Photo taken by the author during the 2025 World Artificial Intelligence Conference

AsianFin -- China's leading technology companies—Alibaba, Tencent, ByteDance, and Meituan—are intensifying their focus on artificial intelligence (AI) agents, signaling a strategic shift towards generative AI applications. These advancements are not only enhancing the capabilities of AI systems but also driving significant investments in cloud infrastructure and computing power.

Alibaba's Tongyi Lab on Tuesday unveiled AgentScope 1.0, a comprehensive agent development framework designed to streamline the creation, deployment, and management of AI agents. This platform aims to simplify the development process, enhance operational security, and provide transparent monitoring throughout the AI agent lifecycle.

In a parallel move, Tencent's Youtu Lab announced the open-sourcing of its agent framework, Youtu-Agent, contributing to the growing ecosystem of AI agent tools. ByteDance also made strides by launching its agent platform, Coze Space, on both the Apple App Store and Android app stores, marking its first direct offering of an AI agent product to consumer-end users.

Meituan introduced LongCat-Flash-Chat, a large language model boasting 560 billion parameters. By activating only a subset of these parameters, the model achieves faster inference speeds, making it particularly suitable for complex agent applications requiring extensive processing. Notably, LongCat-Flash-Chat outperformed models like DeepSeek-V3.1 and GPT-4.1 on the Agentic Tool Use benchmark.

The surge in AI agent development is driven by the increasing commercialization prospects of intelligent agent products. Multi-step reasoning capabilities are fueling a significant rise in token usage, subsequently boosting demand for cloud services and computing power. This growing demand is translating into tangible revenue growth for companies in the AI sector.

Manus, a company specializing in general-purpose AI agents, reported an annual recurring revenue (ARR) of $90 million since its launch in March, with projections to reach $100 million soon. This growth underscores the expanding market for AI agent applications.

Leading Chinese internet companies are making substantial investments in AI infrastructure. In the second quarter, Alibaba's cloud business generated revenue of approximately $4.6 billion, a 26% year-on-year increase. Capital expenditure for the quarter reached $5.3 billion, marking a 220% year-on-year rise. CEO Eddie Wu highlighted that over the past four quarters, the company has invested more than $13.5 billion in AI infrastructure and product R&D.

Tencent's cloud services division reported revenue of about $7.7 billion, with a 119% increase in capital expenditure. Baidu's cloud business revenue stood at $1.4 billion, accompanied by a 79% year-on-year rise in capital expenditure.

Collectively, these companies' capital expenditures in the second quarter exceeded $8.5 billion, reflecting a 168% increase compared to the previous year. This surge in investment highlights China's commitment to advancing AI computing power and infrastructure.

Despite China's significant investments, a report by investment bank Jefferies indicates that China's four major cloud service providers—Alibaba Cloud, ByteDance's Volcano Engine, Tencent Cloud, and Baidu AI Cloud—spent approximately $45 billion over the past 12 months. In contrast, major U.S. cloud companies, including Microsoft and Meta, collectively invested $291 billion, underscoring the disparity in AI infrastructure spending.

However, the report also notes that China is rapidly closing this gap. Since the fourth quarter of 2023, China's capital expenditure on AI infrastructure has been growing at an accelerated pace, increasing threefold by the second quarter of 2025. Furthermore, starting from the fourth quarter of 2024, the ratio of capital expenditure to revenue among Chinese cloud service providers has surpassed that of their U.S. counterparts.

Overseas tech companies are also ramping up their investments. Microsoft's capital expenditure in the second quarter reached $24.2 billion, a 27% year-on-year increase. Meta raised its minimum full-year capital expenditure for 2025 to $66 billion, while Google's capital spending for 2025 increased from $75 billion to $85 billion.

As AI agents become more prevalent, ensuring their security and protection is paramount. According to F5's "2025 State of Application Strategy" report, 96% of enterprises worldwide are deploying AI models, with 91% utilizing Web Application and API Protection (WAAP) to secure their AI and machine learning models.

Mohan Veloo, CTO for Asia Pacific at F5, emphasized the importance of safeguarding AI-powered applications, stating, "In the next one to two years, all app products will transform into AI-powered applications, so we need to protect not only today’s applications but also those future applications built on AI."

F5 has introduced its AI Gateway product, integrated into its Application Delivery and Security Platform (ADSP), to provide enhanced security for AI agent applications. The platform is designed to operate seamlessly across various infrastructure types, including local data centers, virtualized and hybrid environments, and cloud-native environments, ensuring secure and efficient AI operations.

The adoption of AI agents in China is gaining momentum, with enterprises in sectors such as finance, manufacturing, and retail leading the way. According to a report by IDC, 34% of surveyed Chinese companies are conducting testing and validation of AI agents, while 30% are entering the "significant investment + procurement and training" phase.

IDC forecasts that by 2028, the market size for enterprise-level AI agent applications in China is expected to exceed $27 billion, reflecting the growing integration of AI agents into business operations.

In the realm of physical AI, companies like NVIDIA are advancing the development of general-purpose and humanoid robots. NVIDIA's Jetson AGX Thor development kit and production module, launched in August 2025, delivers up to 2070 FP4 teraflops of AI computing power, facilitating real-time inference for high-performance physical AI applications.

However, the deployment of embodied intelligence and humanoid robots presents unique challenges. Professor Zhang Xianghong from Beijing Jiaotong University highlighted the complexities of data acquisition and processing in real-world scenarios, emphasizing the need for extensive data to enable robots to perform tasks autonomously.

Wei Tao, Vice President and Chief Technology Security Officer of Ant Group, underscored the security implications of integrating AI agents into physical systems, stating, "The security guarantees, complexity, and risk level have fundamentally changed."

The market for humanoid robots in China is projected to grow significantly, with estimated sales reaching over 10,000 units by 2025. The market size for embodied intelligence in China is expected to reach approximately $730 million in 2025, with projections to exceed $14 billion by 2030, accounting for nearly 45% of the global market share.