OpenAI builds the "Stargate" and Musk sends computing power "to the sky". What path has China's AI infrastructure taken?

Smart Things Author | Wang Han Editor | Mo Ying The focus of competition in the AI ​​industry has fallen on the competition of computing power infrastructure. In January 2025, OpenAI teamed up with Oracle and SoftBank to launch the "Stargate" project, planning to invest US$500 billion in four years to build computing power supporting facilities with a total power of 10 gigawatts in the United States. But in just one and a half years, the project progress has not been smooth, and problems such as financing, construction, and power supply have continued to be exposed. For example, the expansion of the core campus in Abilene, Texas, has been suspended, project leaders have left one after another, and the original fully self-built plan has been changed to renting computing power. ▲OpenAI "Stargate" project In addition, the construction of data centers across the United States also faces layers of policy and environmental protection resistance. More than 30 states across the United States have promulgated more than 300 relevant regulations, and more than 200 environmental protection agencies have also jointly called for a moratorium on new data center approvals nationwide. The expansion of terrestrial computing power is limited by both electricity and policies. Musk therefore thought of another way out: moving computing power to space. SpaceX has submitted a plan to the U.S. Federal Communications Commission to launch millions of satellites, with an AI computing capacity of 100 gigawatts, and to build an on-orbit data center in low-Earth orbit. Musk believes that power supply is the core shortcoming restricting the development of computing power. A space computing power center that relies on solar power can bypass the limitations of ground power supply. The various bottlenecks exposed in overseas computing power construction have also made domestic manufacturers see the common problems in the industry, and they have begun to look for development paths that suit themselves. They are trying to find a method on the ground to systematically reconstruct AI infrastructure from chips to power grids, from single cabinets to AI factories, from single points of breakthrough to open system collaboration. It is against this background that on July 9, the 2026 Open Computing Technology Conference (OCTS 2026) will be held in Beijing. This conference brings together global open organizations such as OCP, OCTC, SPEC, CXL Alliance, UALink Alliance, UCIe Alliance, Firmware Industry Technology Innovation Alliance, as well as Inspur Information, Alibaba Cloud, ByteDance, NVIDIA, Samsung, China Mobile, Baidu, etc.