The two major sectors of smart life and smart cars became revenue growth points for the company. The company discloses its 2022 annual report.

In 2022, the company achieved revenue of 3.8 billion yuan and gross margin of 67%. The financial level remained stable, with net losses narrowed to $6.1 billion; adjusted net losses amounted to $4.7 billion; and total cash reserves amounted to $16.6 billion. In 2022, the company actively promoted business structure transformation, and revenue growth in the two major sectors of smart life and smart cars accelerated. In terms of the smart living sector, revenue reached record highs. In 2022, the smart living sector's revenue increased 130% year over year to 1 billion yuan; the revenue share of the group increased from 9% in 2021 to 25% in 2022. The number of customers in the smart lifestyle sector was 317, an increase of 23% over the previous year, and single customer revenue increased 86%. The four products of smart life AIGC, AI sensors, AI ISP chips, and smart medical care have achieved a breakthrough in commercialization. In the field of smart phones and smart terminals, the number of newly mass-produced smartphones equipped with the Shangtang AI SDK reached 450 million units, and the super-sharing/portrait series function achieved the highest market share. The company delivered 6 AI sensors with leading performance. RGBW AI sensors were mass-produced and marketed on vivo's flagship aircraft, while 2x2OCLAI sensors have already been delivered to one of the world's leading smartphone manufacturers, and mass production is expected to reach the level of hundreds of millions.

In addition, the company's first AI-ISP chip successfully illuminated, enabling AI to process 4K ultra-high quality video using 50% less power than similar products, and improving video quality and resolution. In the smart car sector, the share of mass production revenue of smart car front mounts has increased, and the value of bicycles has improved markedly. In 2022, the smart car sector achieved revenue of 300 million yuan, an increase of 59% over the previous year. The proportion of the group's total revenue increased sharply from 4% in 2021 to 8%, and the average value of a bicycle improved significantly. In 2022, Genying's smart cabins and intelligent driving products completed the adaptation and mass production delivery of 27 models, mass-produced more than 500,000 vehicles, added more than 8 million new designated vehicles, covered more than 80 models from more than 30 car companies, and continued to lead the industry. “Genie” launched the AI Cloud Platform - Jeongying Vehicle Road Collaboration Platform, integrating the “Car-Road-Cloud” comprehensive product matrix, and achieved the first launch of a variety of innovative functions in the industry within 2022. We believe that Shang Tang's outstanding performance in the smart life and smart car sector means that the company's current revenue structure is getting healthier, and the company's overall development is expected to continue to improve.

Smart commerce focuses on the transformation of SenseCore's external service business, and AIaaS (AI-as-a-Service) revenue has increased. The Shangtang SenseCore AI device continued to expand in 2022, completing the deployment of 27,000 GPUs and achieving 5.0 exaFLOPS computing power output capability, providing services to industry partners using the AIaaS (AIaaS) model. In 2022, SenseCore officially opened up capabilities to industry customers, empowering model development for industry customers. The revenue generated by external services quickly exceeded 20% of the overall revenue of smart business, and the scale of revenue generated by SenseCore's external services will increase dramatically in 2023. SenseCore has supported more than 10 large-scale model training projects, including self-developed models such as language models, graphic models, large visual models, multi-modal models, etc. Among them, SenseCore serves 8 customers for large model training, providing more than 7,000 GPU cards in total; AaaS provided to the outside world specifically includes the three capabilities of computing infrastructure services (IaaS), deep learning platform services (PaaS), model deployment and inference services (MaaS). We believe that smart commerce has achieved good results by focusing on the transformation of SenseCore's external service business. With the development of the AI industry, demand for big model training is also expected to continue to grow, and AIaaS revenue is also expected to continue to rise.

The AI big model is developing smoothly, and the multi-modal big model is the company's strategic investment direction. The company maintains leadership in high-performance computing systems and big model training capabilities, and continues to make breakthroughs in the five major technology fields of computer vision, natural language processing, speech recognition and synthesis, decision intelligence, and AI chips, laying a solid foundation for achieving stronger general artificial intelligence capabilities. In terms of computer vision, the company successfully developed the world's largest general vision model with 32 billion parameters, achieved high-performance object detection, image segmentation, and multi-object recognition algorithms, and was widely used in various fields such as autonomous driving, industrial quality inspection, and medical imaging. In terms of AI-generated content (AIGC), training on a number of large models of Wensheng images has been achieved. They have outstanding abilities for tasks such as understanding text images, drawing quality, and reasoning speed. In particular, they have unique advantages in understanding traditional Chinese culture. In terms of natural language processing, a large pre-training model with 100 billion parameters was successfully developed, which significantly improved the performance of tasks such as text generation and human-computer dialogue. In terms of speech recognition and synthesis, the company's speech recognition model achieves speech-to-text capability with high accuracy and low latency in various scenarios; the speech synthesis model can generate speech with different intonations and sounds according to user needs, providing strong support for the company's digital people, smart cabins and other products. In terms of enhanced learning and intelligent decision-making, the company has successfully developed a series of efficient intelligent decision-making models, which have achieved remarkable application results in fields such as robot control, game AI, and autonomous driving decision systems. In terms of AI chips and hardware, the company has successfully developed several high-performance, low-power AI inference chips and AI ISP chips, showing strong computing power in scenarios such as data centers and edge computing. Furthermore, Shangtang's SenseCore AI device can currently support simultaneous training of up to 20 models with large numbers of 100 billion parameters (in parallel with kilocalories), and can support the training of models with up to trillion parameters. Judging from the strategic outlook, in the future, the company will take the development of multi-modal large models as a strategic investment direction. We believe that Shangtang is leading the industry in accumulating AI technology, and the company has taken the development of multi-modal large models as a strategic investment direction. Against this background, the development of the company's multi-modal large model is also expected to continue to accelerate.

The “New SenseNova” big model system system was announced, the Shangtang language model “Discuss SensecHat” was released, and MaaS services continue to empower the AI industry.

On April 10, 2023, the company held a technical exchange day to share the strategic layout of advancing the development of AGI (general artificial intelligence) with “big model+big computing power”, and announced Shang Tang's “New SenseNova” big model system under this strategy, and launched various big models and capabilities such as natural language processing, content generation, automated data annotation, and custom model training. At the event, Shang Tang not only showcased the big language model under the “New SenseNova” model system, but also showcased a series of generative AI models and applications such as AI graphic creation, 2D/3D digital creation, and big scene/small object generation, and unveiled Shang Tang's R&D system that relies on the AI device SenseCore to achieve “big model+big computing power” integration and innovation. Currently, the industry is training big models in high demand for big computing power, and really easy to use infrastructure is scarce. It took five years for Shang Tang to build an industry-leading AI device. The device has a total of 27,000 GPU chip cards, which can output a total computing power of 5.0 exaFLOPS. It is currently one of the largest intelligent computing platforms in Asia. Based on the capabilities of large devices, Shang Tang has now constructed large models in various fields such as computer vision, natural language processing, AI content generation, multi-modality, and decision intelligence, with the ability to continuously evolve and support a wide range of applications.

“New SenseNova” provides government and enterprise customers with a variety of flexible API interfaces and services, including image generation, natural language generation, and general visual perception tasks and labeling services. According to actual application requirements, customers can use the various AI technical capabilities of the “Nichirishin SenseNova” big model to achieve various AI applications at a low threshold, low cost, and high efficiency.

“Nichinishin SenseNova” has also brought many breakthroughs to Shang Tang's own business. For example, in the field of intelligent driving, based on a large-scale visual model, Shang Tang realized mass production of a BEV that can recognize 3000 types of objects with a general sensing algorithm for circling around, and also constructed a multi-modal model for autonomous driving integrating sensing and decision-making, bringing stronger environmental, behavioral, and motivational decoding capabilities.

Natural language is a key means of human-computer communication, and “NichiRishin SenseNova” also introduced Shang Tang's newly developed large-scale language model “Discuss SensecHat”.

As a natural language processing model with 100 billion parameters, “Consultation SensecHat” uses extensive data training, fully considers the Chinese context, and can better understand and process Chinese text. At the event, “Discuss SensecHat” demonstrated excellent ability to understand multiple rounds of conversation and extremely long texts. Shang Tang also showcased several innovative applications supported by big language models, including: programming assistants that help developers write and debug code more efficiently; health consulting assistants that provide users with personalized medical advice; and PDF file reading assistants that can easily extract and summarize information from complex documents.

The diffusion model sparked the popularity of AIGC applications. Shang Tang demonstrated a series of generative AI models and applications such as “Nichinishin SenseNova” for various AI graphic creation, 2D/3D digital creation, and large scene/small object generation:

The “Photograph SenseMirage” graphic creation platform showcases powerful literary drawing capabilities with realistic lighting and shadows, rich details, and diverse styles. It can support the generation of 6K high-definition images; customers can also train and generate models according to their own needs.

The “SenseAvatar” AI digital human video generation platform takes only 5 minutes of live video footage to generate digital human aliens with natural sounds and movements, accurate lips, and proficiency in multiple languages.

The “Qiongyu SenseSpace” and “Grid SenseThings” 3D content generation platforms can efficiently and at low cost generate large-scale 3D scenes and detailed objects, opening up a new space of imagination for metaverse, virtual and reality fusion applications.

These powerful and easy-to-use content generation capabilities brought by “New SenseNova” will change the production paradigm of the content production industry, break through the ceiling of content creativity, reshape the content production industry ecology and open up new growth space.

Whether it's a big language model, a literary diagram, or a digital person, it's all inseparable from the computing power support of large-scale AI infrastructure. Shangtang AI's big device SenseCore has industry-leading computing power output capabilities, oversized model training, and large-scale reasoning capabilities, and will become an infrastructure service leader in the AGI and big model era.

Based on the AI big device SenseCore and the “New SenseNova” big model system, Shang Tang provides industry partners with a variety of big models as a service (Model-as-a-Service) covering automated data labeling, custom big model training, model incremental training, model reasoning and deployment, and development efficiency improvement:

Automated data labeling based on pre-trained large models can achieve an efficiency improvement of nearly 100 times compared to manual data labeling.

Large model parallel training and model incremental training services can help customers quickly train models using their own data, including developing vertical industry models on pre-trained large models and producing thousands of custom models.

The model inference deployment service can increase the inference efficiency of large models by more than 100% and reduce the cost of using models to provide services.

Shang Tang is also opening up a large number of pre-trained models and AI development toolchains to industry developers, fully empowering customers to improve development efficiency.

Shang Tang will continue to push forward the construction of the “Nichi-Nichi SenseNova” big model system. “New day by day, new day by day”, Shang Tang's model development hopes that it can improve day by day in terms of data volume, parameter structure, and issues that can be processed, and work with the industry ecosystem to usher in a breakthrough in stronger AGI technology, so that AI technology can benefit everyone. We believe that the launch of Shang Tang's “New SenseNova” big model system has given full impetus to the development of its own big model and the development of the entire AI big model industry. “Discussing SensecHat” as a natural language processing model with 100 billion parameters further confirms Shang Tang's leading technical strength in this field. Combined with the continuous promotion of Shangtang's MaaS services, it is expected that the development of China's AI industry will continue to benefit.

Profit forecasts and investment recommendations. We believe that as a leading enterprise in the AI industry, Shangtang's overall business development continues to improve. Among them, the smart life and intelligent driving business foundation is already strong, it has entered the fast track of rapid growth, and its revenue is naturally highly sticky. On the other hand, Shang Tang has a gross profit margin of up to 60-70%, and as the commercialization process of Shangtang AI continues to advance, the company's gross margin is expected to continue to increase. At the same time, the development of the big model of Shang Tang is progressing smoothly, and related models continue to be launched, with the support of the “New SensenOVA” big model system. The development of Shangtang is expected to continue to accelerate. Combining various factors, we gave Shangtang Group 23-25 times PS in 2023, corresponding to the company's reasonable value range of HK$4.05-4.58 (HK$1 = RMB 0.8767), giving it an “superior market” rating.

Risk warning. Risks of AI commercialization falling short of expectations; risks of a company's internationalization expansion falling short of expectations, etc.