Walrus Hits 450TB in First Year, Launches MemWal SDK for Agent Memory
Walrus, the decentralized data storage protocol built by Mysten Labs on the Sui blockchain, is celebrating its first mainnet birthday this week โ and the numbers are substantial.
450TB and Climbing
In just 12 months since launch, Walrus has stored more than 450TB of unencoded data across partners including Team Liquid (250TB of esports archives), blockchain analytics firm Allium (65TB of institutional on-chain data), and media publisher Decrypt. That figure now exceeds the 385TB stored on Arweave, a decentralized storage network that has been live since 2018.
The protocol uses an erasure coding algorithm called Red Stuff, which breaks data into fragments and allows stronger fault tolerance at a lower replication factor โ translating to lower storage costs at scale. In July 2025, Walrus launched Quilt, a batching solution that made storing small files economically viable.
MemWal: Persistent Memory for AI Agents
The most consequential announcement tied to the anniversary is MemWal, a new SDK that allows AI agents to store and retrieve long-term memory on Walrus. Agents equipped with MemWal can persist knowledge across sessions in a tamper-proof, verifiable, and always-accessible format.
Rebecca Simmonds, the Walrus Foundation's Managing Executive, framed the AI storage market as the protocol's biggest opportunity: "As AI agents become more autonomous โ executing financial transactions, making decisions on our behalf โ it becomes critical that we can verify what data those agents used."
What's Next
Walrus also launched a limited-time boosted WAL staking vault on Slush Wallet to mark the anniversary. The protocol's roadmap for year two centers on deeper AI infrastructure integrations and expanding into on-chain finance data delivery.