The latest financial report of SenseTime (0020.HK) reveals a signal: the AI industry revenue explosion may be just around the corner!
We have explored three perspectives in the financial report to analyze the development trends of the AI industry and the strategic layout directions of industry leaders.
1) Performance has experienced “explosive growth”, with generative AI revenue soaring by three digits, and the pace of commercialization has increased again;
2) The strategic “trinity”, AI leading ecological layout of “large device-large model-application”, indicates the future industry competition situation;
3) With a market of “10 trillion-level blue ocean”, DeepSeek is opening the prelude to the leapfrog development of China’s AI industry
01
AI performance: Generative AI customer list shines, revenue doubles to over 2.4 billion
SenseTime released its 2024 annual performance report, in which the generative AI business exceeded 2.4 billion, a year-on-year increase of 103.1%, achieving triple-digit explosive growth for two consecutive years, accounting for 63.7% of the group’s revenue. In 2024, SenseTime Group’s gross profit was 1.62 billion yuan, with a gross profit margin of 42.9%, and its loss narrowed significantly by 33.7% year-on-year. Cash reserves were 12.75 billion yuan.
SenseTime’s financial report has already highlighted the point where AI industry revenue will explode. Strengthening technology iteration capabilities, accelerating commercialization, and reducing costs are all driving the entire AI industry to move faster. As the leader of the AI industry, SenseTime has delivered phased development results with solid data.
It is worth noting that the proportion of SenseTime’s generative AI business in the group’s revenue has increased from 10% in 2022 to 64% in 2024, an increase of more than 50% in just a short period of time. At the same time, the financial report shows that SenseTime, relying on its experience in enterprise-level applications, has built a “one base and two wings” layout with a daily new multi-modal large model as the cornerstone and “productivity tools” and “interaction tools” as the vertical application directions. The application has achieved remarkable results. In the “2024 Big Model Application Market Share” report released by IDC, SenseTime ranks among the top three in China’s big model application market with a market share of 13.8%, second only to Baidu and Alibaba Cloud.
The productivity tools have reached tens of millions of users, with order value increasing sixfold year-on-year. Its customers include China Mobile, Shanghai Telecom, Kingsoft Office, Bank of Ningbo, China Merchants Bank, Haitong Securities, Lenovo, 360, Leapmotor and other leading corporate clients in six industries.
In terms of interactive tools, user usage has increased 8 times year-on-year, helping customers to provide personalized services and content to C-end users and enhance user stickiness. According to SenseTime’s performance announcement, the monthly call volume of a leading anthropomorphic companion APP in its anthropomorphic interaction service has increased by 833% in one year, with a next-day retention rate of 50% and an average daily usage time of up to 130 minutes. SenseTime is accelerating the application of multimodal large models in smart hardware such as robots, smart glasses, and smart cockpits, and has connected more than 70 companies in the months since the open test.
02
The “Trinity” strategy: First, pave a commercialization path for Chinese AI companies
SenseTime’s “trinity” strategy is revealing the trend of future industry changes: the layout of a single link is no longer enough to cope with the fierce competition in the future. Only by building an ecosystem from top to bottom can we gain a firm foothold when the curtain of the generative AI era is opened.
What is the “trinity” strategy? It is the “AI infrastructure (large devices) – large models (daily updates) – applications” “trinity strategy.
They form a virtuous circle: application scenarios drive model development, and large models break through the needs of various vertical fields through differentiated innovation; at the same time, the development of large models also drives the continuous optimization of AI infrastructure, and in turn, powerful AI infrastructure can support the leap in big model capabilities; in addition, AI infrastructure directly supports the explosive growth of various applications, and the actual needs of these applications guide the direction of infrastructure upgrades.
Sense Time’s large-scale facility, as the industrial base of generative AI, is located in Lingang District, Shanghai. It is Sense Time’s intelligent computing center and one of the largest AIDCs in Asia. Computing power AIDC is the core infrastructure that the entire market pays the most attention to in 2025. For example, Alibaba announced that it will invest 380 billion to build AIDC in the next three years, and Tencent’s capital expenditure in 2025 will also be close to 100 billion. Sense Time’s large-scale facility was awarded the first 5A-level intelligent computing center certification in the country by the Ministry of Industry and Information Technology in 2024. As of September 2024, the total operating computing power of Sense Time’s large-scale facility reached 23,000 P. It increased by 92% year-on-year, and the intelligent computing nodes have completed the national layout covering the Yangtze River Delta, Guangdong, Hong Kong and Macao, Beijing-Tianjin-Hebei, and key areas in the central and western regions, and realized unified scheduling of the national network.
Sense Time has made significant breakthroughs in the efficiency of large-scale model training and reasoning, demonstrating its advantages in the coordination of models and infrastructure. Through automated multi-dimensional parallel strategies and communication and operator optimization, Sense Time’s large-scale device cluster training efficiency has been significantly improved, and FP8 mixed-precision training has been successfully achieved. For third-party models such as DeepSeek V3, Sense Time has processed more than 1,600 tokens per second (TGS), which is better than DeepSeek’s official results without special optimization. It is expected that specific optimization of a single model in the future will bring greater performance improvements. In terms of reasoning, Sense Time supports low-bit quantized reasoning of both open source and self-developed dual engines. The latest internal tests show that the reasoning capability of the self-developed multimodal model is 50% higher than that of the daily new 5.5 released in July 2024. For the token generation time of the DeepSeek R1 model, it is 15% more efficient than that of the leading manufacturers, and its performance leads the industry under high service standards (SLA) conditions. According to the “2024 China GenAI Technology Stack Market Report” released by Frost & Sullivan and TouBao Research Institute, SenseTime’s comprehensive capability evaluation of large devices ranks first in the domestic market. Through joint tuning with large model iterations, we create the “AI infrastructure that understands big models best”, which not only serves the training and reasoning of daily new big models, but is also widely used in embodied intelligence, AIGC, AI4S (Al For Science) and other future emerging industries. With the joint optimization of the AI infrastructure system and model algorithms, the training and reasoning costs of daily new large models have reached the industry-leading level.
AI applications are booming, and the tide is turning. But one thing is certain in the industry: with the continuous improvement of technical capabilities and the rapid decline in usage costs, the boom of applications is just around the corner!
SenseTime adopts the “three-in-one” strategy, which is to provide basic infrastructure services and to deepen terminal products and services. It deepens the commercialization path and value chain through integrated capability enhancement and customer service chain. It is this relatively scarce deep collaborative capability that has built SenseTime’s strong moat and formed a unique and leading business model in the market. What awaits the AI industry is undoubtedly a grand banquet, and it may even be a trillion-level opportunity.
03
The “10 trillion-level” blue ocean, the prelude to the generative AI industry has been opened
After the DeepSeek moment, the market generally believes that the curtain of the domestic generative AI industry is accelerating in 2025. From the G end to the B end and then to the C end, we can see that the penetration rate in the industry is rapidly increasing and the market size is growing rapidly. And SenseTime’s financial report has further added a heavy chip to this consensus.
Morgan Stanley pointed out in a report in February that 2025 will be the watershed year for GAI’s return on investment (ROI) to turn from negative to positive, and this technological revolution is expected to generate more than $1 trillion in revenue by 2028. The 34% gross profit margin level marks that the industry has officially crossed the break-even line, and the subsequent industry will enter a stage of accelerated growth. Self-sustaining ability is the core focus of the next stage of industry competition.
Looking ahead to 2028, Morgan Stanley said that total revenue from generative AI is expected to be close to $1.1 trillion: enterprise software will contribute $401 billion, and the potential productivity impact of generative AI solutions on the U.S. labor market will be $6.4 trillion; consumer Internet-related spending will contribute $683 billion, of which generative AI will improve efficiency by $469 billion and generative AI advertising revenue will be $197 billion.
Facing the trillion-dollar GAI era, SenseTime is using the new “1+X” architecture, “three-in-one” strategy, and “one base and two wings” implementation approach to comprehensively promote the group to become an industry-leading AI platform and industry leader.
The above content does not constitute investment advice and does not represent the views of the publishing platform. Users should consider whether any opinions, views or conclusions in this article are consistent with their specific investment objectives, financial situation or needs. The market is risky and investment should be cautious. Please make independent judgments and decisions.