Aptos and Jump Crypto Launch Shelby: The Verifiable Hot Storage Network That Could Reshape AI Data Infrastructure
Every AI model is only as trustworthy as the data it was trained on — yet today, there is no reliable way to prove where that data came from, who owns it, or whether it arrived intact. Aptos Labs and Jump Crypto believe they have built the missing layer. Their new protocol, Shelby, is the world's first verifiable global object storage network designed specifically for AI read workloads, and its early-access testnet is now live.
The Problem No One Solved: Hot, Verifiable, Decentralized Storage
Decentralized storage is not new. Filecoin has been archiving exabytes of cold data since 2020. Arweave offers permanent, pay-once storage for immutable records. But neither was designed for the workloads that dominate the 2026 AI economy — workloads that require millisecond reads, global availability, and cryptographic proof of provenance on every single request.
Think of an AI training pipeline that pulls millions of image–label pairs from a distributed dataset. Or a DePIN network of 50,000 sensors streaming telemetry to inference models in real time. Or a social platform serving personalized content feeds to millions of users simultaneously. In all these cases, the data must be hot: immediately accessible, not archived on tape drives waiting for a retrieval deal to clear.
Cloud providers like AWS S3 handle hot storage effortlessly — but they offer zero verifiability. You cannot prove that the object you received is the same one that was originally stored, that no intermediary tampered with it in transit, or that the content creator actually consented to its use. As AI regulation tightens globally — the EU AI Act now requires training data provenance documentation — this gap is becoming a liability.
Shelby fills it by combining three ingredients that have never been assembled into a single system: cloud-grade read performance, cryptographic verification receipts, and decentralized infrastructure that no single entity controls.
How Shelby Works: Architecture of a New Storage Primitive
Shelby's design reflects lessons learned from both Web2 infrastructure and blockchain protocol engineering. Its architecture cleanly separates the control plane (coordination, payment, verification) from the data plane (actual storage and delivery), a pattern familiar to anyone who has operated cloud infrastructure at scale.
The Data Plane: DoubleZero and Jump Crypto's Edge
The physical layer is perhaps Shelby's most underappreciated advantage. Rather than relying on the public internet — where latency is unpredictable and bandwidth is shared — Shelby runs on DoubleZero, a dedicated fiber backbone spanning 30+ cities across five continents. This is the same class of network infrastructure that high-frequency trading firms use for microsecond-level execution.
Jump Crypto, one of crypto's most technically sophisticated firms, brings storage infrastructure hardened in quantitative trading environments where every millisecond of data retrieval latency translates into real financial cost. The combination of dedicated fiber and trading-grade storage nodes enables Shelby to promise sub-second reads from any region.
Data is encoded using Clay erasure coding, a technique that minimizes replication overhead while maintaining high availability. Unlike simple replication (store three copies), erasure coding splits data into fragments and distributes them across nodes so that any subset of fragments can reconstruct the original. This means lower storage costs per gigabyte and minimal repair bandwidth when nodes go offline.
The Control Plane: Aptos as the Verification Layer
Every read request served by Shelby can return a cryptographic receipt — a verifiable proof that documents:
- What was delivered (content hash)
- When the delivery occurred (timestamp)
- Where the data originated (storage node identity)
- Under what rights the access was granted (licensing and consent metadata)
These receipts are anchored on the Aptos blockchain, which provides the settlement and coordination substrate. Aptos was chosen for its performance characteristics: 600-millisecond finality, 30,000 transactions per second, and gas fees as low as $0.000005. This means that anchoring millions of verification receipts per day is economically viable — something that would be prohibitively expensive on Ethereum mainnet.
While Aptos is the native coordination layer, Shelby is designed to be chain-agnostic. Support for Ethereum, Solana, and other chains is planned, allowing applications on any blockchain to leverage Shelby's verified storage without migrating their entire stack.
The Economic Model: Pay-Per-Read Micropayments
Shelby introduces a pay-per-read economic model through micropayment channels, a departure from the upfront storage deals used by Filecoin or the one-time permanent fees charged by Arweave. This aligns incentives with actual usage: storage providers earn revenue proportional to how frequently their data is accessed, creating natural economic pressure to keep popular data hot and available.
This model is particularly well-suited for AI workloads, where a training dataset might be read billions of times during a single training run but then accessed infrequently afterward. Under Shelby's model, the cost tracks actual demand rather than requiring upfront commitment for archival permanence.
Why AI Needs Verifiable Storage Now
The timing of Shelby's launch is not accidental. Three converging forces are creating urgent demand for verifiable data infrastructure.
The Data Provenance Crisis
As AI models grow more powerful, the question "what data was this model trained on?" has moved from academic curiosity to regulatory requirement. The EU AI Act mandates documentation of training data sources. The US Executive Order on AI Safety emphasizes data provenance. China's generative AI regulations require content traceability.
Yet today's infrastructure provides no systematic way to enforce these requirements. Training data is typically downloaded from S3 buckets, scraped from websites, or licensed through opaque agreements with no cryptographic chain of custody. Shelby's verification receipts create an auditable trail that can prove, on-chain, that every piece of training data was accessed with proper consent and licensing.
The AI Data Market Opportunity
The market for AI training data is projected to exceed $50 billion by 2030. But the current marketplace model is broken: data providers have no way to verify that buyers are honoring usage restrictions, and buyers have no way to verify that data hasn't been tampered with before delivery.
Shelby enables a new category of verifiable data marketplaces where consent, attribution, and licensing terms travel with the data object itself. A dataset owner can publish training data to Shelby with embedded licensing metadata, and every subsequent access generates a cryptographic receipt that proves compliance — or exposes violations.
The DePIN Data Explosion
The DePIN sector has grown from a $5.2 billion market cap in 2024 to over $19 billion in 2025, representing nearly 270% growth. CoinGecko now tracks 250+ DePIN projects generating sensor data, telemetry, and user content at massive scale.
These networks need hot storage that is both fast enough for real-time inference and verifiable enough to prove data authenticity. A weather prediction model consuming readings from 10,000 decentralized weather stations needs to know that each reading is genuine, unmodified, and delivered from the claimed geographic location. Shelby's architecture — combining low-latency delivery with per-read verification — is purpose-built for this use case.
Shelby vs. the Decentralized Storage Landscape
Understanding where Shelby fits requires examining how the decentralized storage market has matured.
| Feature | Filecoin | Arweave | Akave Cloud | Shelby |
|---|---|---|---|---|
| Primary use case | Archival / cold storage | Permanent immutable storage | S3-compatible enterprise backup | Hot storage for AI reads |
| Latency | Seconds to minutes | Seconds | Sub-second | Sub-second |
| Verification | Proof of storage | Proof of access (bundled) | Cryptographic audit trails | Per-read cryptographic receipts |
| Cost model | Storage deals (upfront) | One-time permanent fee | Pay-as-you-go (80% cheaper than AWS) | Pay-per-read micropayments |
| Network | Public internet | Public internet | Public internet | Dedicated fiber (DoubleZero) |
| Best for | Long-term archival, backups | Permanent records, publishing | Enterprise cloud replacement | AI pipelines, DePIN, streaming |
The key insight is that these are complementary, not competing, solutions. Filecoin excels at long-term archival. Arweave is ideal for data that must exist forever. Akave provides a familiar S3-compatible interface for enterprises. Shelby occupies the previously empty quadrant: decentralized storage that is both fast enough for production AI workloads and verifiable enough to satisfy regulatory requirements.
What Developers Can Build
The Shelby early-access testnet opens several new application categories.
AI Data Marketplaces: Platforms where data providers publish datasets with embedded licensing terms, and every model trainer's access generates verifiable compliance receipts. This could transform the legal landscape around AI training data rights.
Verifiable AI Training Pipelines: Organizations can prove to regulators and auditors that their models were trained exclusively on properly licensed data, with a complete on-chain audit trail from data ingestion to model deployment.
Real-Time DePIN Applications: Sensor networks, autonomous vehicle fleets, and IoT meshes can stream data through Shelby with per-read verification, enabling downstream AI models to trust their input data without relying on centralized gatekeepers.
Decentralized Content Delivery: Social platforms and media applications can serve content globally with sub-second latency while maintaining cryptographic proof of content origin and creator rights — a potential solution to the deepfake attribution problem.
The Road Ahead
Shelby's early-access testnet represents the beginning of a multi-phase rollout. A developer-focused devnet launched in late 2025, and the current public testnet allows AI teams and developers to begin integrating Shelby into real workloads. Full production launch is expected later in 2026.
The broader blockchain AI infrastructure market is projected to grow from $6 billion in 2024 to $50 billion by 2030, a 42.4% compound annual growth rate. Within that market, verifiable data infrastructure — the ability to prove not just that data exists, but that it was accessed correctly, consensually, and without tampering — may prove to be the most valuable primitive.
If Shelby delivers on its architecture's promise, it won't just be another storage protocol. It will be the trust layer that the AI economy requires but has never had — the cryptographic bridge between "data available" and "data verified."
As blockchain infrastructure continues to evolve with protocols like Shelby pushing the boundaries of what's possible on Aptos, developers need reliable access to the networks powering these innovations. BlockEden.xyz provides enterprise-grade RPC and API services for Aptos and 20+ other chains, giving builders the foundation to integrate with next-generation protocols.