Tech leaders like Elon Musk and Jeff Bezos propose launching data centres into orbit to power AI's massive computing needs, but experts highlight formidable hurdles. From vast solar panels and cooling issues to radiation risks, building such facilities in space remains far off. Projects like Google's 2027 prototypes show early interest, yet production-scale viability is distant.
The surge in demand for generative AI, such as ChatGPT, has intensified the need for enormous data centres requiring gigawatts of power—comparable to that consumed by millions of households. On Earth, these facilities increasingly rely on unsustainable energy sources like natural gas, as renewables struggle to provide the necessary scale and reliability.
To address this, figures like Elon Musk and Jeff Bezos have floated the idea of orbital data centres in low Earth orbit, harnessing constant sunlight via solar panels for uninterrupted power. Bezos, through his company Blue Origin, predicts gigawatt-scale facilities could emerge in 10 to 20 years.
Google is advancing more tangible efforts with Project Suncatcher, planning to launch two prototype satellites equipped with its TPU AI chips in 2027. Meanwhile, Nvidia-backed Starcloud made strides this year by deploying a single H100 graphics processing unit in space, though this pales against the million such chips reportedly used by OpenAI.
Experts remain skeptical about near-term feasibility. Benjamin Lee from the University of Pennsylvania states, “From an academic research perspective, [space data centres] are nowhere near production level.” Key obstacles include the immense physical footprint: AI's power demands necessitate square kilometres of solar panels, while cooling in the vacuum of space relies solely on radiating heat away, without Earth's evaporative methods. Lee notes, “Square kilometres of area will be used independently for both the energy, but also for the cooling.” Starcloud envisions a 5000-megawatt centre spanning 16 square kilometres—400 times the area of the International Space Station's solar arrays.
Additional challenges encompass high-energy radiation that can induce computational errors, necessitating restarts and error corrections, thus imposing a “performance discount” compared to Earth-based systems. Coordinating thousands of satellites would demand precise laser communications, complicated by atmospheric interference with Earth links.
Krishna Muralidharan at the University of Arizona views these as surmountable: “It’s not a problem, it’s a challenge,” citing potential innovations like thermoelectric devices to recycle heat into electricity. He adds, “It’s a question of when and not if.”
Looking ahead, even if AI's computational hunger plateaus—as early signs suggest with training requirements potentially peaking—space data centres could still aid lunar exploration or Earth observation, per Muralidharan.
