One Year, 450TB, Zero Chill: Walrus Gobbles Data and Eyes AI's Long-Term Memory

Walrus turned one on March 27, celebrating 12 months of hitting milestones since launch like a degen chasing yields after a lucky flip. The decentralized data storage layer, built by Sui developer Mysten Labs, went live barely a week after the Walrus Foundation raised $140 million in a private round led by Standard Crypto, with a16z, Electric Capital and Franklin Templeton Digital Assets joining in. Sometimes the fundraising happens so fast you wonder if the term sheet was drafted on the same napkin as the birthday cake order.

Built on programmable storage and scalability, Walrus lets developers customize how their apps store and access data. Its Red Stuff encoding algorithm boosts throughput and resilience—and it's drawn serious adoption over the past year. Partnerships with real-world asset blockchain Plume, developer CCP Games, esports org Team Liquid, and Decrypt (now storing news articles, videos and photos on Walrus) have shown real-world utility. No, that's not a typo—they really are storing the news. On-chain. Good luck trying to edit that history.

That adoption drove serious growth. Walrus hit 409 TB of total data stored in early March, then passed 450 TB this week—beating Arweave's 385 TB. For those keeping score at home, that's more data than most people's entire digital lives multiplied by roughly 450,000. We're in the terabyte era now, baby.

"The fact that we've now surpassed 450TB of unencoded data in under a year is meaningful precisely because the data comes from real organizations," said Rebecca Simmonds, Walrus Foundation's Managing Executive. That includes Team Liquid migrating 250TB of esports archives, Decrypt moving its media library, and Allium bringing 65TB of institutional-grade blockchain data from Bitcoin, Ethereum, Sui and other top networks. That's a lot of hot takes, match footage, and chain data—all happily living on decentralized infrastructure.

As AI models get faster and larger, an important question looms: can we verify the data behind their outputs? "Most AI systems rely on data pipelines that nobody outside the organization can independently verify," Simmonds noted. Imagine asking an AI for facts and just... taking its word for it. That's not a vibe.

Organizations pick Walrus partly because its erasure coding—breaking data into fragments—offers stronger fault tolerance at lower replication factors. "That translates directly into lower costs at scale, and it makes us viable for organizations storing hundreds of terabytes, not just small files," she said. Basically, less redundancy tax while still sleeping at night.

Walrus kept building post-launch, adding Quilt in July (efficient small file storage) and Seal in September (data privacy and access levels). Quilt optimized costs so well it actually reduced network revenue initially. "That responsiveness, combined with strong underlying tech, is what created the adoption flywheel," Simmonds said. Sometimes you gotta burn a little to earn a lot. The market respects the strategy.

Current Web3 storage has data public by default, encryption left to app developers, and access control that's "hard-coded, brittle, or missing entirely," according to Simmonds. Seal's mainnet launch aims to change that with decentralized secrets management. Finally, secrets that stay secret—even from the blockchain's prying eyes.

Walrus sees AI as its biggest growth opportunity. "As AI agents become more autonomous—executing financial transactions, making decisions on our behalf—it becomes critical that we can verify what data those agents used to make those decisions," she explained. Since Walrus data is verifiable, tamper-proof and always accessible, it could serve as a long-term memory layer for agentic AI.