Sensetime-W (0020.HK), a leading AI company listed in Hong Kong, released its fiscal year 2025 results at the end of March. The report contained a milestone figure: the company achieved positive EBITDA of 380 million yuan in the second half of 2025, marking the first positive reading since its IPO. Furthermore, Sensetime's operating cash flow also turned positive for the first time in the latter half of 2025. For the full year 2025, the company's cash conversion cycle was significantly shortened from 228 days to 129 days. This series of improvements indicates that Sensetime has developed a self-sustaining,良性循环 capability.
The group's revenue performance was equally remarkable. Total revenue exceeded 5 billion yuan, a 33% year-on-year increase, reaching a record high, while losses narrowed substantially by 58.6% compared to the previous year. Breaking down the business segments, Sensetime's visual AI business is transitioning from a phase of technological investment to one of scaled returns. As a cornerstone business, visual AI plays a crucial role as a cash cow. The 2025 financial report shows that the CV2.0 business not only achieved revenue growth but also reported a positive net profit for the first time, maintaining positive cash flow for two consecutive years. This success is attributed to Sensetime's integrated general and specialized algorithm production system, which uses large models to understand complex scenarios and small models for efficient task execution, effectively addressing the traditional AI challenge of balancing computational cost with recognition accuracy. Having led the Chinese computer vision market share for nine consecutive years, Sensetime has now refined this business into a stable source of cash flow.
If visual AI is Sensetime's shield, then its generative AI business is the spear piercing the sky. The financial report reveals that generative AI business revenue surged 51% year-on-year to 3.63 billion yuan, accounting for over 70% of total revenue. Throughout 2025, Sensetime continued to invest in cutting-edge R&D, solidifying its leading position globally in the field of multimodal large models. Its '日日新' series of large models continually raised the bar for multimodal reasoning; the SenseNova-SI spatial intelligence model pioneered the blurring of boundaries between virtual and real worlds; and the release of the new NEO architecture represented a complete overhaul of the underlying logic for native multimodality. These breakthroughs not only redefine model inference efficiency and training paradigms but also signify that Sensetime has entered a new generation in its pursuit of multimodal integration.
In terms of large model applications, Sensetime has deeply penetrated various sectors including general office work, finance, intelligent marketing, and content creation, serving customers across numerous industries such as automotive, smart devices, consumer goods, internet, embodied AI, finance, education, and healthcare. In its collaboration with SMZDM Technology, Sensetime helped improve store operation efficiency by 20 times, live streaming operation efficiency by 6 times, traffic placement efficiency by 5 times, and real-time monitoring efficiency by 3 times.
As the solid technical foundation of the integrated strategy, the AI Data Center achieved a deep transition from technological strength to industrial闭环 in 2025. While pursuing computational efficiency, Sensetime, through its globally first fully integrated AI intelligent system encompassing computing power management, IDC operations, and energy storage systems, utilizes large models to predict power loads and dynamically schedule computing tasks, which is expected to lead to a 7% reduction in electricity costs.
Despite delivering what could be termed its strongest financial report ever, the capital market's reaction has been relatively muted. A significant disparity exists between the current stock price and the company's record revenue, positive EBITDA, and world-leading technological capabilities. This undervaluation may stem from the market overlooking that Sensetime has reached a crossroads in its valuation framework. Previously, the market generally applied a PS valuation method to unprofitable AI companies. Now, with both EBITDA and cash flow turning positive, the valuation anchor should shift towards PE and PCF metrics used for profitable growth companies.
Projecting from Sensetime's current situation, if it can maintain the positive EBITDA trend and achieve continued improvement in operating cash flow in 2026, its valuation framework is poised for a systematic re-rating. Considering that most AI companies are still loss-making, using global SaaS company valuation levels as a benchmark is appropriate. Mature SaaS companies typically trade at EV/Revenue multiples between 8x and 12x, whereas Sensetime currently trades at less than 5x this metric, indicating significant room for valuation repair.
Although market expectations for the AI industry exhibit inertia, when a company simultaneously possesses a cash cow (the profitable CV2.0 business), a growth engine (the surging generative AI business), and a moat (the AI Data Center with compute-power-electricity synergy), coupled with its unique position in China as the only company with a full-stack闭环 from computing power to models to applications, a value re-rating appears to be a matter of time. Historical experience shows that capital market sentiment is short-term, while a company's intrinsic value is long-term. Companies that maintain financial health and continuous technological investment even during industry downturns ultimately gain market recognition.
Comments