StarCloud is pioneering the development of data centers in space to capitalize on abundant, 24/7 solar energy, which is 8x more efficient per square meter than on Earth.
The company has successfully deployed the StarCloud 1 satellite, proving the viability of running commercial-grade GPUs like the NVIDIA H100 in space by training a nanoGPT model and running Gemini inference.
StarCloud has filed with the FCC for an 88,000-satellite constellation, a $100 billion project designed to provide 20 gigawatts of low-latency compute capacity primarily for AI inference workloads.
The economic feasibility of this venture hinges on decreasing launch costs, with SpaceX's Starship target of $10-$20/kg being the key inflection point that makes space-based energy cheaper than terrestrial alternatives.
9 quotes
Concerns Raised
Thermal management and heat dissipation in the vacuum of space.
Radiation effects, such as bit-flipping, on non-hardened commercial GPUs.
The risk of space debris and collisions (Kessler syndrome) with a mega-constellation of 88,000 satellites.
The economic model is heavily dependent on future launch vehicles like Starship achieving their target cost-per-kilogram.
Opportunities Identified
Accessing vast, continuous, and low-cost solar energy, bypassing terrestrial grid limitations.
Building a global, low-latency compute network for the rapidly growing AI inference market.
Overcoming terrestrial constraints on data center expansion, such as land availability and power shortages.
Establishing a first-mover advantage in the new market of space-based compute infrastructure.