As the concept of "space data centers" continues to gain momentum, orbital computing is entering its early commercial phase. Currently, while the global scale of GPUs in orbit remains limited, relevant infrastructure and application scenarios are beginning to take shape gradually. In January this year, Canadian company Kepler Communications launched what is currently known as the largest on-orbit computing cluster, consisting of 10 satellites equipped with approximately 40 Nvidia Orin edge processors interconnected via laser communication links. The company now serves 18 clients and announced this week a partnership with startup Sophia Space. Under the agreement, Sophia will upload its self-developed operating system onto Kepler's satellites and attempt deployment and configuration across six GPUs located on two spacecraft. While such operations are routine at ground-based data centers, they represent a first-of-its-kind trial in an orbital environment. This test will provide critical technical validation for Sophia's planned launch of its inaugural satellite by late 2027. Industry consensus suggests that large-scale space data centers envisioned by entities like SpaceX or Blue Origin may not materialize until the 2030s. Until then, the core value of orbital computing will primarily manifest through "edge processing"—performing computations directly at the point of data collection—thereby enhancing the response speed and efficiency of satellite sensors. Mina Mitry, CEO of Kepler, stated that the company positions itself not as a traditional data center operator but rather as an infrastructure platform providing networking and computing services for satellites, drones, and aircraft. Meanwhile, Sophia focuses on developing passively cooled space computers to address thermal management challenges for high-performance processors in the orbital environment. With growing demand for satellite data processing among institutions including the U.S. military and increasing restrictions on terrestrial data center construction in certain regions, industry observers note rising appeal for space-based compute capacity as an alternative solution. Although widespread deployment remains distant, distributed inference computing models could emerge as the first practical implementation scenario for orbital computing.