The rapid evolution of large model technology is driving the market to leap rapidly from the early capability exploration phase to the value implementation phase. At the same time, changes in the external environment are reshaping the underlying logic of industry development—geopolitical games continue to heat up, the uncertainty of computing power and technology supply chains rises, AI sovereignty is gradually becoming a national strategic proposition; national policies are accelerating their layout, and the institutional framework from top-level design to industry regulation is becoming increasingly perfect; the digital transformation of various industries is continuously extending into core business scenarios, posing higher demands for stability, continuity, and security. The superposition of multiple factors is driving gradual changes in the market situation and procurement logic of large models in China.
PART.01
The large model market has entered the "value realization phase" after the "technology release phase".
Since 2022, China's large model industry has undergone a three-stage evolution from 'parameter competition' to 'business embedding' and then to 'value realization', with development focusing on shifting from a single technical dimension to comprehensive capability development. The first stage (2022-2023) was the technology release period, dominated by parameter competition, with unprecedented market enthusiasm, but there were very few cases where models were truly embedded into business processes. The second stage (2024-2025) was the scale implementation period, when enterprises began to shift their focus to the practical application value of technology and sustainable business closed loops, but the common challenge became 'it's easy to pilot but difficult to replicate'. The third stage (since 2025) is the value realization period, with development gradually shifting from a single dimension of 'stronger models' to a comprehensive dimension that includes whether models can integrate into business processes, form quantifiable ROI, and achieve balance between cost, security, iteration, and other aspects.
Driven by both the continuous intensification of external technological blockades and the systematic response of internal policy systems, the strategic significance of computing power and model sovereignty continues to rise. Autonomous controllability has evolved from a single technical route choice to gradually become a core element of national information security strategy.
Data source: Analysis by Frost & Sullivan
PART.02
Market demand has exploded, and enterprises' focus on application has accelerated from tool trials to in-depth operations.
In 2025, the average number of projects won by large models reached 628, nearly quadrupling compared to the average of 127 projects in 2024. The disclosed amount of bids for large models in 2025 was about 29.52 billion yuan, more than tripling compared to about 6.467 billion yuan in 2024. The large-scale implementation of large models is driving the industry towards a new collaborative innovation paradigm: central enterprises generally adopt the 'innovation consortium' model, forming a seamless collaboration where 'central enterprises provide scenarios, data, and requirements, while technology companies provide technologies, algorithms, and solutions.'
Based on the tracking of AI deployment trajectories of sample enterprises, it was found that in the first half of 2025, enterprises were still in the 'tooling trial' phase. The core driving force was mainly 'empowering existing products and services' (accounting for about 34%), reflecting the initial tendency to use AI as an incremental plugin to existing businesses. However, in the second half of the year, the driving force underwent a restructuring: 'improving enterprise operational and R&D efficiency' jumped to the top with a proportion of 29%; at the same time, the weight for maintaining technological leadership to cope with competitive pressure also rose to 26%. This indicates that enterprises have shifted from early single-point product optimization to a deep reconstruction of internal R&D workflows and office collaboration systems, achieving a leap from external empowerment to internal deep cultivation. As technology applications enter the deep waters, enterprises are reorganizing the underlying operational foundation through large models, transforming AI capabilities into the core R&D competitiveness and a driving force for native business innovation. This dual internal and external drive marks that enterprises have entered a new normal where they drive organizational structure evolution with AI and enhance market response speed through global collaboration.
Data source: Analysis by Frost & Sullivan
PART.03
Industry distribution and characteristics: The core industries all take self-control and autonomy as the main procurement focus.
Throughout 2025, in terms of the industry distribution of projects awarded by large models, education and research, government affairs, communications, energy, and finance ranked among the top five industries. These industries led significantly in both project quantity and award amount compared to other sectors. The industries of transportation and healthcare followed closely behind, consistently positioned between sixth and seventh place, with an overall momentum for penetration continuing to improve. Overall, the top five industries share common characteristics such as 'high safety requirements, strong scenario standardization, and great repurchase potential.' Autonomous control is the core thread running through procurement decisions across these five industries.
Taking the energy industry as an example, it is a pillar industry of China's economy and is crucial to the national economic development. Driven by goals such as 'clean, low-carbon, safe, and efficient', the industry is moving from informatization infrastructure construction towards a new stage of deep integration of technology and business.
Data source: Analysis by Frost & Sullivan
PART.04
Manufacturer landscape: iFlytek, Volcano AI and Alibaba Cloud rank at the forefront, with real differentiation occurring in high-barrier industries
Looking at the performance of key general-purpose large model manufacturers, in 2025, six leading manufacturers— iFlytek, Baidu, Huoshan AI, Alibaba Cloud, Tencent Cloud, and Zhipu—totalled over 500 winning bids with a disclosed amount of approximately 4.503 billion yuan. Among them, iFlytek won 210 bids with a disclosed amount of about 2.316 billion yuan, ranking among the top in key manufacturers.
Industries such as energy, industry, finance, civil aviation, and government affairs have higher requirements for security, stability, industry knowledge accumulation, and complex delivery capabilities. These industries can also test whether manufacturers truly possess the ability to be self-controllable, integrate industry Know-how, deliver in complex scenarios, and sustain operations. Looking at the public bidding results of leading manufacturers, all three paths have entered the practical stage, but the distribution of results is not the same. iFlytek's public results continue to appear in high-barrier scenarios such as oil and gas, metallurgy, finance, and civil aviation. iFlytek Spark has become a large model partner for multiple central state-owned enterprises including National Energy, PetroChina, Sinopec, CNOOC, and the National Pipeline Network Group. Volcano Engine's samples are more distributed in scenarios such as automotive, education, and consumer goods, reflecting the rapid expansion of multi-modal capabilities and intelligent agent platforms on the application side. Alibaba Cloud is more reflected in its ability to undertake enterprise-level needs through cloud infrastructure, model platforms, and enterprise service ecosystems. All three paths constitute important samples for the commercial implementation of general-purpose large models in China today.
PART.05
Judgment on Four Major Trends: Scale Growth, Vertical Deepening, Hybrid Architecture, and Sovereign Overseas Expansion
Looking ahead, the Chinese general-purpose large model market will evolve along four major directions. First, procurement scale will continue to maintain a high growth rate, with value realization becoming the core driving force; second, after deep integration of AI with industry Know-How, vertical specialization will become the main battlefield, and long-term operational capabilities will build differentiated barriers. Against this backdrop, security and trust super assistants represented by iFlytek's AstronClaw are becoming typical paradigms for large models to realize large-scale value realization through deep embedding into industry core business processes; third, hybrid architecture will become the mainstream deployment mode. Manufacturers with lightweight deployment and full-stack self-developed capabilities can precisely match this dual demand of "flexible deployment + data security," and occupy a significant comprehensive competitive advantage in the era of hybrid architecture; fourth, domestically developed large models with autonomous control will accelerate their overseas expansion, becoming a new core choice for the global AI market: domestically developed large models will rely on their multi-language capabilities to focus on emerging markets such as ASEAN and Latin America, providing full-process technology and data security guarantees for Chinese-funded enterprises going global; under international cooperation frameworks such as the Belt and Road Initiative, domestically developed large models will be deeply integrated with local digital infrastructure construction, building a global AI ecosystem that is both resilient and open.
Data source: Analysis by Frost & Sullivan
