Google and SpaceX race to launch AI data centers into orbit by 2027

Major tech companies are turning to space to meet the soaring energy needs of artificial intelligence. Google and SpaceX are leading efforts to deploy data centres in orbit, where solar power could be far more efficient than on Earth. The move comes as AI workloads demand ever-greater computing resources and sustainable energy solutions.

Google’s 'Suncatcher' project plans to launch two test satellites with Tensor Processing Units (TPUs) in early 2027. These TPUs have already passed radiation tests simulating five years in low Earth orbit. However, additional error-correction systems would still be required for training new AI models.

Research by Google indicates that satellites in a dawn-dusk orbit could generate around eight times more solar energy annually than ground-based panels. This advantage could make space-based data centres a viable alternative to terrestrial ones. Yet, the economic feasibility depends on reducing launch costs to roughly $200 per kilogram. Meanwhile, SpaceX is preparing to integrate AI computing into its Starlink satellites, targeting a total capacity of 500 gigawatts per year. Both SpaceX and Blue Origin are also exploring concepts for orbital data centres. Despite the potential, challenges remain, particularly in thermal management for high-density TPUs operating in space. Google estimates that replacing a single 1-gigawatt data centre on Earth would require a swarm of about 10,000 satellites. The company’s findings highlight the scale needed to match current computing infrastructure in orbit.

The push for space-based AI data centres is gaining momentum, with Google and SpaceX leading the way. If launch costs fall and technical hurdles like heat dissipation are overcome, orbital facilities could become a reality. For now, test missions and energy-efficient designs remain key priorities.