Revenue grew steadily, and the generative AI business contributed 60% of revenue. The company announced its 2024 interim results announcement. With 2024H1, the group's overall revenue increased significantly, Japan's new model capabilities increased by leaps and bounds, model inference costs dropped rapidly, and the number of model calls increased significantly. 2024H1, the group's total revenue was 1.74 billion yuan, up 21.4% year over year. Among the three major businesses, generative AI revenue was 1.051 billion yuan, a sharp increase of 255.7% year on year, accounting for 60.4% of the group's revenue; “Brilliant” smart car business revenue was 0.168 billion yuan, up 100.4% year on year; and traditional AI revenue was 0.52 billion yuan, down 50.6% year on year. At the same time, the company's overseas business also grew steadily, with a year-on-year increase of 40.1%, accounting for 18.5% of the group's revenue. 2024H1, the Group's overall gross profit was 0.767 billion yuan, up 18.2% year on year, and gross margin was 44.1%. While increasing investment in research and development related to large models, the Group actively optimized operating expenses to reduce losses. Sales expenses were 0.329 billion yuan, a decrease of 21.2% year on year, administrative expenses were 0.735 billion yuan, down 7.7% year on year, R&D expenses were 1.892 billion yuan, up 6.1% year on year, and the number of R&D personnel was 3,399. Compared with the same period in 2023, the overall loss for the period decreased by 21.2% year-on-year to 2.477 billion yuan.

Generative AI has become the company's leading business, and AIDC's service share ranks among the top three in the industry. Generative AI ushered in a golden period of development in 2024. Enterprise customers in various industries cooperated with companies to shift traditional computing to intelligent computing worth trillions of yuan. Large models and intelligent computing became new types of productivity, significantly improving the efficiency of various industries and helping customers win revenue growth opportunities in the generative AI era. The company's generative AI business achieved a sharp increase of 255.7% and became the group's dominant business. This stemmed from the strong demand for large models and AIDC in the Chinese market. In addition to the Internet industry, industries such as smart hardware, new energy vehicles, robotics, medical care, and finance are also actively adopting generative AI technology, forming a number of huge vertical markets. The company is well positioned for the next wave of growth. The Japan-Japan Big Model and Large Device AI Cloud Service has been fully put into operation, effectively supporting the training of trillion-parameter multi-modal models and large-scale AI inference. Through an extensive partner network, the company provides comprehensive enterprise-grade generative AI solutions covering the cloud, local data centers and terminal devices. According to a report recently released by IDC, the company's large model platform and application ranked second in the industry with 16% market share, and AIDC services ranked among the top three in the industry with a 15.4% market share.

Shang Tang's strategic goal of transforming and upgrading to generative AI has been achieved in stages. The company's core advantage lies in the deep collaboration of “Shangtang Big Install+Big Model”, which enables the company to quickly iterate the model and reduce inference costs. 2024H1, more than 3,000 leading companies in the industry used Shangtang's big model and AIDC services, and the overall call volume of the “Daily New” big model increased by 400%. As of August 2024, the total operating computing power had surpassed 20,000 PetaFlops, an increase of about 70% compared with March 2024. The number of GPUs launched by the company reached 0.054 million, making it a leading position in the industry. The company has mature capabilities in design, construction, and flexible scheduling and operation from 1000 GPU clusters to 10,000 GPU clusters, and can support stable and efficient training and inference of MoE multi-modal models with up to 2 trillion parameters. In the inference scenario, the company achieved a four-fold increase in requests per second (QPS) under the same computing power and power costs with an innovative technical architecture, and achieved elastic scaling of inference services as needed, and optimized the overall cost of large-scale AI inference. In addition, the company continues to increase the GPU resource utilization rate of the 10,000 Card Cluster to a maximum of 80% based on technological innovations such as integrated training and promotion, computational collaboration, etc., and enhance the overall return on AIDC's operation.

The new big model continues to iterate day by day, and the company's generative AI industries are developing steadily. The Shang Tang Ri-ri Big Model brought together the company's cutting-edge R&D resources. Starting at the end of 2022, it invested more than 10,000 GPUs in computing power resources, and iterated from the 1.0 version to the 5.5 version released in July 2024. The 5.5 multi-modal capabilities of the Japan-Japan New Large Model have been fully enhanced to achieve comprehensive performance and real-time interaction experience against the GPT-4o standard, making it the first multi-modal real-time interaction model in China. The company distilled models of various scales based on Nishi-Nisshin 5.5 to provide comprehensive enterprise-grade generative AI solutions covering the cloud, local data centers, and terminal devices. The company has also expanded from customers in the Internet industry to smart hardware, electric vehicles, robotics, medical care, finance and other industries. The overall call volume of Nishi-Nisshin 2024H1 achieved a 400% increase, and both the number of users and the average number of calls made by users increased dramatically. In the financial field, corporate clients represented by Bank of China, China Merchants Bank, Bank of Ningbo, Bank of Shanghai, and Haitong Securities use the company's financial model capabilities to build financial vertical scenario applications, covering business scenarios such as employee assistants, compliance risk control, programming assistants, digital people, and marketing. In the medical field, large hospitals such as Shanghai Ruijin Hospital, Shanghai Xinhua Hospital, and the First Affiliated Hospital of Zhengzhou University use the company's “Big Doctor” medical model to build the big model as an intelligent hospital center to achieve functions such as intelligent guidance, pre-consultation, medication consultation, post-diagnosis follow-up, and multi-modal data management, to enable patients to seek medical treatment throughout the entire process and specialized research work before and after diagnosis. In the field of Copilot assistants, the Shangtang Code Assistant and the office assistant product “Little Raccoon” have become one of the fastest growing Copilot products in China. Large enterprise users such as China Telecom and Jinshan Office, as well as hundreds of thousands of individual users and developers, have become “Little Raccoon” users. In the field of anthropomorphic interaction, Shang Tang's Anthropomorphic Congress model supports Internet applications such as Weibo, Reading Literature and Building Dream Island, iQiyi, and Idea Flow. User stickiness and number of active users continued to be at the forefront of the industry, and daily calls increased nearly 22 times within half a year. The company's small end-side model maintains the lead in inference speed and model ability, bringing a new user interaction experience on various IoT devices such as mobile phones, cars, personal computers, smart speakers, etc. The inference speed can reach 100 words per second, serving customers such as Xiaomi and OPPO.

Mass production and delivery of Shang Tang Zhuying is progressing rapidly, with the goal of forming a large-scale model for autonomous driving in Shanghai by 2026. Shang Tang Zhuying is positioned as a leader in smart car technology in the era of general artificial intelligence. 2024H1, the smart car business achieved revenue of 0.168 billion yuan, an increase of 100.4% over the previous year. In terms of mass production delivery, the 2024H1 delivered 0.705 million new vehicles, an increase of 80.0% over the previous year. The cumulative number of delivered vehicles reached 2.6 million, covering 104 models. The 2024H1 added 15 new fixed-point models, and added about 6 million fixed-point vehicles, including fixed points from overseas brands such as Volkswagen and Lexus. In addition, Zhuying has also participated in overseas projects for many Chinese car companies, and Zhuying's global influence is becoming more and more prominent. In the field of intelligent driving, Zhuying continues to lead the mass production and development of one-stage end-to-end autonomous driving. With the abundant computing power, high-quality simulation data, and industry-leading large model performance of the Shangtang device, the UniAd end-to-end solution brings a higher limit of intelligent driving capabilities. At the Beijing Auto Show in April 2024, UniAd successfully displayed the actual vehicle. It relied only on visual perception to complete complex live road tests without high-precision maps, and was hailed by the media as “China's own FSD (fully automatic driving)”. The company uses the world model to synthesize multi-view video data to enhance training and testing of the UniAD model. The 2024H1, FAW Hongqi EH7, is the designated model for smart driving, achieved mass production. So far, Zhuying has delivered high-speed pilot functions to 5 models of GAC, Hezhong, and FAW. In June 2024, the Ministry of Industry and Information Technology announced the first batch of nine car companies in the country to launch L3 autonomous driving pilot projects. Two of the nine companies — GAC and FAW — provided L3 smart driving solutions. As a lead unit, Zhuying is also participating in the Shanghai Autonomous Driving Vertical Model Technology Innovation and Application Project initiated by the Shanghai Economic and Credit Commission. The goal is to form an autonomous, controllable, mature and implementable Shanghai Autonomous Driving Vertical Model by 2026, with mass production scale of 10,000 vehicles. In the field of smart cockpits, based on a combination of capabilities such as the Japan-Nisshin multi-modal model and the Wenshengtu model, Zhuying has established cooperation with more than 10 domestic and foreign car manufacturers, including Audi, Volkswagen, Honda, BMW, Xiaomi, SAIC Zhimi, and Geely, to establish large cabin models. Zhuying was the first in the industry to achieve the car-side deployment of the new 5.5 multi-modal model with real-time interaction. The first frame delay was reduced to 60 ms, which significantly improved the real-time experience of human-computer interaction in the car.

Profit forecasting and investment advice. We believe that at present, generative AI is in an important period of rapid development. The company's core advantage is deep collaboration with “big device+big model”, the ability to build first-class large models, and build unique advantages on native multi-modality, natural video stream interaction, and low-cost model inference architectures, so it is expected to stand out from the fierce market competition. Currently, the company's strategic goal of transforming and upgrading to generative AI has been achieved in stages, and the company's long-term development is worth looking forward to in the future. Combining various factors, we gave Shangtang Group a rating of 16-20 times PS in 2024, corresponding to the company's reasonable value range of HK$2.32-2.90 (1 HKD = RMB 0.9135; the exchange rate reference date is August 27, 2024), and “superior to the market” rating.

Risk warning. Risks that AI commercialization falls short of expectations; risks that the company's internationalization falls short of expectations, etc.