Market Share No. 1! SenseTime SenseCore Ranks First Among China's AI Native Cloud Vendors

Recently, leading research firm Frost & Sullivan and the LeadLeo Research Institute, jointly released the 2025 H1 China Full-Stack AI Cloud Services Market Report. The report indicates that SenseTime ranks 4th in China's full-stack AI cloud services market in overall market share, while topping the list of AI Native Cloud vendors, continuing to lead the AI Native Cloud sector.

Frost & Sullivan pointed out that AI native technology is reshaping cloud services, defining "AI Native Cloud" as a cloud computing service model specifically designed and optimized to support AI workloads. This model deeply integrates AI technology into every layer of cloud services, rather than merely adapting traditional cloud services to meet AI needs.

In response to the AI native trend, Frost & Sullivan divided full-stack AI cloud vendors into two categories: those undergoing comprehensive AI transformation from traditional cloud computing vendors; and AI Native Cloud vendors, like SenseTime, which have been intrinsically linked to AI since their establishment, possess AI genes, and have gradually evolved into full-stack AI cloud vendors.

The report stated, "SenseCore, as a 'cloud grown from AI', has been designed from the ground up for AI native applications, embedding AI as its core gene." Through continuous breakthroughs in underlying capabilities, such as computing resource scheduling, software-hardware synergy, and heterogeneous training, SenseCore has established a full-stack AI cloud capability system that is "more attuned to AI, models, and industries", creating AI native cloud services for enterprises and industries, and fostering the systematic implementation of industrial ecology and application scenarios.

End-to-End AI Infrastructure Technological Breakthroughs, Empowering the Entire Model Lifecycle

The Frost & Sullivan's report emphasized that the core advantage of AI Native Cloud vendors lies in their ability to "form an end-to-end technical closed loop from precise model tuning at the model layer to efficiency improvements at the inference layer, and infrastructure adaptation and refinement".

Computing-power and power-supply synergy: SenseCore innovatively connects AI platform operation data, computing performance indicators, and power-side infrastructure data through a synergy platform. This significantly improves the utilization rate of per-unit computing power and enables the scale of computing clusters to double within per-unit power indicators.

Stable operation of ultra-large-scale training clusters: SenseCore employs mechanisms such as checkpoint resume training, automatic fault tolerance, task migration, and topology-aware scheduling to ensure stable cluster operation amid node fluctuations, resource switching, and dynamic scaling. Additionally, with the SRE-Agent intelligent operation and maintenance agent, it can achieve minute-level, closed-loop response to task abnormalities, greatly improving efficiency and system autonomy.

Efficient and stable training on large-scale domestic heterogeneous computing power: SenseTime’s SenseCore took the lead in realizing large-scale heterogeneous mixed training on a cluster of 5,000 domestic GPUs, achieving a computing power utilization rate of up to 80% and efficiency reaching 95% of homogeneous training, successfully overcoming the limitation of single GPU. Ecologically, SenseTime SenseCore has completed full adaptation with the Ascend 384 super node and established strategic cooperation with domestic chip companies like Cambricon.

Strong training and inference optimization capabilities: SenseTime’s SenseCore achieves collaborative optimization between infrastructure and models. Its online inference performance is 15% higher than that of leading vendors, while its offline inference speed is 5 times faster in the Prefill phase and 3.5 times faster in the Decode phase. This reduces the inference cost of multimodal large models to the same level as large language models.

Full-link tool system: SenseCore provides a complete large model production tool system, covering computing power, data, models, and applications. It has also developed a comprehensive model family that includes general-purpose models, multimodal perception and generation, as well as code and 3D models, offering systematic tool support for the entire lifecycle of large model R&D.

Strong Scenario Orientation + High Responsiveness, Adapting to the Personalized Needs of Traditional and Emerging Industries

The Frost & Sullivan report noted that one of the representative characteristics of AI Native Cloud vendors is the formation of a "scenario-oriented + high-response" service system, closely aligning with customers' personalized needs. Leveraging profound scenario-based solution capabilities and a high-response expert service system, SenseCore collaborates with industry customers to co-develop "Lighthouse Projects" and benchmark cases, meeting personalized demands while forming long-term, highly sticky cooperative relationships. Examples include the following:

Emerging industries: Activating end-to-end innovation efficiency. For instance, in the field of embodied intelligence, SenseCore offers the only complete end-to-end solution in the market, enabling full-link empowerment for industrial innovation. In the field of video generation, it has established in-depth cooperation with the top three AIGC startups. And in the AI4S field, it has partnered with multiple national scientific research institutions to drive scientific research innovation through full-stack AI infrastructure.

Traditional industries: Addressing intelligent upgrading challenges. For example, SenseTime’s SenseCore and the China RAILWAY FIRST Survey and Design Institute jointly developed the industry's first multimodal large model application platform, integrating professional knowledge across 28 industries, effectively addressing industry pain points and accelerating the digital and intelligent transformation of the national railway industry.

Agentic AI: Building "Self-Evolving" Agent Infrastructure

The Frost & Sullivan report highlights that Agentic AI has driven a paradigm shift in the field of AI, redefining how cloud services deliver value and enhancing enterprise productivity from "tool assistance" to "autonomous evolution".

SenseCore has progressively developed a future-oriented AI Agent Infra, integrating model capabilities, tool ecology, and agent self-evolution mechanisms, providing enterprises with full-link support from generation to verification and gray-scale launch. For example, in a large-scale project, it realized unified management and scheduling of over 100 agents through a single business portal, most of which were independently generated by top-level agents. Covering more than 30 internal business systems, over 4,000 tools, and more than a dozen vertical models, it achieved autonomous AI positioning and task execution, verifying the scalability and efficiency of its Agent Infra in real-world industry-level projects.

Yang Fan, Co-Founder of SenseTime and President of the SenseCore Business Group, stated: "With AI native genes at its core, SenseCore will continue to drive technological innovation, industrial landing, and ecological co-construction, helping enterprises efficiently embrace intelligence and accelerating the large-scale application and value release of AI across thousands of industries."