Many people evaluate storage projects by racing to see which one is cheapest, most reliable, and most decentralized. Walrus is aligned with Arweave, Filecoin, and even AWS, but comparing them this way can lead to tunnel vision. The real issue isn't "who has better redundancy," but rather—the architecture of the public blockchain itself simply can't support massive, persistent, and transferable data as first-class citizens.

What is the true situation of on-chain applications now? Ownership and transaction records are written on the chain, but what about actual images, audio, model parameters, game map states, sensor data? Most of these are not the core data on the chain; they only contain links pointing to the backend.

Where is the pitfall here? When the pointer behind it is a data center operated by a cloud provider or telecom operator, the so-called decentralization is just a shell of "settlement decentralization." Once they change policies, shut down services, or get banned, the "assets" recorded on the chain become a pile of waste paper— the ledger is intact, but the data has disappeared.

Walrus's idea is quite clear: don't treat storage simply as "cost reduction" or "performance competition." This is a fundamental flaw of the underlying infrastructure. For a public chain to truly support ecosystems that require continuous iteration and generation of new data (large-scale social applications, evolving game worlds, AI models and training data, data streams from physical network devices), it must build a truly reliable "long-term memory system"—not temporary fast storage, nor centralized resources that can be reclaimed at any time.