Starcloud (blogs.nvidia.com)

Starcloud, a Redmond startup in NVIDIA’s Inception program, announced it will launch Starcloud‑1 in November — a 60‑kg, fridge‑sized satellite carrying an NVIDIA H100 data‑center GPU, marking the first deployment of a state‑of‑the‑art, data‑center‑class GPU in orbit. The satellite promises roughly 100× more GPU compute than prior space operations and leverages orbit’s near‑continuous solar power and the vacuum of space as an infinite heat sink to avoid Earth’s evaporative cooling needs. Starcloud projects roughly 10× energy and cost savings (and a 10× reduction in CO2 over a data center lifetime) even after accounting for launch, and plans to run Google’s open model Gemma in orbit to demonstrate large‑model inference and fine‑tuning in space. For the AI/ML community this is significant because in‑orbit compute enables low‑latency, bandwidth‑efficient processing where massive sensor data is collected — a key advantage for Earth observation workflows such as optical, hyperspectral and SAR imaging (SAR can produce ~10 GB/s). Running inference in space can cut response times for wildfire detection, distress signals and rapid map generation from hours to minutes while reducing downlink load. Starcloud’s use of NVIDIA accelerated computing — and future plans to adopt the Blackwell platform for potentially another ~10× performance jump over Hopper — signals a new path for scaling AI workloads off‑planet, with implications for model placement, data locality and sustainable high‑performance inference.