The future of data centers may soon be found beyond Earth's atmosphere. Starcloud, a startup participating in NVIDIA's Inception program, is preparing to launch an AI-powered satellite, marking a significant step towards creating data centers in space.

This initiative aims to address the growing energy demands and cooling challenges associated with terrestrial data centers.

The Promise of Space-Based Data Centers

Starcloud envisions building a 5-gigawatt orbital data center, utilizing expansive solar panels and cooling systems spanning approximately 4 kilometers in width and length. Philip Johnston, cofounder and CEO of Starcloud, highlights the potential for nearly unlimited, low-cost renewable energy in space. He also emphasized that the environmental impact, limited to the launch phase, would result in a 10x reduction in carbon dioxide emissions compared to Earth-based data centers.

NVIDIA H100 GPU's Cosmic Debut

The upcoming satellite launch in November will feature the NVIDIA H100 GPU, representing its first deployment in space. The 60-kilogram Starcloud-1 satellite, roughly the size of a small refrigerator, is projected to deliver 100 times more GPU computing power than any previous space mission.

Sustainability Advantages

Space-based data centers offer unique sustainability benefits. Unlike Earth-bound facilities that often rely on freshwater for cooling, Starcloud plans to use the vacuum of space as a natural heat sink, conserving significant water resources. Continuous solar exposure in orbit eliminates the need for batteries or backup power.

Starcloud estimates that energy costs in space could be 10 times lower than those on Earth, even when factoring in launch expenses. Johnston predicts that within a decade, most new data centers will be constructed in outer space.

Applications in Space

Early applications for space-based data centers include:

Analyzing Earth observation data to improve crop monitoring.
Predicting local weather patterns.

Real-time data processing in space can also significantly enhance critical operations like wildfire detection and distress-signal response, reducing response times from hours to minutes.

Why NVIDIA?

According to Johnston, Starcloud chose NVIDIA GPUs for their superior performance in training, fine-tuning, and inference, making them competitive with Earth-based data center workloads.

The company also plans to integrate the NVIDIA Blackwell platform in future launches, anticipating a tenfold increase in AI performance compared to the NVIDIA Hopper architecture.

Starcloud is also planning to run Gemma, an open model from Google, in orbit on H100 GPUs, demonstrating the viability of running large language models in space.

Johnston emphasized that being part of NVIDIA Inception has been critical, providing the startup with technical support, access to NVIDIA experts, and NVIDIA GPUs.