Good News on Listing | Frost & Sullivan Assists Beijing Zhipu Huazhang Technology Co., Ltd. to Successfully List on the Hong Kong Stock Exchange (2513.HK)

Beijing Zhipu Huazhang Technology Co., Ltd. (hereinafter referred to as 'Zhipu Huazhang' or 'the Company') successfully listed on January 8, 2026. The Company issued 3,741.95 million H shares at a price of HK$116.2 per share, raising a net amount of approximately HK$4.17 billion.

During the process of listing in Hong Kong this time, Frost & Sullivan mainly undertook the following tasks: helping the issuer accurately and objectively understand its positioning in the target market, using objective market data to discover, support and highlight the issuer's competitive advantages, assisting the issuer, investment banks and other intermediaries in completing the writing of relevant parts of the prospectus (such as overview, competitive advantages and strategy, industry overview, business and other important chapters), helping the issuer complete communication with the Hong Kong Stock Exchange and investors, assisting investors in quickly understanding the market ecosystem and competitive landscape, and providing assistance to the issuer in completing feedback on various industry-related issues from the Hong Kong Stock Exchange, etc.

Frost & Sullivan has always been a leader in helping companies go public in Hong Kong. According to LiveReport's big data, from January to December 2025, and over the past 36 months, Frost & Sullivan provided listing industry advisory services for 83 (accounting for 72%) and 180 (accounting for 71%) Hong Kong-listed IPOs respectively, ranking first in terms of number. It has a wealth of industry experience and communication skills with regulatory authorities, exchanges, investment and financing institutions, and various related agencies.

• Based on 2024 revenue, the company is the largest independent large language model provider and the second-largest in China;

   

• The company released China's first proprietary pre-trained large model framework, the GLM framework, in 2021, and in 2022, it open-sourced its first model with a scale of 100 billion (GLM-130B);

   

• According to the evaluation results of 12 industry standard benchmark tests conducted in July 2025, GLM-4.5 ranked third globally, first in China, and at the top of the list for open-source models worldwide. Within just 48 hours after its launch, GLM-4.5Hugging FaceRise to the top of the global charts on the (world's largest open-source model platform);

   

• AutoGLM is inAgentBenchAchieved through the intelligent agent AI benchmark certified by the Stanford University's 2024 AI Index reportSOTA refers to State-of-the-Art TechnologyPerformance;

   

• CogVideoX achieves top-tier performance and ranks among the top in comprehensive SuperCLUE-I2V benchmark tests;

   

• GLM-4V is China's first open-source bilingual multimodal dialogue model;

   

• GLM-4-VoiceIt is China's first end-to-end super-humanoid voice model, further consolidating the company's comprehensive leadership position in multimodal capabilities and demonstrating its pioneering abilities in all major modalities;

   

• CogView-4 is the first open-source text-to-image model that supports Chinese character generation. It also ranks first in the comprehensive score of the DPG-Bench benchmark test, achieving SOTA performance among open-source text-to-image models;

   

• GLM4-Plus (the latest model in the GLM-4 series) has achieved SOTA performance in multiple benchmark tests and is significantly superior to comparable models in long video processing and fine action understanding.

Fourth Industrial RevolutionIt represents a profound technological transformation that has occurred globally, characterized by the integration of the physical, digital, and biological worlds. Unlike the previous three industrial revolutions (powered by steam engines, electricity, computers, and the internet), the fourth industrial revolution is characterized by faster change rates, a broader scope, and more profound social and economic impacts. Artificial intelligence is one of the most critical driving forces and decisive factors of this industrial revolution, playing an irreplaceable role in empowering other digital technologies, achieving intelligence and automation, creating new business models and industries, and integrating the physical and digital worlds.

Artificial intelligence is an important branch of computer science, referring to the technology that simulates and extends human intelligence, enabling machines to possess capabilities such as perception, understanding, reasoning, learning, and autonomous decision-making (these abilities allow machines to complete complex tasks without human intervention). The development of artificial intelligence is profoundly changing economic growth, business operations, and human life. It is estimated that by 2030, AI will empower at least 20% of global daily business decisions, support mainstream intelligent devices for at least 80% of consumers worldwide, and create an AI-impacted economy worth over $20 trillion.

Computing power, algorithms, and data are the key drivers of growth in the artificial intelligence market:

• computing powerComputing power is the foundation for AI model training and inference, and a prerequisite for realizing the large-scale commercial application of artificial intelligence technology. As models become more complex and the scale of model parameters grows exponentially, the demand for computing resources continues to rise.

   

• algorithmAlgorithms are essential building blocks that empower artificial intelligence to mimic (and even surpass) human intelligence. The optimization and innovation of algorithms directly affect the learning efficiency, comprehensive capabilities, and adaptability of artificial intelligence. From the relatively simple algorithms used in the early days (such as decision trees, support vector machines, etc.) to the advanced methods today (such as deep neural networks, transformer-based algorithm architectures, etc.), the evolution of algorithms has driven breakthroughs in fields such as multimodal interaction and agent-based task execution. Excellent algorithms not only improve model performance but also significantly reduce the dependence on computing power, thereby enhancing the feasibility and economic viability of artificial intelligence applications.

   

• dataData is the fundamental element for training and continuously optimizing artificial intelligence models. The improvement of AI capabilities depends on the input of a large amount of high-quality data. Through in-depth data mining and pattern recognition, AI models gradually enhance their abilities in perception, understanding, and decision-making tasks. Therefore, the scale, quality, and diversity of data directly determine the learning effect and intelligence level of AI systems.

The market scale of artificial intelligence in China increased from RMB 937 billion in 2022 to RMB 1607 billion in 2024, with a compound annual growth rate of 31.0% from 2022 to 2024; it is estimated that by 2030, the market scale of artificial intelligence in China will further increase to RMB 9930 billion, with a compound annual growth rate of 35.5% from 2024 to 2030.

Data source: Analysis by Frost & Sullivan

Artificial intelligence is in a transitional phase from Artificial Narrow Intelligence (ANI), which is limited to specific tasks, to Artificial General Intelligence (AGI), which refers to a complex level of artificial intelligence that matches or even surpasses human capabilities in all cognitive tasks. As the core of this transformation, large language models are increasingly becoming a key element driving a new era of artificial intelligence development. With continuous leaps in parameter scale, semantic understanding, multimodal integration, and self-evolution capabilities, large language models have broken the limitations of traditional discriminative AI application scenarios. They have initially demonstrated technical potential to approach general intelligence.

According to the capabilities of large language models, their abilities can be divided into five levels:

• Pre-training phaseSuch models can understand, write, and speak natural language, possess basic language communication abilities such as text dialogue, and represent an early form of language intelligence.

   

• Alignment and Inference PhaseThese models possess multimodal understanding and output capabilities, align with human intentions, and further develop into reasoning and planning. This can enhance security, reduce hallucinations, and extend the alignment function from text to images, videos, audio, and actions.

   

• Independent learning phaseThese models can use tools and solve real-world problems by calling external resources (such as APIs, standard software, or physical devices). They can also plan and execute multi-step tasks through self-criticism, self-reflection, and contemplation, marking a shift in AI from closed models to open ecosystem collaboration.

   

• Self-Cognition StageThese models operate independently of human supervision, forming autonomous attitudes and simulating emotions by observing and interpreting their own behavior.

   

• conscious intelligence stageSuch models have a certain understanding of their internal processes and external environment, capable of exploring scientific laws and solving philosophical propositions. These models demonstrate systematic thinking and organizational capabilities, and may integrate themselves into complex social structures or develop self-organizing systems.

In terms of revenue, the market scale of large language models in China reached RMB 5.3 billion in 2024, of which enterprise-level customers contributed RMB 4.7 billion. Enterprise-level customers will still be the core driving force behind market growth. It is estimated that by 2030, the market scale of large language models for Chinese enterprises will reach RMB 90.4 billion, with a compound annual growth rate of 63.7% from 2024 to 2030.

Data source: Analysis by Frost & Sullivan

Among them, in terms of revenue, the market scale of enterprise-level large language models in China reached 4.7 billion yuan in 2024. The market scale for cloud deployment was 900 million yuan, and the market scale for local deployment was 3.8 billion yuan.

Data source: Analysis by Frost & Sullivan

● Significantly reduces the threshold and cost for enterprises to access large language models, driving demand for cloud deployment

Cloud deployment provides "ready-to-use" services in the form of APIs. Enterprises do not need to worry about underlying model structures, computing power management, or maintenance issues. Cloud deployment operates on a pay-as-you-go basis, significantly reducing initial investment and operational costs. For example, if an enterprise chooses non-cloud deployment, the cost of a single H100 GPU is about $25,000. For enterprises that need to configure multiple GPUs in a single system, the related expenses can be quite high, especially for small and medium-sized enterprises. Cloud deployment reduces or eliminates the need for large capital expenditures such as GPU purchases by adopting a pay-as-you-go operational expenditure model, thus avoiding large upfront hardware investments. Cloud deployment enables small and medium-sized enterprises and innovation teams to access high-quality large language model capabilities with lower thresholds through standardized services and tiered pricing mechanisms. The inclusive nature of cloud deployment helps AI technology transform from a tool exclusive to large enterprises into a general infrastructure for all types of enterprises, accelerating the intelligent transformation of the entire industry.

● The agile features of cloud deployment have also driven demand growth

In a rapidly changing and highly competitive industry environment, enterprises increasingly need flexible and scalable AI capabilities. Cloud deployment provides standardized interfaces through the cloud, allowing enterprises to complete model integration within hours, achieving a rapid transformation from trial to deployment. The 'elastic supply and rapid delivery' model enables enterprises to quickly and flexibly experiment with and implement solutions. In contrast, local deployment may be slower due to internal infrastructure construction progress, and its scalability may be limited by the capacity of purchased hardware.

● Powerful and rapidly iterating technical capabilities

Leading cloud large language model vendors typically invest a significant amount of computing power, training data, and engineering resources in training specialized large language models, endowing them with powerful generalization capabilities and reliability. These models are continuously fine-tuned and optimized by dedicated teams and updated through cloud push. This rapid iteration ensures that enterprises can always obtain the most advanced artificial intelligence capabilities, which is difficult to achieve with most local deployment of large models.

● Market share is expected to further increase as we lead market participants

As the cloud solution market matures, leading participants are consolidating their competitive advantages in model performance, ecosystem development, customer resources, and service delivery. With their large-scale delivery capabilities and technical expertise, leading players can build full-cycle business models and diverse monetization strategies. Given the service dependence and high customer loyalty of cloud solutions, market leaders are more likely to retain and expand their user base. As enterprises increasingly focus on model effect stability, service reliability, and continuous optimization, small players face challenges in computing resources, product maturity, and domain-specific customization. This could lead to further market concentration.

●Multimodal fusion

Multimodal fusion is gradually becoming a key direction. Enterprises are no longer satisfied with single text processing and generation, but are seeking models that can handle and generate visual, auditory, and motion data. For example, manufacturing companies need models that can analyze images and video data on production lines for quality inspection and fault detection. The financial industry hopes to improve the accuracy of risk assessment and customer interaction experience through multimodal data fusion. In the future, large language models will expand from simple text processing to various application scenarios such as image recognition, video analysis, and 3D modeling, forming more comprehensive artificial intelligence solutions that provide better decision support for enterprises.

● Demand for secure and controllable AI capabilities

For enterprises with core task systems, localized deployment can reduce dependence on external networks and minimize service interruption risks due to connection issues or cloud service failures. Enterprises can independently control system architecture and resource allocation, improving system reliability and response speed. Moreover, since data privacy has always been one of the primary concerns for enterprises deploying large language models, localized deployment keeps data flowing within internal infrastructure, helping to meet strict requirements for data security, privacy protection, and compliance. This is particularly crucial for institutions in industries such as finance, healthcare, and government affairs, where the importance of data protection, autonomy, and direct operational control far exceeds the advantages of cloud deployment.

● Customized needs for specific industries

Large language model application scenarios typically exhibit a high degree of specialization, involving fields such as intelligent customer service, legal document review, financial analysis, and medical diagnosis. Localized deployment enables enterprises to use proprietary data and domain-specific knowledge bases for private training and fine-tuning of models. This allows models to better align with business workflows and significantly improve performance and value in practical applications.

● Open-source and commercialization-driven model

Open-source and closed-source models will develop together in a complementary manner. Open-source models, due to their flexibility and accessibility, may still be attractive to enterprise-level customers. Open-source models help reduce innovation costs and make it easier for enterprise-level customers to develop customized applications. The openness of open-source models also promotes rapid technology dissemination and collaborative innovation, accelerating the adoption of models. Meanwhile, closed-source models are developed and maintained by professional technical teams, providing more stable and efficient performance and more comprehensive support services. In the future, open-source models will play an important role in driving technological innovation and fostering community collaboration, while closed-source models will lead in commercial applications and enterprise services. This dual-driven development model will provide enterprises with a wider range of choices to meet diverse needs in different industries and scenarios.

● Improvement of the value chain and ecological construction of large language models

The rapid development of the artificial intelligence market in Chinese enterprises is inseparable from the support of a mature value chain, ranging from upstream data collection and annotation, and computing resource infrastructure, to midstream model development and algorithm optimization, and finally to downstream specific industry applications and commercialization. This ecosystem is showing a more collaborative trend, with increasingly close partnership relationships on the value chain promoting sustainable market development.

● Universal bases coexist with vertical ecosystems

The market will form a pattern where general-purpose bases coexist with vertical ecosystems. General models (usually provided by tech giants) will serve as digital infrastructure, offering standardized services for general semantic understanding and generation. At the same time, more models and solutions tailored to specific industries will emerge in vertical fields. For example, in the healthcare industry, large language models will combine with data such as medical images and electronic health records to form a healthcare vertical ecosystem. In the financial industry, large language models will focus on scenarios such as risk analysis and investment decisions to build a financial ecosystem. This pattern of coexistence between general-purpose bases and vertical ecosystems will provide enterprises with more flexible options, allowing them to leverage both the broad capabilities of general models and the deep functions of vertical solutions, driving the market towards a more diversified and specialized direction.

Based on 2024 revenue, the company is the largest independent large language model provider in China and the second-largest.

Data source: Analysis by Frost & Sullivan

● Technological barriers

Leading manufacturers, relying on their independently developed large language model pre-training frameworks, have constructed a comprehensive multi-level model combination with breadth and depth. This combination forms a complete matrix covering from lightweight end-side models to flagship models with hundreds of billions of parameters, supporting various application scenarios such as text generation, image understanding, code generation, multimodal interaction, retrieval-enhanced generation, and video synthesis. This model architecture, which covers multi-dimensional task capabilities, not only meets the increasingly diverse needs of customers but also continuously improves the comprehensive ability and training efficiency of models through iterative development. In addition, leading enterprises can quickly customize and optimize models based on actual customer feedback, and in the continuous accumulation of scenario data and industry demands, they can reverse-engineer underlying technology evolution, constructing a positive feedback loop system with low training costs, strong model performance, and high application adaptability. The systematic collaborative ability from the underlying framework, model design to application feedback greatly increases the technical difficulty for newcomers to replicate and surpass, constituting a core technology entry barrier in the Chinese large language model market.

● Flexible business model and delivery strategy

Enterprise customers vary significantly in terms of industry attributes, data sensitivity, computing power infrastructure, and budget scale. Therefore, the ability of vendors to provide flexible and customizable business models as well as diverse delivery strategies has become a key factor determining their customer coverage breadth and market penetration capability. Leading large language model vendors typically offer various business models such as on-demand invocation, subscription-based payment, and one-time deployment options to meet the multi-level needs of customers ranging from small and medium-sized enterprises to large enterprises.

●Ecological construction capacity

Building a large-scale and deeply integrated ecosystem has become an important barrier for leading manufacturers to create sustainable competitiveness. Leading enterprises have established comprehensive ecological networks covering developer communities, hardware partners, industry customers, and public domain users. This not only expands the application scope of model capabilities, improves deployment efficiency, but also significantly raises the threshold for new entrants. In terms of the developer ecosystem, some manufacturers have taken the lead in implementing open-source strategies. By continuously iterating models and building active communities, they have accumulated a large base of developers and expanded the reach of model technology. Upstream, leading manufacturers have carried out close collaboration with mainstream computing power chip manufacturers, achieving efficient adaptation across multiple hardware platforms and optimizing model inference and training performance. Downstream, through cooperation with independent software vendors and key customers in industries such as finance, healthcare, government affairs, and manufacturing, they jointly develop intelligent solutions to promote the implementation of practical applications. This 'multi-directional expansion and closed-loop feedback' ecological model that connects open-source communities, software-hardware collaboration, and industry applications not only accelerates technical optimization and product iteration but also achieves deep integration of resources, capabilities, and customers, constructing a unique ecological barrier that is difficult to replicate.

● Talent barriers

The Chinese large language model market has a strong demand for talent, especially experts with a profound technical background and rich experience. Industry-leading participants have attracted top talents and built strong technical teams, while newcomers face fierce competition in acquiring professional skills talent.