When "AI Native" has become a buzzword in the tech world, Li Feifei, the head of Alibaba Cloud's database division, poured cold water on the trend. In a recent industry sharing session, he bluntly pointed out that claiming to be an "AI Native Database" at this stage is akin to a cognitive "Great Leap Forward." In his view, the AI track itself is still rapidly evolving, and no one in the industry has yet reached that ultimate form.

To put an end to this conceptual confusion, Li Feifei set two hard-core criteria for what constitutes an "AI Native Database": first, more than half of the database instances must be directly driven by AI Agents (intelligent entities); second, more than half of the database outputs should be Tokens. This statement quickly elevated databases from cold storage containers to "intelligent cores" deeply integrated with large models.

The industry logic of data is undergoing a profound transformation. Large models are consuming cold and warm data, but high-frequency updated "hot data" in enterprise business systems remains the high-value fuel that keeps intelligent engines running. Li Feifei pointed out that the future bottleneck may not be computing power, but data and storage, especially as hardware costs rise, databases must learn how to handle multimodal structured and unstructured data more intelligently.

Facing this wave, Alibaba Cloud did not choose aggressive slogans but instead adopted a steady and solid "4+1" evolution strategy. From building a multimodal AI data lake, to achieving real-time synchronization of metadata; from built-in model operators support, to making the database the best platform for AI Agents, Alibaba Cloud PolarDB is accelerating from "cloud-native" to "AI-ready."

This is not empty talk, as practical data has already given the most powerful feedback. In just a few months, the amount of Tokens consumed by the Alibaba Cloud Yaochi Database product system calling large model operators such as Bailian has seen explosive growth of more than 100 times. This "downward strike" in databases has already been fully implemented in core businesses in fields such as finance, automotive, and internet.

Behind this battle for definition lies a dual contest of technical routes and market positioning. When future database outputs are no longer neat tables but dynamic Tokens, we may have truly touched the threshold of the super-intelligent era.