According to data released by global AI market tracker Aicpb.com, ByteDance's flagship AI chatbot Doubao saw its monthly active users (MAUs) plummet by 6.1 million in May following the introduction of a paid subscription option. Analysts suggest that the era of free AI services in China is far from over, and premature commercialization could jeopardize Doubao's leading position in China's highly competitive consumer AI market.
At the same time, AI powerhouse Anthropic has published a blog post calling for a global slowdown in AI development. Anthropic warned that the rapid pace of AI advancement could soon enable systems to achieve autonomous self-improvement without human intervention, posing severe societal risks. Internal data disclosed by the startup highlights the sheer velocity of their frontier models' capabilities.
On the hardware front, Nvidia CEO Jensen Huang confirmed that memory giants SK Hynix, Samsung, and Micron have all cleared qualification to mass-produce and supply next-generation HBM4 memory. These advanced chips are poised to fuel Nvidia’s upcoming Vera Rubin architecture, stepping up the compute power available for frontier AI research.
Domestically, Tencent is showcasing massive AI adoption. At the Tencent Cloud AI Conference, Executive VP Dowson Tong revealed that most of Tencent's code is now AI-generated, with developers shifting focus to high-level architecture design and oversight. Furthermore, Tencent Cloud's LLM portal, TokenHub, reported that its daily token consumption has exponentially scaled to break the 5 trillion mark within just three months of its launch.
[AgentUpdate Depth Analysis] The sharp drop in Doubao's MAUs after introducing a paywall, contrasted with Tencent's surging 5-trillion daily token consumption, highlights a fundamental shift in the AI landscape: consumer willingness to pay for basic chatbot wrappers is shrinking, while enterprise-level developer agent adoption is skyrocketing. For the AI Agent ecosystem, this signals a transition from 'chat' to 'action'. Future commercial value lies not in static Q&A assistants, but in autonomous agents that can seamlessly integrate into enterprise workflows. Meanwhile, Anthropic's warning on self-improving AI emphasizes the urgency of designing safe, steerable Multi-Agent coordination frameworks. As agents begin to recursively write and deploy code, establishing robust execution environments and hardware-accelerated safeguard layers will be the defining challenge for the next wave of AGI infrastructure.
