Nvidia's H100 GPUs are going to space– space based solar-powered AI data centers (www.tomshardware.com)

Nvidia H100 GPUs will be launched into orbit as part of a partnership between AI factory Crusoe and satellite data-center startup Starcloud: Starcloud plans to send H100s on an initial mission in November 2025, with Crusoe deploying its Crusoe Cloud on a Starcloud satellite by late 2026 and offering limited space-based GPU capacity in early 2027. The satellites are intended to host AI accelerators and very large solar arrays, with Starcloud aiming for data-center scale up to gigawatts of power generation. This matters because the partners claim an “energy-first” model: unfiltered sunlight in orbit can yield much higher and more consistent solar throughput than on Earth, potentially cutting energy costs by an order of magnitude (their estimate: ~10x lower, including launch costs). For AI/ML this could enable cheaper, renewable-powered large-scale training and inference. Key technical implications include using the vacuum of space for radiative heat rejection (Starcloud calls deep space an “infinite heat sink”), the need for radiation-hardened or error-mitigated GPU stacks, massive radiator and panel infrastructure (the article cites multi-kilometer arrays), and susceptibility to orbital debris and space weather. Practical constraints—thermal engineering for high-power GPUs, fault rates from radiation, collision avoidance, latency to Earth, and the economics of launches—remain open questions that will determine whether this “space cloud” model is viable at scale.