Generative AI now accounts for close to 80% of SenseTime's revenue, and repeat-paying customers are becoming an important pillar supporting the company's loss reduction.
On August 26, SenseTime Group released its 2026 interim results, with first-half revenue of RMB 2.911 billion, up 23.4% year on year, and an adjusted net loss of RMB 386 million, narrowed by 67.3% year on year.
Among this, generative AI revenue reached RMB 2.327 billion, up 28.2% year on year, accounting for 79.9% of total revenue. SenseTime, which started out in computer vision, has clearly shifted its revenue structure toward large models, compute services and related applications.
Under International Financial Reporting Standards, SenseTime recorded a net profit of RMB 617 million in the first half, achieving profitability on this basis for the first time since listing. Higher fair value of investments and gains from the disposal of subsidiaries had a significant impact on the book profit turnaround.
From an operating perspective, the company's gross margin rose to 41.4%, up 2.9 percentage points year on year, while declines in research and development and administrative expenses also helped narrow losses.
A change more noteworthy than single-period profit is that SenseTime has begun to secure more recurring revenue. The company disclosed for the first time that first-half recurring revenue was RMB 1.145 billion, up 124.4% year on year, with its share of total revenue rising to 39.3% from 21.6% in the same period last year.
For AI companies that must continuously bear research and development and compute costs, repeated use of services by customers helps improve revenue predictability and reduces reliance on the delivery schedule of any single project.
But whether this portion of revenue can continue to expand still depends on customer usage frequency, renewals and service pricing. After model call volumes increase, how much gross profit each call can leave also affects the quality of commercialization.
SenseTime is expanding both compute services and agent applications. The former provides infrastructure for model training, fine-tuning and inference, while the latter increases opportunities for customers to use AI through scenarios such as office work, programming and content creation. SenseTime disclosed that enterprise users of its office agent Xiaohuanxiong grew more than sixfold, and its video creation product Vimi is also expanding its application scale.
The two businesses have different ways of generating returns. Compute services require higher equipment utilization and control of chip, energy and maintenance costs; agent applications need to prove that, after paying, enterprises can reduce manual operations, shorten working hours or improve delivery quality. For SenseTime, its existing compute infrastructure and enterprise customer resources provide conditions for expansion, but user scale still needs to be further converted into stable payments.
Xu Li, executive chairman of the board and chief executive officer of SenseTime, said at the results release that the company will combine model capabilities, compute infrastructure and agent control systems to drive AI to complete more complex real-world tasks. This direction means the company needs to further embed model capabilities into customers' business processes, and the importance of continuous service rises accordingly.
Expansion also requires funding. Wang Zheng, chief financial officer of SenseTime, said capital expenditure in the first half reached RMB 2.34 billion, mainly for large-scale compute infrastructure construction. New facilities can accommodate more customer demand, but equipment investment comes first and revenue recovery comes later, so utilization and collection speed will directly affect investment efficiency.
SenseTime has already narrowed its losses through revenue growth, gross margin improvement and expense control. Next, whether recurring revenue can continue to grow and whether compute expansion can secure corresponding customer demand will determine the pace of loss reduction. If agent applications can improve customer retention and usage depth, they may also bring more stable business volume to the infrastructure.
As these two parts of the business expand, the relationship among recurring revenue, service gross margin and capital recovery will reflect SenseTime's operating progress more than any single model release.
Comments