Apply for Trial
Newsroom

SenseTime Group Announces 2022 Annual Results

2023-03-28

Artificial General Intelligence (AGI) Era Begins

Smart Auto and Smart Life Recorded High Growth


[Hong Kong, 28 March 2022] – SenseTime Group Inc. ("SenseTime", or the "Company", stock code: 0020), a leading global Artificial Intelligence (AI) software company, today announced its audited results for the year ending 31 December 2022 (“2022” and “the Year”).

 

In 2022, the Company made a series of impressive achievements in R&D. To support artificial general intelligence (AGI) industry’s development, SenseTime officially offered SenseCore’s services to industry customers, assisting them to efficiently train large models. During the Year, the Company expanded its AI applications and strategically adjusted its income mix as revenue from Smart Life and Smart Auto segments increased significantly by 130% and 59% respectively, which also substantially grew as a proportion of total revenue. Moreover, four products of Smart Life: AIGC, AI sensor, AI ISP chip and Smart Healthcare, achieved commercial breakthroughs. The proportion of pre-installed mass production revenue from Smart Auto increased and so has single vehicle value. Simultaneously, the Smart Business and Smart City segments have initiated business transformation. Smart Business segment focuses on commercialization of SenseCore and increasing AIaaS revenue while the Smart City segment focuses on serving top customers and optimizing the quality of revenue, as it steadily promotes the business development.

 

The Company maintained a leading position in high-performance computing systems and large model training capabilities and has continued to make breakthroughs in the five major technical fields: computer vision (CV), natural language processing (NLP), speech recognition and synthesis, decision intelligence and AI chips, laying a solid foundation for more advanced AGI capabilities.

 

Dr. Xu Li, Executive Chairman of the Board and Chief Executive Officer of SenseTime, said, “In 2022, artificial general intelligence (AGI) made remarkable progress, driven by large computing power and large models. SenseTime has always been forward-looking, focusing on the research and development of large models and actively investing in the construction of high-performance computing infrastructure for artificial intelligence (AI). Facing the future, we will focus on AGI as our core development strategy, with a view to achieving major breakthroughs over the next few years. We believe AI is a vast frontier with endless possibilities. SenseTime will, as always, firmly invest in the cutting-edge R&D and commercialization of AGI, as it makes the greatest contribution to the development of AI+ society.”

 

Core Business highlights in FY2022:

● Revenue reached RMB3.8 billion and gross profit margin recorded 67%.

● Revenue growth from two core business segments accelerated. Smart Life segment recorded a year-on-year (“YoY”) increase of 130% while Smart Auto segment achieved a YoY increase of 59%.

● The proportion of revenue to the Company generated from Smart Life and Smart Auto increased significantly, accounted for 25% and 8% of total revenue respectively in FY2022 as compared with 9% and 4% in F2021.

● SenseTime continuously leads China’s AI Industry with its comprehensive R&D, R&D expenditure was RMB 4 billion in FY2022, representing a YoY increase of 11% while per capita R&D efficiency continued to improve by 90% YoY and the average number of models produced by each R&D personnel per year increased to 9.35. The cumulative number of commercialized models increased by 93% to 67,000.

● The financial status remained stable. The net loss improved to RMB6.1 billion and the adjusted net loss was RMB4.7 billion. The total cash resources were RMB16.6 billion.

 

Breakthroughs in Core Technologies in FY2022:

● SenseCore and high-performance computing: SenseCore achieves computing power output of 5.0 exaFLOPS based on a parallel computing system with more than 27,000 GPUs, supporting up to 20 large models (paralleled operation with thousands of cards) with parameters of hundred-billion level for simultaneous training.

● Computer vision: the Company successfully developed the largest CV generic model in the world with 32 billion parameters, enabling high-performance object detection, image segmentation and multi-object recognition algorithms, which are widely used in fields such as autonomous driving, industrial quality inspection and medical imaging.

● AI-generated content (AIGC): the Company achieved training on multiple text-to-image large models, with advanced capabilities such as understanding text to generate graphics, high image quality and low inference speed. In particular, it has unique advantages in the understanding of traditional Chinese culture.

● Natural language processing: the Company successfully developed a 100 billion parameters pretrained large language model, which significantly improves performance in tasks such as text generation and human-computer dialogue. This model and its capabilities will be commercially available by mid-2023.

● Speech recognition and synthesis: the Company’s speech recognition models achieved high accuracy and low latency speech-to-text conversion in a variety of scenarios. The speech synthesis models can generate speech with different tones and timbres according to user needs. These technologies provide strong support for the Company’s digital human and Smart Auto Cabin products.

● Reinforcement learning and decision intelligence: the Company successfully developed a series of efficient decision intelligence models, which have achieved significant application results in areas such as robot control, game AI and decision-making systems for autonomous driving.

● AI chips and hardware acceleration: the Company successfully developed several high-performance, low-power AI inference chips and AI ISP chips, which have powerful computing capabilities in data centers, edge computing and other scenarios.

 

Summary for Business Reviews

“Large model + AI infrastructure” lays a solid foundation, SenseCore empowers the AGI’s development

● SenseCore continued to expand, completing the deployment of 27,000 GPUs and achieving a computing power output capability of 5.0 exaFLOPS. It provided services for industry partners in the form of AI-as-a Service (AIaaS). Currently, SenseCore supports simultaneous training of 20 large models with hundreds of billions of parameters (paralleled operation with thousands of cards), and can support the training of super large models with trillions of parameters.

● With the support of SenseCore, the Company realized 10+ large model training projects, including large language models, AIGC models, large CV models, large multi-modal models and other proprietary and customized models.

● SenseCore supported 8 key customers to for their customized large model training tasks with 7,000 GPUs. The AIaaS provided by SenseCore to external users specifically includes three major capabilities, including infrastructure-as-a-service (IaaS), platform-as-a-service (PaaS), and model-as-a-service (MaaS).

 

Actively adjusted its business mix as revenue growth from Smart Life and Smart Auto accelerated

● The revenue generated from Smart Life segment reached a historical high: in 2022, revenue from Smart Life increased by 130% YoY to RMB955 million. The proportion of revenue to the Company expanded from 9% in FY2021 to 25% in FY2022. The number of Smart Life customers reached 317, representing a YoY increase of 23% and revenue per customer improved by 86%.


In 2022, the Company comprehensively laid out pre-trained large models and developed full capabilities for content generation covering text, speech, image, video, and 3D scenes. The Company also upgraded the SenseMARS mixed reality platform products and functions to serve industry clients and fuel revenue growth. The AIGC-related ability of SenseMARS empowers more than 200 APP such as Xiaohongshu, Weibo, bilibili, ZEPETO, and Line, and has a renewal rate of over 95% for top customers. The expanded offline scenes cover an area of 15 million square meters, including top scenes such as  large amusement parks, shopping malls, museums, banks etc. According to Frost & Sullivan’s “China Digital Human Market Observation Report 2022 – Milestones”, SenseTime’s digital human products have entered the “mature stage”, making the Company a market leader. The Company ranked first in overall competitiveness as it received six top scores in ten evaluation indicators.


In smartphone and smart devices, the number of mass-produced smartphones equipped with SenseTime AI SDK reached 450 million units, and its super resolution and portrait series function ranked first in terms of market share. The Company also delivered 6 AI sensors which deliver leading performance. The Company’s RGBW AI sensor is mass-produced on vivo’s flagship smartphone, and the 2x2 OCL AI sensor have been delivered to a leading smartphone manufacturers, with expected mass-production orders in the tens of millions. In terms of smartphone ISP chip development, the Company successfully developed its first AI-ISP chip, which reduces power consumption by 50% for AI processing 4K UHD video versus comparable products while providing high-quality and high-resolution video quality.


SenseTime maintained its development layout and exploration on Smart Health and consumer-grade products. In the medical segment, SenseTime Smart Health generated significant revenue growth in 2022 by providing a variety of intelligent products and services to hospitals. So far, SenseCare intelligent diagnosis and treatment platform has obtained a total of 22 domestic and foreign medical device certifications, including NMPA Class III and Class II medical device registrations, CE of European Union, Singapore, Saudi Arabia and UAE.  During the year, it  was installed in more than a hundred hospitals and medical institutions to assist doctors in analyzing tens of millions of medical examinations. Additionally, the home consumer AI products, SenseRobot chess-playing robot, was released in August 2022 and quickly became the top selling smart robot product in Tmall and JD.com. It has won numerous awards such as “Product Innovation Exploration Brand”, “2022 Annual Intelligent Hardware Innovation Award” and “Top 10 Best Products of Artificial Intelligence”.

 

● Significant increase in pre-installed mass production revenue from Smart Auto Segment: In 2022, the Smart Auto segment achieved revenue of RMB293 million, representing a year-on-year increase of 59%. Its proportion to the Company’s revenue increased from 4% in FY2021 to 8% in FY2022 and the single vehicle value has improved.


SenseAuto Cabin and SenseAuto Pilot products completed adaptation and mass production delivery of 27 models. More than 500,000 vehicles that are equipped with SenseAuto technology have been mass-produced and put on the road, with more than 8 million new pipelines, covering more than 80 car models of more than 30 automotive companies and continued to lead the industry. SenseAuto also released an AI cloud platform – Vehicle-to-Road Collaboration Platform which integrates the "vehicle-road-cloud" all-round product matrix. Additionally, during the Year, the Company debuted a variety of innovative functions that advance the application of AI within the industry.

 

● Smart Business focused on the commercialization of SenseCore AI capabilities: in 2022, SenseCore officially offered its services to industry customers, empowering their model development. It generated revenue from external services, exceeding 20% of the overall revenue of the Smart Business segment. The revenue generated by SenseCore is expected to increase significantly in 2023. During the Year, 38% of the Company’s total revenue came from the Smart Business segment, representing a decrease as compared with 42% in FY2021. However, more than 110 customers were Fortune 500 companies and listed companies, demonstrating the Company’s strong competitive position.

 

● Smart City maintained its leading advantages in industry: in 2022, the Company adjusted its Smart City business to focus on serving top customers, and continuously optimized the business’ cash flow to address the impact of Covid-19.


Against this backdrop, revenue from Smart City segment accounted for 29% of the Company’s total revenue, which is lower than in FY2021. According to IDC MarketScape: China Smart City Computing Platform Vendor Assessment, 2022, SenseTime was shortlisted as a leader and ranked first in terms of strategic capability. As of December 31, 2022, SenseFoundry has incorporated around 35,000 AI models, increased by 56% YoY.  With China’s support for the digital transformation and upgrading of cities, a total of 162 cities have deployed SenseFoundry, including 16 large cities with over 10 million population, and 4 overseas cities.


AI governance and information security

Advocating the development of “responsible and assessable” AI and proactively promoting “Trusted AI”

 

● In August 2022, the SenseTime AI training platform (AutoLink) won the “Trusted AI Platform Application Benchmark Case” award by the AI Industry Development Alliance, and SenseTime’s ethical governance system and algorithm evaluation tool also won the “Trusted AI Practice Excellence Case” award.

● “Trusted AI Promotion Plan” released by China Artificial Intelligence Industry Development Alliance (AIIA) and the AI Key Technology and Application Evaluation Laboratory, SenseTime was selected as one of the “Top 10 Enterprises with Outstanding Contribution in 2022”.

● SenseAuto received ISO/SAE 21434:2021 Automotive Information Security Process Certification in December 2022, following the ISO 26262 Functional Safety ASIL B product certification and ASPICE CL2 certification in July 2020. SenseTime is one of the few companies in the AI industry that has obtained all three certificates.

 

Strategic Outlook

Facing the future, with AGI being the core of SenseTime’s development strategy, the Company looks to continue achieving major breakthroughs in AGI technology over the next few years. In addition, the Company will further research and explore the following key areas to expand AGI’s core capabilities:

 

1. Large computing power: SenseTime will use SenseCore as the core platform for large model production, continue to expand its capabilities, including investing in high-performance computing infrastructure such as high-speed networks, large-scale data storage and powerful computing nodes.

 

2. Large multimodal model: These models handle multiple data types, including text, images, audio and video, and have stronger data understanding and generation capabilities, as well as multi-task generalization capabilities. SenseTime uses the R&D of large multimodal models for its strategic investment direction.

 

3. AI-specific hardware plus chips plus open research cooperation: SenseTime will continue cooperating with the world’s leading chip manufacturers, global research institutions, universities and enterprises to jointly explore new paths to optimize computing power and promote the development of AI computing power.

 

4. Establish and implement “AI for All”: SenseTime will ensure the development of AI technology benefits all mankind as a whole, strive to reduce the cost and threshold of AI technology, as well as will work with global AI researchers and developers to advance the development of AI technology through open source projects, shared resources, and collaborative research.

 

5. Promote the empowerment of SenseCore and large model capabilities to four major sectors: by making full use of the large models’ capabilities, the Company will continuously upgrade and strengthen the product matrix of its four business segments, improve product competitiveness, and increase the application of AI via commercialization.