for the application layer running on NVIDIA chips.
Data generation, processing locations, communications, power, and software—they are prepared to enclose the entire value chain NVIDIA is not just selling chips and walking away; they are attempting to design the entire environment where AI runs in orbit, centered around their own platform. It is clear they are attempting to use the same strategy in space that they used to make the terrestrial CUDA ecosystem the de facto standard.
Starcloud is also summarized in this article.
4. No convection, only radiation—The cooling problem and the future of orbital data centers
The vision is grand, but there is a physically troublesome challenge: cooling.
Terrestrial data centers can dissipate heat using air or water. Space has neither. Since there is no convection, heat exhaust must be handled solely through radiation. The 25x increase in computational performance compared to the H100, conversely, comes with that much heat generation. The size and efficiency of the radiative cooling panels are highly likely to become the bottleneck for the amount of computation that can actually be used in orbit.
The launch timing and specific orbital insertion plans have not yet been announced. Cooling technology is an area that NVIDIA cannot solve alone, and collaboration with spacecraft manufacturers will be key.
Nevertheless, the impact of NVIDIA officially establishing "AI computing in space" as a business domain cannot be overlooked. Until now, radiation-hardened processors from defense manufacturers like BAE Systems and Microchip have been the mainstream for space-grade computing chips. If a trend emerges where cutting-edge consumer AI chips are sent into orbit, the data processing capabilities of the space industry will change by an order of magnitude.
Considering NVIDIA's ability to build ecosystems, there is a possibility that they will move from the "vision and partnership" stage to implementation surprisingly quickly. The progress of cooling technology and the timeline on which partner companies actually send satellites into orbit—that is the next point to watch.
Reference
NVIDIA unveils AI computing module for space-based data centers | SpaceNews
Nvidia chips for orbital data centers in space AI push | CNBC
※ This article is for informational purposes only and does not recommend the buying or selling of any specific financial products. We do not guarantee the completeness or accuracy of the information posted.