The Zhitong Finance App learned that Guosheng Securities released a research report saying that according to a report from the Science and Technology Innovation Board Daily, on December 18 local time, Trump signed an executive order formulating a vision for an “America First” space policy and increasing support for space policy. At the same time, Karpathy was given nano-GPT training, and the first H100 space GPU training LLM was born. In the future, Starcloud will also integrate modules from cloud infrastructure company Crusoe to enable customers to deploy and run AI workloads directly from space. In the domestic market, Suzake-3 rocket recycling has accumulated valuable experience in China's commercial space technology; once recycling technology is implemented in the future, launch costs will be drastically reduced and the commercial space profit model will mature.
Guosheng Securities's main views are as follows:
Trump signs executive order to increase space policy support
According to a report by the Science and Technology Innovation Board Daily, on December 18, local time, the US White House issued a statement saying that Trump signed an executive order on the same day, setting out a vision for an “America First” space policy to ensure that America “leads the world” in space exploration, safety, and commerce. The order requires Americans to return to the Moon by 2028 and establish initial facilities for a permanent lunar outpost by 2030; directs the deployment of nuclear reactors on the Moon and in orbit. The order stimulates private sector innovation and investment by upgrading launch infrastructure and developing commercial pathways to replace the International Space Station by 2030.
In the lunar landing program, launch vehicles play an extremely important role. Starship, for example, has carried out many test flights and can stably implement a next-generation recycling method for “chopstick holder rockets.” At present, Starship has begun verifying the V2 version of the spacecraft and carrying out payload deployment tests, laying a solid foundation for subsequent Starlink V3 satellite launches and deep space exploration experiments, including lunar landing.
The first LLM to be trained by H100 space GPUs was born
According to Xinzhiyuan's report, the first H100 space GPU training LLM was born on December 11. It is based on Karpathy nano-GPT training. Not only that, but Google Gemma also successfully operated in space, sending the first greeting to the world: Hello, Earthlings. Based on the Starcloud-1 satellite supported by H100, the startup completed training using Shakespeare's corpus on Karpathy's nanoGPT.
Starcloud CEO Philip Johnston firmly believes, “Anything you can do in terrestrial data centers, I believe it can be done in space in the future.” Ultimately, Starcloud will build a 5GW orbital data center equipped with solar panels and cooling panels about 4 km wide and high. Currently, Starcloud has announced the next phase of the roadmap — October 2026. The next launch will be equipped with multiple H100s at once, and the Blackwell platform will be integrated to improve AI performance. It will also integrate modules from cloud infrastructure company Crusoe to enable customers to deploy and run AI workloads directly from space.
China's recyclable rocket “finale” in December to accelerate the breaking of industrial bottlenecks
According to the 21st Century Economic Report, 2025 is a year in which China's commercial space industry moves from technology accumulation to explosion. This year, multi-model private space companies completed rocket launch tests. With the acceleration of satellite internet networking, the commercial logic gradually became clear, and the pace of development of multi-model new rockets was fully pressed. On December 3, the Suzake-3 Yao1 carrier rocket was launched at the Dongfeng Commercial Space Innovation Test Zone. The rocket stage 2 accurately entered the intended orbit and completed the mission according to the procedure. The mission also carried out first-level rocket recovery verification, but abnormal combustion occurred during the process, and a soft landing on the recycling site was not achieved. Despite the failure of the recycling test, this achievement is still regarded as China's private space forces have accumulated valuable experience in rocket recycling technology.
Following Suzake-3, the latest developments on December 18 show that the Long March 12A reusable launch vehicle, developed by the Eighth Academy of China Aerospace Science and Technology Corporation, is scheduled to take its first flight in Jiuquan in mid-late this month. The intensive connection between the two major missions can be seen as the most iconic advance in China's commercial aerospace in 2025, which is expected to reshape market rules. Previously, most rockets were “disposable”. After the spacecraft entered orbit, the rocket itself could not return to the landing site and be reused. Once recycling technology is implemented, launch costs will be drastically reduced and commercial space profit models will mature.
Related targets
1) Space computing power: Shunhao Co., Ltd. (002565.SZ), Putian Technology (002544.SZ), Maiwei Co., Ltd. (300751.SZ), etc.
2) Rockets: Western Materials (002149.SZ), Aerospace Dynamics (600343.SH), Aerospace Mechatronics (600151.SH), Chaojie (301005.SZ), Sirui New Materials (688102.SH), Platinum (), Feiwo Technology (301232.SZ), Guangzhou Union Airlines (300900.SZ), etc. 688333.SH
3) Satellites: China Satellite (600118.SH), Shanghai Hanxun (300762.SZ), Zhenlei Technology (688270.SH), Aerospace Hongtu (688066.SH), Zhongke Star Map (), Haige Communications (002465.SZ), Gaohua Technology (Dubai), Shaanxi Huada (301517.SZ), Aerospace Electric (002025.SZ), Electronics Digital (), Jiayuan Technology (301117.SZ), Mengsheng Electronics (688568.SH 688539.SH 600850.SH 688311.SH), Zhenyou Technology (688418.SH), Tongyu Communications (002792.SZ), Xinwei Communications (300136.SZ), etc.
Risk Alerts
Downstream demand falls short of expectations; industry competition intensifies; industry development falls short of expectations.