Aptos Labs and Jump have moved their Shelby platform into private beta, marking a significant shift toward real-world AI data infrastructure built on the Aptos Layer 1 blockchain.
Shelby shut down its testnet on August 3, 2026, and rather than closing operations entirely, the team invited selected builders into private production access.
Approximately eight weeks after the testnet closure, on September 29, 2026, the Aptos Labs team formally announced the transition into private beta.
The stated focus for this new phase centers on supporting real customer applications and optimizing the platform specifically for genuine AI data workloads.
Months of testing yielded largely positive results, with the core platform holding up under pressure, though real gaps emerged in tooling, APIs, and developer compatibility.
Shelby is designed as S3-compatible object storage, built to serve AI agents and distributed compute environments that require flexible, scalable data management.
A key architectural principle behind Shelby is that compute is stateless while data is not, meaning the data layer should not dictate where compute operations are ultimately run.
Under the hood, Shelby relies on Aptos smart contracts working in combination with RPC servers, private networks, and erasure coding to maintain data integrity.
One notable feature of the platform is that it creates a sequenced cryptographic record with each write operation, reducing the number of data copies required as compute moves across environments.
The breadth of Shelby’s testnet SDKs, which span React, Ethereum, Solana, and S3-compatible tools, signals a deliberate effort to meet developers within the ecosystems where they already operate.
For Aptos Labs, Shelby represents a direct test of whether the Aptos blockchain can function as a coordination layer for data-heavy infrastructure beyond its existing financial applications.
There is currently no confirmed native token, airdrop, or points program tied to testnet participation, with official communications revealing no token or tokenomics details at this stage.
