Chinese startup Adaspace Technology has deployed the neural network in a space-based computing center that is part of the Star-Compute Project — a constellation of 2,800 satellites designed to power physical artificial intelligence and support training and inference tasks.

The large language model successfully completed several tasks, according to Wang Yabo, Executive Vice President of the company. He said that uploading prompts from Earth and receiving responses took less than two minutes.

Adaspace launched its space computing center in May 2025, becoming the world’s first constellation of 12 satellites dedicated to AI computing.

Space as the future of AI

Space is increasingly seen as a promising environment for deploying AI hubs, although most projects are still at the planning stage.

In January, Elon Musk announced that Tesla would resume work on Dojo3 — a previously shelved third-generation chip project originally intended for electric vehicles. Its new purpose is space-based computing.

Musk and several other tech leaders believe the future of data centers lies beyond Earth, arguing that the planet’s energy infrastructure is nearing its limits.

Key advantages include virtually unlimited access to solar energy and ample physical space for hardware deployment. The main drawback remains the high cost of rocket launches and the necessary infrastructure.

Analysts at research group 33FG estimate that orbital AI computing could become economically viable by 2030.

Google was among the first to show interest, announcing plans to build a satellite network in low Earth orbit to generate power for data centers.

The idea is also supported by OpenAI CEO Sam Altman, but Musk holds a strategic advantage due to his control over launch capabilities.

The entrepreneur plans to use SpaceX’s upcoming IPO to fund his vision of deploying clusters of computing satellites using Starship — systems capable of operating under constant sunlight and generating energy around the clock.