Published on Feb. 06, 2026

While much media attention around the world has focused on the latest large language models (LLMs) released in both the United States and China, a relatively underreported aspect of the “AI race” concerns the infrastructure needed to power data centers, especially to meet their enormous need for energy and cooling.

Both countries have begun deploying a new type of satellite into Earth’s orbit: computing satellites that compete for geostationary positions with traditional navigation, communication, and remote-sensing satellites in an increasingly crowded orbit.

Alibaba Cloud's Qwen-3 model was uploaded to several satellites last November, a leading executive involved with China's "Star Compute Program Phase-01" told participants of a symposium, reported the newspaper Nanfang Dushi Bao on January 27, 2026.

What followed was one of the first examples of using a general-purpose LLM in outer space. The entire process of uploading queries from a ground station to the GPUs deployed on a satellite, completing the inference workload, and sending the answers back to Earth reportedly took less than two minutes.

This means China has now conducted a similar AI-in-space experiment as Starcloud, the Nvidia-backed startup in the United States that launched a satellite with an H100 GPU onboard a SpaceX rocket last November.

Starcloud ran a customised version of Google's open-weight LLM Gemma in space. China's Qwen-3 LLM deployed by Adaspace is also open-weight. Both experiments happening in the same month highlight the increasingly intense race to develop AI capabilities in space between the two countries.

SpaceX founder Elon Musk, trying to impress investors ahead of his company's planned IPO rumoured to raise up to US$25 billion later this year, made it all sound easy at the recent World Economic Forum meeting in Davos.

"It's a no-brainer building solar-power data centers in space ; the lowest-cost place to put AI will be space and that will be true within two years, three at the latest," Musk said.

Both China and the US plan to build more solar-power stations and data centers in orbit. China's main space contractor, China Aerospace Science and Technology Corporation (CASC) has announced plans to "construct gigawatt-class space digital-intelligence infrastructure" within the next five years, reported the Chinese state broadcaster CCTV.

Critics warn that there are still a lot of technical hurdles to overcome, though. The long-term effects of radioactive radiation in space on sensitive semiconductors and the challenge of effectively cooling large data centers without air or water are still waiting to be addressed, experts say.

However, both China and the United States seem determined to move early for a number of reasons. One is that the lower orbit is getting increasingly crowded, increasing complexity and costs for late-comers.

The other is that technological and physical dominance of space is seen not only as a huge commercial opportunity but also as a necessity from strategic and military perspectives.

Beijing has made the integration of space-based solar power stations with AI a core item of its 15th Five-Year Plan, expected to be officially adopted in the coming weeks. Adaspace, which has deployed the first constellation of 12 satellites for its Starcompute Program in May last year, plans to launch two more clusters of AI satellites this year alone.

The AI race has now been extended to outer space for good.