Walrus doesn’t replicate blobs the way people expect, and that’s deliberate.
When data is uploaded, Walrus splits it into fragments and encodes it using erasure coding. Each node stores only a piece, not a full copy. As long as enough fragments are available, the original blob can be reconstructed.
This avoids the waste that comes with keeping many full copies of the same data. Instead of multiplying storage costs, Walrus spreads responsibility. No single node matters, but the group does.
Recovery is built into the design. If some fragments go missing, the remaining ones are enough to rebuild the data. Nodes that fail to serve their part lose rewards, so the network naturally favors reliability.
The result is quieter than replication, but more efficient. Storage scales without exploding costs, and durability doesn’t rely on any one operator staying online forever.
Walrus chooses math over duplication.