Frost & Sullivan: The implementation of large models has reached a turning point, and industry competition is shifting from 'technology competition' to 'value delivery'

Large models are transitioning from the pilot verification phase to large-scale implementation, with their core value lying in 'improving quality and efficiency'. Currently, large-scale implementation is mainly reflected in several major scenarios: the highest proportion is 'question-answering enhancement', followed by 'code assistants', 'document processing generation', and 'intelligent customer service'. In addition, industries such as finance, government affairs, and manufacturing are also accelerating the deployment of RAG, industry agents, and digital employees. Currently, the core obstacle to technology implementation is no longer cost but higher-dimensional strategic and application challenges, with the most prominent being 'unclear application scenarios that can generate real value', followed by 'lack of relevant technical talent', 'difficulty in ensuring data security and privacy', and 'difficulty in integrating with existing business systems and workflows of enterprises'.

Digital human technology is being widely applied in scenarios such as customer service, virtual anchors, online education, and corporate image endorsements. In terms of replacing real people, it has achieved a high degree of natural voice, expression, and body movement simulation, especially in standardized and repetitive real-time interactions, where it is stable and efficient. Taking Alibaba as an example, its 'digital human' technology has formed a strategic combination of 'cloud capabilities + Taobao implementation'. At the technical level, Alibaba Cloud provides an open platform for virtual digital humans, supporting text and audio-driven generation, supplemented by a low-threshold free experience; at the business level, Taobao Live has opened public domain interfaces to service providers, which has promoted the launch of nearly a hundred digital human live streaming rooms and brought an average viewing increase of about 5 times. Combining with the platform's potential energy of breaking through 1 billion views during the 618 period, a year-on-year increase of 53%, digital human e-commerce and store broadcasts are accelerating their popularization. However, large-scale implementation of digital human technology still faces major obstacles, including how to balance generation quality and cost, break through technical bottlenecks driven by real-time, clarify ethical regulatory responsibilities, and establish user trust.

B-side implementation shows clear characteristics: the focus of enterprise decision-making is shifting from 'pursuing the strongest single model' to 'seeking optimal solutions for specific business scenarios'. This means that the market has entered a new stage of 'value-driven' over 'technology-driven', with enterprises placing more emphasis on scenario fit and commercial value, and their needs evolving into 'flexible integration + technology autonomy and control' solutions to balance cost-effectiveness, flexibility, and security and controllability. In terms of implementation challenges, the biggest obstacle is 'unclear application scenarios that can generate real value', followed by 'lack of relevant technical talent' and 'data security and privacy' issues. It is worth noting that enterprise acceptance is also an obstacle, and 'high training and inference costs' are no longer the primary pain point, accounting for a low proportion.

Yes, AI application implementation has reached a critical turning point, transitioning from the pilot verification phase into a new stage of large-scale implementation. The main criteria for judgment are threefold: 1. First is the 'explosive increase' in call volume. In the first half of 2025, the average daily call volume of Chinese enterprise-level large models reached 1018.65 billion tokens, a surge of about 363% compared to the second half of 2024, marking the full release of market demand. 2. Second is the shift in market focus, with the industry transitioning from 'technology-driven' to 'value-driven', and market attention shifting from 'extreme performance competition' to 'equal emphasis on scenario fit and commercial value'. 3. Finally, the core pain points faced by enterprises have undergone a structural change, and enterprises have entered the 'deep water zone'. The main obstacles have shifted from high costs in the past to 'unclear application scenarios' and 'difficulty in system integration', indicating that enterprises have passed the technology trial phase and begun to face the challenges of deep integration.

From the perspective of model call volume, the industry pattern shows a highly concentrated feature, with the advantages of domestic manufacturers 'accelerating solidification'. Alibaba Tongyi (17.7%), ByteDance DouPao (14.1%), and DeepSeek (10.3%) together account for more than 40%. The key factors determining call volume lie in the ecosystem and differentiated strategies: Alibaba Tongyi relies on its 'integrated deployment capabilities' and the delivery closed loop formed by Alibaba Cloud's foundation and PAI platform; ByteDance DouPao transforms enterprise-level large model calls through rapid iteration and layout of application construction platforms such as Coze and HiAgent; DeepSeek quickly breaks through with high cost-effectiveness and open source compatibility. At the same time, the impact of open source and ecosystems is extremely profound, and open source is becoming the preferred path for enterprise model selection. Driven by TCO pressure and data sovereignty demands, up to 70% of enterprises plan to increase open source models more in the future.

Given the agglomeration effect of top talents, high capital investment barriers, and computing infrastructure barriers, the future market pattern is expected to further converge towards leading manufacturers. Latecomers face extremely high competitive barriers and find it extremely difficult to catch up. If they seek a breakthrough, they need to deeply cultivate in niche tracks, such as in private deployment operation and maintenance response capabilities, in-depth customization of industry solutions, and the accumulation of expertise in vertical fields to build differentiated advantages.