After the unexpected passing of its founder two weeks ago, Chinese AI company SenseTime Group Inc. (00020.HK) seems to have steadied itself amid the initial turmoil, with its shares recovering some lost ground. Investors are now reevaluating the firm’s stability and future prospects in the evolving AI landscape.
On December 17, 2023, SenseTime announced that its founder, controlling shareholder, and executive director, Professor Tang Xiao’ou, passed away due to illness on December 15, 2023. As a result, SenseTime’s stock price initially plummeted significantly the next day but then continued to rebound.
Investor concerns have been alleviated by the stability of SenseTime’s core management team, which continues to oversee daily operations. The equity shareholding structure further reassures that control firmly remains with the founding team. With generative AI applications poised to deepen, SenseTime’s core businesses appear set for significant growth in the coming years.
Around as early as 2019, Tang Xiao’ou had already stepped back from the day-to-day operations of SenseTime to focus on technological research. This allowed SenseTime’s management team to gain experience and not rely on a single leader.
Per Hong Kong Stock Exchange regulations, a major shareholder’s demise results in the automatic conversion of their super voting rights to ordinary voting rights. SenseTime announced that the differential voting rights of the 6.906 billion Class A shares held by Tang Xiao’ou ended on December 15. These shares were subsequently converted into Class B shares on a one-for-one basis, remaining under a voluntary two-year lock-up agreement commencing December 30, 2022.
SenseTime’s Chairman Xu Li, Executive Director Wang Xiaogang, and Xu Bing collectively hold Class A share voting rights amounting to 15.75%. Along with their Class B shares, which account for an additional 20.60% in voting rights, their combined total voting rights reach 36.35%. This ensures that control of SenseTime remains firmly and stably in the hands of the founding team.
Meanwhile, SenseTime is actively advancing and commercializing its large language model (LLM) products. With AI emerging as a national strategic priority in China, SenseTime stands among the foremost Chinese companies at the forefront of the AI surge.
Around 2018, SenseTime began exploring LLMs and initiated research for its Artificial Intelligence Data Center (AIDC). In 2021, SenseTime launched a new AI infrastructure – the SenseTime MegaDevice SenseCore. The AIDC, an important computational base for the SenseTime MegaDevice, started construction in July 2020 and officially began operations in January, 2022. Currently, it can output a computing power of 6000 Petaflops.
In April 2023, SenseTime released its large-scale model system “SenseNova,” one of the first in China to launch a series of LLM products. This includes several models including Chinese language model “ShangLiang,” the text-to-image model “MiaoHua,” the AI digital human video generation platform “RuYing,” the high-precision real-scene 3D reconstruction platform “QiongYu,” and the 3D object generation platform “GeWu.”
In vertical commercialization, SenseTime released an upgraded version of its medical and health large language model DaYi in October. DaYi is based on the “ShangLiang” model, which has a scale of hundreds of billions of parameters and is pre-trained with trillions of tokens.
It is further trained with over 20 billion tokens of high-quality medical knowledge data, including massive amounts of medical textbooks, guidelines, clinical pathways, drug libraries, disease libraries, physical examination reports, and data from 40 million real medical records and doctor-patient Q&As. Currently, DaYi has cooperated with institutions such as the Ruijin Hospital in Shanghai and many other medical institutions in China.
Additionally, SenseTime’s text-to-image model “MiaoHua” has served over 60 enterprise-level clients from sectors including architecture, education, entertainment, gaming, and automotive, with nearly a million users registered.
In December, SenseTime released an intelligent programming assistant based on SenseTime’s self-developed large language model, Raccoon, for public testing. Raccoon covers software requirement analysis, architecture design, code writing, and software testing. With the application of Raccoon, developers can delegate 80% of the writing work to AI, says SenseTime.
AI large-scale models are key drivers for SenseTime’s growth and the company appears on track to further expand this part of its businesses. Its latest financial report shows that SenseTime’s main revenue sources include smart business, smart living, smart cities, and intelligent vehicles, accounting for 59.6%, 21.8%, 12.8%, and 5.9% of the group’s total revenue, respectively.
In the first half of 2023, revenue related to SenseTime’s AIGC (AI-Generated Content) contributed over 20% of the income.
In the first half of 2023, SenseTime’s intelligence automobile platform production SenseAuto achieved a 573% year-on-year growth. Based on the large language model product ‘ShangLiang,’ SenseTime also launched several Chinese interactive products for the automobile sector.
Founded in 2014, SenseTime initially focused on facial recognition, and was used by entities such as the Chinese law enforcement to track down criminals. It raised massive financing to support its rapid growth in the following years and completed an IPO at the end of 2021.