AI and the Move Off Earth: Why Companies Are Testing Data Centers in Space

The growth of large AI models has stretched the limits of power grids and available land, prompting engineers and investors to explore far-reaching alternatives. The IEA reports that global data centre electricity use in 2022 was 240–340 TWh, and that ML-related workloads have risen even as hardware efficiency improves. Faced with rising demand, some startups have proposed placing computing modules into low Earth orbit, where sunlight is stronger and radiators can reject heat to space.

Starcloud, which began as Lumen Orbit, has raised about $21 million in seed funding and plans a demonstrator mission to test data‑centre grade GPUs in orbit in 2025. The company has described a long-term vision that could scale to gigawatts of solar power, although achieving such ambition will require breakthroughs in launch economics, large deployable arrays, and in-orbit maintenance.

In-principle advantages are clear: orbital solar can deliver higher insolation per square metre, radiative cooling removes reliance on evaporative systems, and proximity to satellites can reduce raw data downlink needs. Nevertheless, the technical trade-offs include battery sizing for eclipse periods, radiation hardening for commercial GPUs, and the need for high-bandwidth optical links, challenges that have been highlighted in feasibility studies and industry reports.

Public and private research efforts are already under way: the European ASCEND feasibility study was published in June 2024, and experiments such as Microsoft’s Project Natick have shown unconventional sites can offer reliability gains. Ultimately, whether orbital data centres become a major part of the AI infrastructure mix will depend on the outcomes of near-term demonstrations and on continued reductions in launch and operations costs.