Innovation Record! SenseTime's 2025 Revenue Exceeds 5 Billion Yuan, EBITDA Turns Positive in Six Months

On March 24, SenseTime Technology (0020.HK) released its full-year earnings report. For 2025, the group’s total revenue increased by 33%, surpassing 5 billion yuan, setting a new record high and marking the fastest growth in nearly three years; net loss significantly narrowed by 58.6%. Adjusted net loss achieved four consecutive half-yearly year-over-year reductions, with EBITDA for the second half reaching 380 million yuan, turning positive for the first time since listing, exceeding market expectations. The company successfully transitioned from a period of technological investment to a sustainable period of explosive growth. During the reporting period, SenseTime’s revenue-generating capability achieved a qualitative leap, with continuous improvement in capital efficiency, steady cash flow, and trade receivables reaching 4.87 billion yuan, a record high; in the second half, operating cash flow turned positive for the first time since listing. SenseTime will launch a new model based on the second-generation NEO architecture in Q2 2026, expected to achieve a multiple leap in efficiency and cost performance, broadly empowering intelligent AI applications.

Chairman and CEO Xu Li stated: “We firmly believe that the deep integration of language and vision is an efficient path to突破智能上限. Based on the native NEO architecture, we efficiently realize unified understanding and generation, exploring a new ‘scale law’ for multimodal AI. These technological breakthroughs combined with deep integration of intelligent agents will open new application possibilities and empower new vertical scenarios. While maintaining technological innovation at the core, SenseTime has also achieved dual growth in revenue and EBITDA, demonstrating outstanding growth resilience and operational efficiency in the industry, steadily moving towards high-quality development.”

Deepening Native Multimodal Large Models, “Efficiency + Cost Reduction” Drive Business Acceleration

In 2025, SenseTime continued to invest in cutting-edge technology R&D, achieving breakthroughs in large model architecture innovation, training paradigms, inference efficiency, and spatial understanding, maintaining a leading position domestically.

SenseTime’s daily-updated multimodal large models continued to improve performance and dominate rankings, ranking first in comprehensive evaluations such as SuperCLUE and OpenCompass in January, May, and December 2025. Meanwhile, SenseTime released and open-sourced the spatial intelligence model SenseNova-SI series, ranking first globally in multiple authoritative spatial intelligence metrics; the open-source Kairos-SenseNova became the first integrated open-source embodied native world model capable of multimodal understanding, generation, and prediction. In December 2025, SenseTime launched and open-sourced the new native multimodal model architecture NEO, which requires only one-tenth of the training data and computing power of comparable models to achieve top performance. These breakthroughs not only redefine inference efficiency and training paradigms but also mark SenseTime’s entry into a new generation of multimodal fusion development.

Leveraging the leading daily-updated multimodal model technology, SenseTime explores the closed-loop value of intelligent agents in strategic sectors such as office, finance, marketing, and content generation, across complex tasks and fault tolerance. Clients span automotive, smart devices, consumer, internet, embodied intelligence, finance, education, and healthcare industries.

In the general office domain, the Little Raccoon family has served 15 million individual users and thousands of enterprise clients, achieving sevenfold monthly active user growth and a 90% reduction in business analysis cycle, significantly improving office efficiency. In intelligent marketing, SenseTime covers the entire e-commerce process, forming strategic partnerships with companies like ValueBuy Technology, boosting store operation efficiency by 20 times, live broadcast efficiency by 6 times, traffic investment efficiency by 5 times, and real-time monitoring efficiency by 3 times. In content creation, the creative agent Seko directly outputs episodes up to 3 minutes long, generating 100 episodes continuously, rapidly attracting over 300,000 creators, with monthly active users reaching 100,000. The “Financial Digital Decision Platform” has rapidly expanded revenue in banking and insurance sectors, and these vertical scene practices contribute steady growth to the group, while high-value decision data also enhance the deep logical reasoning capabilities of the “Daily New” models. Relying on the core advantages of the “Daily New” large model in full-modal interaction, low latency, and long memory, SenseTime promotes deep embedding of multimodal models into various hardware such as smartphones, AR glasses, in-vehicle interfaces, and embodied intelligent devices, with clients including Xiaomi, BYD, Transsion, and others.

In addition to building a scalable and replicable B2B business model, SenseTime has made progress in exploring next-generation AI native consumer applications. The Kapi series has accumulated millions of users, demonstrating the explosive potential of AI-native applications and showcasing SenseTime’s ability to integrate AI technology into daily life.

Large-Scale Infrastructure: Deep Coordination of Computing Power and Model Development, Witnessing Global Expansion Potential

As the core of the “Trinity” strategy, large-scale infrastructure achieved a deep leap from technological advantage to industry closed-loop in 2025. It supported nearly one million model development tasks throughout the year, connecting hardware to applications and software stacks to model adaptation. Among them, the LightX2V world model inference system surpassed overseas chip performance on domestic hardware platforms.

The infrastructure continued to accelerate domestic ecosystem collaboration, positioning itself as a “chain leader” in the domestic ecosystem. It jointly launched the “SenseTime Large-Scale Computing Power Mall” with chip manufacturers such as Huawei Ascend, Hygon, and Cambrian. The infrastructure has become a key partner for leading research institutions, internet giants, entertainment companies, embodied intelligence firms, and large model unicorns, and established China’s first overseas domestic computing cluster in Saudi Arabia.

Supported by CATL, the AI intelligent system built on the infrastructure can accurately predict power load through large model inference tasks, dynamically generate optimal energy dispatch strategies, and is estimated to save 7% on electricity costs and reduce over 4,000 tons of carbon emissions. As of the release of this earnings report, the total operational computing power of the infrastructure has increased to 40,400 PetaFLOPS (FP16).

CV2.0 Achieves First Net Profit and Positive Cash Flow! “X” Innovative Business Highly Recognized by External Capital

Visual AI is entering a scale-up phase from technological investment. During the reporting period, CV2.0 achieved profitability for the first time and maintained positive cash flow for two consecutive years, becoming a solid pillar driving revenue growth and cash flow improvement. SenseTime has ranked first in the CV market share for nine consecutive years. Customers in Southeast Asia, Northeast Asia, the Middle East, and other regions continue to repurchase visual AI products and services, with increasing interest from South America and Europe over the past year, forming a “Chinese-style overseas expansion” model that is replicable and scalable.

In 2025, SenseTime steadfastly implemented the “1+X” strategy, establishing an efficient collaborative system where the “mother ship (group)” provides the foundation and “sub-ships (ecosystems)” compete in various tracks, enabling a leap from single business growth to ecosystem value multiplication. The ecosystem companies incubated by SenseTime have made smooth progress in primary market financing and received high recognition from top internet giants, top venture capital firms, and industry funds. The edge-side chip and autonomous driving business units have successfully completed financing and achieved independent operation.

Looking ahead to 2026, the group is ready to迎接全球人工智能产业的“决赛”, continuing to deepen native multimodal architectures, establishing a leading position in the integration of native multimodal large models and spatial intelligence; simultaneously seizing the opportunity in the native AI intelligent agent market to achieve both traffic scale and business value explosion. The group will also fully promote domestic chip adaptation, continuously reduce inference costs for large models, and solidify product competitiveness with extreme cost performance. In visual AI, the group will continue to leverage both domestic and international efforts to expand scale and establish a global benchmark in the intelligent industry. The group will also continue to implement the “1+X” strategy, capturing the value-added dividends brought by vertical industry intelligence explosion.