Walrus

• Walrus is a decentralised storage protocol built on Sui by Mysten Labs.

• Its Red Stuff erasure coding stores large files at substantially lower replication cost than competing networks.

• WAL is used to pay for storage and stake by node operators.

• Supply allocated across community, ecosystem and contributors.

• Storage is paid for in WAL; nodes stake it as a service commitment.

Pros:

• Erasure coding cuts storage overhead materially compared with full replication.

• Suits large files such as AI datasets, video and game assets rather than only small blobs.

• Built by the Sui core team rather than a separate project.

Cons:

• Sui ecosystem dependency for coordination.

• Competes against established decentralised storage networks with larger node counts.

• Storage demand from crypto applications is smaller than infrastructure capacity.

• Exposure to decentralised storage with a genuine cost advantage.

• Core team credibility from Sui.

• Weigh the competition and the size of actual storage demand.

• Erasure coding that stores large files with far less replication overhead than networks that keep full copies at every node.

• Decentralised storage networks replicate files across many nodes, making large files prohibitively expensive.

• Erasure coding reconstructs data from fragments, cutting the storage multiple substantially.

• Developed by Mysten Labs, the team behind Sui, founded by former Meta Diem engineers.

• Backed by Standard Crypto and Franklin Templeton.

• No public ICO. Distributed through ecosystem and community allocations in 2025.

• Sui ecosystem dependency.

• Established decentralised storage competitors.

• Storage demand smaller than available capacity.

• Node operator economics depend on sustained storage payments.

The Walrus paper addresses why decentralised storage has failed for large files: existing networks replicate each file at many nodes, so storing a terabyte costs many terabytes of network capacity, and the economics never work for video or model weights. Its Red Stuff encoding splits a file into fragments such that any sufficient subset reconstructs the original, reducing the replication multiple by roughly an order of magnitude while tolerating node failure. The paper is specific that this shifts cost from storage to reconstruction: retrieving a file requires gathering fragments from several nodes, which is slower than fetching a complete copy, and it argues that tradeoff is correct for large infrequently-read data and wrong for small hot data.