Most people assume decentralized storage means every node holds everything. Walrus never did that.
On Walrus, blobs are assigned to committees. Each epoch, the network selects a subset of storage nodes responsible for storing and serving a specific blob. Selection is weighted by stake, but it’s not winner-takes-all. More stake increases the chance of being selected, not exclusive control.
This matters because Walrus avoids global replication by design. Not every node stores every blob. That would be wasteful and expensive. Instead, committees are large enough to guarantee durability through erasure coding, but small enough to keep costs under control.
I picture it like rotating shifts in a data center. Different crews are responsible for different racks, and the roster changes over time. If a crew fails to show up, the system notices quickly.
The risk is coordination. Fewer nodes per blob means performance depends on committee health. The upside is efficiency that scales.
Walrus chose precision over brute force.