In December 2025, Starcloud (NVIDIA-backed) trained the first LLM ever trained in space, on an H100, roughly 500km above Earth. Same GPU class we orchestrate across RunPod, Vast.ai, Lambda, and Hyperstack every day for ground-based AI teams.
The infrastructure story here is genuinely interesting: orbital compute is still experimental (thermal management, radiation, power constraints are hard problems), but the model development behind space-bound AI, satellite imagery analysis, autonomous navigation, mission planning, still happens on Earth, on the same class of hardware we already route jobs to.
We're not building orbital infrastructure. We're the ground layer serving teams building toward it, using the exact multi-provider GPU orchestration we've been live-testing and publishing results on all week.
Curious if anyone here is building for space/aerospace AI applications and what your actual GPU requirements look like, happy to compare notes.
Top comments (0)