Starcloud, which launched a satellite with a Nvidia H100 chip in November, says the satellite is running and querying responses from Google's Gemma
Nvidia-backed startup Starcloud trained an artificial intelligence model from space for the first time, signaling a new era for orbital data centers … X: @philipjohnston X: Philip Johnston / @philipjohnston : We just trained the first LLM in space using an @Nvidia H100 on Starcloud-1! 🚀 We are also the first to run a version of @Google's Gemini in space! This is a significant step on the road to moving almost all compute to space, to stop draining the energy resources of Earth and [image]
Context & Ripple Effects
Starcloud’s H100 experiment turns orbital AI infrastructure from an ambition into a demonstrated workload: a satellite is not merely carrying an accelerator but training an LLM and serving queries from a Google model. That makes the adjacent reports of Blue Origin and SpaceX work on orbital AI compute more concrete competitive context rather than distant concept-stage activity.
The later emergence of a dedicated Nvidia GPU platform for orbital data centers and Starcloud’s subsequent $170M Series A show how quickly the experiment became part of a capital-and-hardware roadmap for space-based compute.
First-order effects
- Starcloud gains a live technical proof point for its Starcloud-1 architecture, demonstrating that an Nvidia H100 can support model training and Google-model queries in orbit.
- Nvidia’s H100 becomes a validated component in a new deployment environment, while Google’s Gemma/Gemini-family models gain an additional demonstrated inference setting.
Second-order effects
- Orbital-compute rivals and prospective launch partners face pressure to move from broad data-center claims toward demonstrable on-orbit workloads, reliability, and model-serving performance.
- Hardware vendors can treat space as a distinct AI-compute target, encouraging purpose-built systems such as Nvidia’s later Space-1 Vera Rubin platform rather than relying solely on terrestrial data-center designs.
Third-order effects
- If repeatable missions follow, AI capacity may be planned across terrestrial and orbital sites, making deployment constraints, power availability, and launch cadence part of infrastructure strategy rather than only data-center operations.
- The key industry question shifts from whether AI can run in orbit to whether orbital capacity can be operated and scaled competitively against Earth-based infrastructure; this single-satellite result does not resolve that economics test.
The trend: This is an early data point in the push to diversify AI infrastructure beyond terrestrial data centers through specialized, distributed compute environments.