In 2025, a new race is unfolding in low Earth orbit. Google, Amazon, Elon Musk’s xAI, and Chinese companies including ADA Space and Zhejiang Laboratory are vying for dominance in building orbital data centers for artificial intelligence. Their goal: move AI computing and connectivity off-planet to relieve overloaded terrestrial data centers, harness solar power, and gain an edge in the global machine learning surge. How? By launching constellations of satellites equipped with onboard AI processors—designed to work faster, more efficiently, and almost nonstop [1][2][3].
Satellites Instead of Server Farms – Google Pilots Project Suncatcher
Google was first to unveil a detailed vision. Project Suncatcher, announced in November 2025, aims to create an orbital network of satellites fitted with Tensor Processing Units, all powered by solar energy. These satellites will orbit in a sun-synchronous trajectory, where solar panels can generate up to eight times more energy than on Earth’s surface. The vacuum of space provides natural cooling for the heat-intensive AI chips. Google, in partnership with Planet, plans to launch two prototype satellites by early 2027. The mission: test TPU radiation resistance and the performance of ultra-fast optical links [2][3][4][5][1].
Amazon has been less vocal but is keeping pace. Its Project Kuiper—a broadband satellite constellation—is set to underpin future cloud and AI services. Internal documents reveal Amazon’s plans for direct orbital data processing, integrated with AWS infrastructure [7]. Meanwhile, Elon Musk envisions orbital compute farms linked to the Starlink network. His startup xAI wants to train large AI models in space, leveraging unlimited solar power and sidestepping the constraints of Earth-bound server farms [8][9].
Amazon, xAI, and China: Competing Strategies for Orbital Data Processing
China has already moved from planning to action. In May 2025, it launched 12 satellites as part of the Three-Body Computation Constellation. Each is equipped with an AI model boasting 8 billion parameters and processors for edge computing. The initial cluster delivers five peta operations per second, with plans to scale to 2,800 satellites. In December, Zhejiang Lab reported its mini-constellation could analyze remote sensing data and detect gamma-ray bursts in real time. The ambitions are sky-high: over 1,000 satellites and computing power exceeding 100 trillion operations per second [4][5].
Other Chinese players, like Zhongke Tiansuan, are already testing space-based supercomputers with proprietary GPUs. The Aurora 5000 is slated for orbit next year. Beijing Astro-future Institute of Space Technology is planning a megawatt-scale data center in space by 2035, projected to outperform China’s entire current ground-based fleet [5][4].
Arguments, Risks, and Ambitions: Will Space-Based AI Rewrite the Rules?
Proponents of orbital AI argue that shifting workloads to space will ease pressure on Earth’s power grids, cut latency for emergency applications, and deliver rapid services even in remote regions. “It’s a way to process data closer to its source and with near-zero emissions,” engineers claim [1][2][3][4][5][6][7].
But this isn’t a risk-free endeavor. Google’s proposed Suncatcher constellation (81 satellites) would operate in one of the most crowded orbital bands. Analysts warn of rising debris hazards and the complexities of managing space traffic. Chinese officials are candid: it’s not just about technology, but also about setting the standards for a new layer of information infrastructure—much as they did with satellite navigation systems [3][5][8][9][10].
By 2025, orbital artificial intelligence is no longer science fiction. It’s becoming a new infrastructure layer—one that could shape everything from rural connectivity to military surveillance. Who will win this race, and what will it mean for Earth?