China's AI Software Market Surges 25% in Q2 2025, Driven by Advanced AI Agents and Multimodal Models

Goldman Sachs has released a new report highlighting the robust growth of AI-native applications in China's software market during the second quarter of 2025. The report identifies three key drivers of this growth: AI agents, multimodal AI models, and model deployment. These technologies are propelling the Chinese software industry forward, with AI vendors transitioning from basic integration to more advanced, use-case-specific large language models (LLMs).

The report underscores the shift in the AI landscape, where companies are moving beyond foundational models to develop applications tailored to specific needs. This evolution is evident in the increasing sophistication of AI agents, which are becoming more capable of handling complex tasks and interactions. Multimodal AI models, which integrate multiple types of data, are also gaining traction, enabling more comprehensive and accurate AI solutions.

Model deployment, the process of integrating AI models into real-world applications, is another critical area of growth. As AI models become more advanced, the ability to deploy them efficiently and effectively is crucial for their commercial success. This trend is particularly notable in China, where the market for AI-native applications is expanding rapidly.

The commercialization of AI agents is also beginning to take shape. These intelligent entities are designed to perform tasks autonomously, interacting with users and other systems to achieve specific goals. As AI agents become more sophisticated, they are expected to play an increasingly important role in various industries, from customer service to healthcare.

AI agents are evolving from being mere concepts discussed in edge devices and enterprise settings to becoming commercially viable products. These agents are not just an addition of AI to existing systems but are poised to become new user interfaces for enterprises and knowledge workers. Their core capability lies in responding to and even anticipating environmental changes, planning, and executing workflows autonomously.

For instance, a private company has launched a general AI agent capable of independently completing complex tasks such as resume screening and property searches, with a monthly subscription fee of 39 dollars. Additionally, software providers are accelerating the integration of AI agents into their professional platforms. Kingdee, for example, introduced five specialized AI agents for finance, travel, and recruitment in May. Meitu also plans to release an AI agent in the second half of 2025 to enhance the user experience of its entertainment and productivity tools.

Despite the strong demand for AI applications in the enterprise sector, the consumer market faces significant challenges in monetization. The report reveals that the conversion rates for consumer-facing AI tools are generally low. Even for tools with relatively better performance, such as WPS AI, the conversion rate is only 13%. Most tools struggle with conversion rates between 3% and 6%.

Moreover, the revenue contribution from AI features remains minimal. For instance, the AI functionality in Xunfei's Xinghuo contributes only 1% to its total revenue, while Meitu's KaiPa contributes 1%. Only a few mature products, like Wondershare's Filmora, have seen significant revenue from AI features, contributing 46% to its total revenue. However, most new AI applications are still exploring effective monetization strategies.

AI vendors are continuously advancing multimodal models to achieve lower costs, higher quality, and longer context lengths. The localization of EDA/IP software is also accelerating, presenting a reliable growth opportunity for AI vendors. Companies like Jieyuexing, Minimax, and ModelBest have released upgraded multimodal AI models in recent months. These models focus on generating and editing images, videos, audio, and 3D content, achieving professional-level results while reducing computational costs and entry barriers.

For example, Wondershare's "Tianmu 2.0" large model can achieve cinematic camera effects, while Desay SV and ModelBest have collaborated on an AI audio solution for smart cockpits, demonstrating the broad potential of multimodal technology in various applications. Additionally, companies are reducing model parameters and computational costs. Stepfun's Step1X-Edit image editing model, for instance, has only 19 billion parameters but ensures high efficiency and quality. The pursuit of high-definition visuals and precise physical simulations is also a new direction, as seen in Stepfun's Step1X-3D model, which focuses on 3D content generation.

EDA software, a core tool for chip design, is seeing accelerated product releases from domestic suppliers in the second quarter of 2025. This is driven by the growing demand for localization from simulation/digital foundries and fabrication plants. HuaDa Jiutian has released an upgraded parasitic parameter analysis tool and a Liberal IP tool for mixed-signal SoCs. Jiutongfang has launched digital simulation and verification tools and plans to offer trial periods to specific customers starting in June 2025.