SenseTime Announces 2024 Annual Results The Integrated Strategy of “AI Infrastructure – Large Model – Application” Yields Significant Results Generative AI Achieves Triple-Digit Growth for Two Consecutive Years

SenseTime Group Inc. (“SenseTime” or the “Group”, stock code: 0020), a leading global AI software company, announced its annual results for the year ended December 31, 2024 (“2024” or the “Year”).

In 2024, SenseTime’s total revenue increased by 10.8% year-on-year, reaching RMB 3,772.1 million. Its Generative AI revenue exceeded RMB 2,404.0 million, marking a year-on-year increase of 103.1%. This is the second consecutive year that the Generative AI business has maintained triple-digit growth, making it as the Group’s largest business segment. The Group’s gross profit for 2024 was RMB 1,619.7 million with a gross margin of 42.9%. SenseTime remains committed to its integrated strategy of “AI Infrastructure (SenseCore) - Large Model (SenseNova) - Application”. Over the past three years, it has continuously invested in building and operating its Artificial Intelligence Data Center (AIDC) at Lingang, Shanghai, which is also China’s first 5A-rated AIDC. Through its operating model, SenseTime has raised its computing power to 23,000 Petaflops, an increase of 92% year-on-year. By fine-tuning its infrastructure with each iteration of the large model, SenseTime positions itself as the “AI infrastructure that best understands large models”, delivering value not only to its proprietary SenseNova large model, but also widely applied in emerging fields such as embodied AI, AI generated content (AIGC), and AI For Science (AI4S). The positive reinforcement between SenseTime’s AI infrastructure and the algorithms of its large models has helped SenseNova achieve industry-leading efficiency in both training and inference. The Group has deployed industry-leading open-source models, such as DeepSeek V3 and R1, onto SenseCore, with training efficiency exceeding DeepSeek’s public reports. The SenseNova large models, supported by SenseCore, continued to rank among the top in China. The substantial decline in model costs has catalyzed explosive growth in applications. SenseTime focused on developing applications and products in two key areas, productivity tools and interactive tools. Productivity tools directly enhance productivity in business scenarios like enterprise copilot assistants, finance and public services. The aggregate amount of order value for productivity tools, which represents the customers’ willingness to pay, increased by sixfold compared to 2023. Interactive tools empower enterprise clients and customers with a 2B2C strategy, enhancing user experience and fulfilling user demands in intelligent companionship, smart hardware interactions, and intelligent marketing, with monthly user activity increasing by eightfoldcompared to 2023.

Dr. Xu Li, Chairman of the Board and CEO of SenseTime, said, “The positive reinforcement between large model algorithms and AI infrastructure systems has become the core driving force for the rapid development of Generative AI. This positive reinforcement, coupled with SenseTime’s integrated strategy of ‘AI Infrastructure – Large Model – Application’, are deeply aligned with this trend, positioning us for a phase of accelerated growth. We have strategically focused our resources as we transitioned to the ‘1+X”’organizational structure, emphasizing sustainable growth and profitability at our core, while strengthening our position as an industry leader.”

Key Highlights of the Group’s Business Operations in 2024

- Revenue reached approximately RMB 3.8 billion, representing a year-on-year growth of 10.8%. Gross margin stood at 42.9%, with net loss narrowed by 33.7% year-on-year. As of December 31, 2024, the Group’s total cash reserves reachedapproximately RMB 12.8 billion.

- Generative AI revenue reached approximately RMB 2.4 billion, achieving a year-on-year increase of 103.1% and maintaining triple-digit growth for the second consecutive year. The proportion of Generative AI in the Group’s revenue further increased from 34.8% in 2023 to 63.7% in 2024. Generative AI applications achieved significant adoption, with productivity toolsreaching millions of users and order value increasing sixfold year-on-year. Interactive tools achieved an eightfold year-on-year increase in user activity.

- SenseCore’s computing power grew by 92% year-on-year. The total computing power operated by SenseCore has exceeded 23,000 Petaflops, with the initial rollout across China completed.  Chinese chips have demonstrated outstanding compatibility, while various initiatives have significantly reduced the costs of large model training and inference, achieving industry-leading training efficiency. 

- The transition to the "1+X" organizational structure delivered strong results, with five ecosystem companies securing external capital between late 2024 and early 2025.

Business Highlights

Generative AI: Seamless Integration of the “AI Infrastructure – Large Models – Applications” Strategy, Maintaining Leading Market Share and Technological Capabilities in China

SenseCore expanded scale in terms of computing power and operation capabilities with a combination of proprietary ownership and operating model. the Shanghai Lingang AIDC was awarded China’s first 5A-rated intelligent computing center certification issued by the China Academy of Information and Communications Technology, demonstrating SenseTime’s market leadership position in operating large-scale AIDCs. AI computing nodes have been deployed across China, covering key areas including the Yangtze River Delta, the Greater Bay Area, the Beijing-Tianjin-Hebei region, and the central and western regions of China, achieving nationwideunified scheduling of computing power. For its AI cloud service capabilities, SenseTime has enabled the rapid creation and scaling of virtual AI clusters through a series of engineering optimization, achieving speeds six times faster than the industry average and more than 90% savings on virtual cluster management costs. The Group has achieved the large-scale commercial deployment of Chinese chip clusters. This enables the scheduling and execution of a single large-model training task across a 5,000-GPU heterogeneous chip cluster, achieving 95% of the efficiency of homogeneous training. The SenseCore AI cloud services continued to rank among the top three by IDC in the “China AI Computing Services Market Tracking Report (H1 2024)”.

The Group made significant breakthroughs in the efficiency of large-model training and inference, reflecting the advantages of model and infrastructure collaboration. Through automatic multi-dimensional parallelism strategies and optimization of communication and operators, SenseCore’s training efficiency has been significantly improved, and FP8 mixed-precision training has been successfully achieved. For third-party models such as DeepSeek V3, SenseTime has processed over 1,600 tokens per second (TGS), outperforming DeepSeek's official results without dedicated optimization; it is expected that specific optimization of a single model in the future will bring greater performance improvements. In terms of inference, SenseTime supports low-bit quantization inference of both open source and proprietary engines. Latest internal tests show that the inference capability of the proprietary multimodal model is 50% higher than that of SenseNova 5.5 released in July 2024. For the token generation time of the DeepSeek R1 model, it is 15% more efficient than the leading suppliers, with industry-leading performance under high service standards (SLA) conditions.

The SenseNova large model series was the first large model in China to surpass GPT-4 Turbo in terms of performance and was also the first to introduce audio-visual streaming interaction. SenseNova 5.5 ranked first in China on both the SuperCLUE and OpenCompass benchmarks, outperforming Gemini 2.0 Flash-Exp and GPT-4o. SenseNova 6.0 is set to launch on 10 April 2025, with performance expected to benchmark to Gemini 2.0 Pro.

Leveraging its deep expertise in enterprise-level applications, SenseTime has established the “One Foundation, Two Wings”application strategy. The commercialization of productivity tools has delivered strong results, significantly enhancing efficiency in scenarios such as enterprise copilots, financial insight agents, and public service assistants. SenseTime’s customers include leading enterprises such as China Mobile, China Telecom (Shanghai), Kingsoft Office, Ningbo Bank, China Merchants Bank, Haitong Securities, Lenovo, 360, and Leapmotor, with the aggregate amount of order value, representing customers’ willingness to pay,increasing by sixfold year-on-year in 2024. Among these, penetration of office copilot and coding copilot are accelerating, with number of users exceeding 1.5 million, the number of daily API calls reaching millions of times, the number of tokens processed daily exceeding 3.5 billion, and reaching 10 million developers, making it the fastest growing AI office tool in China. SenseTime’s human-like companionship model ranks among the top in deep conversation engagement rate and paid conversion rate, with the monthlyAPI call volume of a top customer increasing by 833% over the past year, with a next-day retention rate of 50%, and an average daily time spent of 130 minutes. In smart hardware interaction, the Group is accelerating the deployment of multimodal large models across robots, smart glasses, smart cabin and other smart hardware. Since the beta testing, the system has already been integrated by over 70 enterprises. 

SenseTime’s multimodal large model combination with digital human technology empowers e-commerce marketing, with the duration of AI-generated videos for smart marketing increased by 805% in 2024, leading to a 94% year-on-year growth in Gross Merchandise Value (GMV). According to IDC’s “2024 Large Model Application Market Share Report”, SenseTime was ranked among the top three in China’s large model application market with a market share of 13.8%, trailing only Baidu and Alibaba Cloud.

Computer Vision (Traditional AI): Maintaining Leadership of General Vision Model, While Deeply Engaging in Premium Businesses and Expanding Globally

In the field of Computer Vision, SenseTime continues to be deeply engaged in mature industries with a strong differentiation, high profit margins and abundant cash flow in the Chinese market, while actively expanding its overseas markets. The Group is also fostering a wider network of overseas partners and a more sustainable growth model. In 2024, the customer repurchase rate increased by 31 percentage points, while number of multi-year partners (>3 years cooperation) reached 389, representingapproximately 40% of the total Computer Vision clientele. SenseTime has held the number one market share in China’s computer vision market for eight consecutive years, and its Computer Vision project won the first prize of science and technology excellence in national ministry- level evaluations in 2024. 

Smart Auto: SenseAuto Develops an Advanced Smart Pilot System for All Scenarios and an End-to-End Smart Pilot Product Ecosystem, Achieving Breakthroughs in Mass Production 

Leveraging its “Smart Pilot - Smart Cabin – Cloud” product system and strategic roadmap, SenseAuto continues to strengthen its differentiated capabilities in the smart auto sector, building on SenseTime’s capabilities in Computer Vision, multimodal large models and SenseCore. In 2024, SenseAuto’s new vehicle deliveries exceeded 1.67 million vehicles, a year-on-year increase of 29.2%, with 42 new models. The cumulative deliveries reached 3.6 million vehicles across over 130 models. SenseAuto also received confirmation letters as the designated supplier for 41 models and 11 million vehicles. SenseAuto achieved breakthrough in mass production and maintains technological leadership in high-end Smart Pilot solutions with its unified end-to-end architecture. SenseAuto has ranked first in the “Smart Cabin Computer Vision Software Market Share” for five consecutive years. SenseAuto’s “A New Member For U”, the industry’s first Smart Cabin interactive product based on the multimodal large model, has secured a hundreds-of-millions RMB contract from a top EV manufacturer. SenseAuto also launched the “Kaiwu” World Model, which set an industry landmark and has secured first-place victories for two consecutive years since 2023 in globally renowned autonomous driving simulation competitions.

Corporate Social Responsibility

SenseTime Group consistently upholds green development principles, prioritizing societal needs while leveraging AI technology to foster a sustainable ecosystem. The Group used AI innovative technology to complete an environmental, social and governance (ESG) project for Saudi Arabia’s Public Investment Fund in the Middle East Red Sea region. The Group used underwater IoT technology to monitor the underwater ecosystem in real time and accurately assess the level of underwater pollution and the health of wild animals. The Group is committed to build a low-carbon green future to achieve sustainable development goals.

Strategic Outlook

SenseTime will continue to reinforce the in-depth synergy in its “AI Infrastructure – Large Model – Application” integrated strategy, maintaining its leadership in multimodal large models, and promote breakthroughs in commercialization. The Group will continuously enhance its Computer Vision business model, focusing on mature use cases and premium clients. SenseTime will also further unlock the commercial viability of its ecosystem companies, to unleash new value from synergies.