A new stack is emerging for how AI applications store and move data. It does not look like the cloud infrastructure we have spent the last decade building on. And we think the pieces that make it work are starting to come together.
Using AI generally requires more reading. Remember the first time you started using your favorite LLM? You might have had the same experience. You put in a short question or thought, and then have to go grab your favorite pair of Warby Parkers because what comes out is six paragraphs of well-researched, nuanced thinking.
A similar phenomenon happens with computer architecture. AI reshapes every infrastructure it touches. As soon as AI features enter production, the data layer stops behaving like traditional object storage. Reads multiply, retrieval requests fan out across datasets and regions, and the cost and performance of delivering data becomes the limiting factor.
Shelby is building storage for that reality. Shelby Early Access is now live in testnet, introducing verifiable global object storage designed for read-heavy workloads. Instead of copying data across regions or stitching together routing layers after the fact, Shelby focuses on serving data once and delivering it globally with predictable performance.
That architecture depends on something the public internet was never designed to provide: deterministic, low-latency delivery. That is where we come in. DoubleZero is the high-performance fiber network underneath Shelby. Dedicated bandwidth, reduced latency, optimized routing, and so much more. The reason Shelby's global read performance can actually deliver on its promise.
Instead of relying on best-effort internet routing, DoubleZero provides bandwidth and predictable routing across a global fiber backbone. This allows systems like Shelby to deliver data across regions without the instability that comes from traditional internet paths.
Where Shelby focuses on global storage and verifiable reads, DoubleZero ensures the delivery path remains consistent. Shelby’s architecture combines several layers of specialized infrastructure.
Every Network Problem
Global reads are only useful if the delivery path itself is reliable. Most distributed systems still rely on best-effort public internet routing to move data between regions. That routing was never designed for the kind of consistency required by large-scale read workloads. Performance varies. Latency shifts. Delivery paths change.
For applications that depend on predictable global access, those fluctuations quickly become the bottleneck. DoubleZero is a dedicated, high-performance network designed for distributed systems.
Shelby provides a global object storage system designed for AI retrieval patterns, regardless of the use case:
- Jump Crypto provides high-performance storage infrastructure tailored to the demands of quantitative trading environments, where reliability and throughput are non-negotiable.
- Aptos serves as the coordination and verification layer, anchoring data provenance and ownership in a consistent control plane.
- DoubleZero provides the network layer underneath it all, delivering the deterministic connectivity required for global read performance.
Each layer solves a different part of the same problem.
Together, they form the infrastructure required to serve data globally, reliably, and verifiably.
What Does Early Access Mean Here?
Shelby Early Access marks the beginning of the testnet phase. Developers can start experimenting with Shelby’s global object storage system and explore how read-heavy applications behave when the storage and networking layers are built specifically for those workloads. For DoubleZero, Shelby represents another tenant operating on top of the network.
The long-term vision for DoubleZero has always been multi-tenant infrastructure: blockchains, applications, and data systems all sharing a high-performance network designed for distributed workloads.
Shelby is an early example of that model in practice. We’re excited to welcome them to DoubleZero’s growing ecosystem and look forward to supporting them in their journey to mainnet with high-performance infrastructure.
Get Started
Shelby Early Access is live on testnet! Developers interested in building AI systems that depend on global reads can explore the documentation now in their Docs.