Architecture Overview
The Nexus Layer 1 consists of three fundamental architectural layers:1. Execution Layer

A new compute paradigm: The IVC machine
Rather than simply executing code, the Execution Layer is a machine for proving computation. It’s defined by three core properties:- Incremental verifiability: Every step in a computation produces a succinct proof that can be verified independently and aggregated recursively.
- Global parallelism: Computation is distributed across a dynamic, decentralized mesh of nodes — orchestrated through a DAG-style topology.
- Recursive aggregation: Proofs are recursively folded into higher-order proofs, enabling scalable throughput with constant verification cost.
Hierarchical architecture: Orchestration and delegation

- Delegators: Lightweight compute nodes. Any user running Nexus OS can contribute CPU cycles. These nodes execute tasks via zkVMs and emit zero-knowledge proofs.
- Orchestrators: Coordinators that manage delegators, aggregate proofs, and maintain high-quality throughput. Currently centralized for speed, they will transition to a permissionless market over time.
- Proof trees: The network forms a DAG where delegators feed into orchestrators, recursively combining local proofs into the Universal Proof — a single, succinct statement verifying global computation.
From theory to deployment
Nexus isn’t just conceptual. Multiple public testnets have proven the viability of the Execution Layer:- Millions of nodes have participated globally via Nexus OS.
- Full zkVM integration allows for arbitrary EVM-like workloads to be proven and verified.
- Universal Proofs — verifying entire blocks of computation — are in active development, with the first milestone imminent.
Why it matters
As AI models, rollups, and onchain infrastructure strain the limits of traditional blockchain compute, Nexus offers an alternative: trustless, scalable, verifiable execution. Developers can:- Run untrusted computation with cryptographic confidence.
- Delegate workloads to a global proving network.
- Scale applications without compromising on transparency.
2. Consensus Layer

- Early Phases: Tendermint-based consensus with delegated staking and single-ring validator topologies.
- Mid Phases: Introduction of orchestrator and re-delegator roles, supporting nested security rings and capital layering.
- Later Phases: Deployment of HotStuff-2 with epoch-based committee rotation, adaptive quorum selection, and probabilistic fork-choice rules — all designed to withstand long-duration adversarial coordination.

3. Storage Layer

How it works: Hyper-dimensional encoding and DAS
At the heart of the Storage Layer lies a novel approach to data encoding and retrieval:- Files structure enables distributed representation with strong cryptographic properties.
- Only a small subset of this encoded data is stored across the network.
- Retrieval involves sampling this subset and computationally decoding the file.
The architecture: Scaling from Ethereum to a universal state
The Storage Layer will be deployed across three main phases: Mainnet Season 1 – Replicated storage A familiar model where storage nodes replicate the entire L1 state, Ethereum-style. Simple and robust. Mainnet Season 2 – Delegated staking Inspired by Lido, this phase introduces storage orchestrators who manage nodes and stake to guarantee liveness. Nexus OS users can contribute spare storage to the network. Mainnet Season ∞ – Data availability sampling The final form. Storage becomes probabilistically sampled and externally accessible. Clients can request data directly, while the system retains resilience through redundancy and decentralized orchestration.A new kind of economic engine
With a target of $100B+ in storage-layer staking, this layer is designed to maximize indestructibility. Participants are incentivized not just to store data, but to secure its permanence and integrity over time. As AI systems become more powerful and more foundational to society, they must become more verifiable. That means:- Training data that can be audited
- Model checkpoints that can be proven
- Inference outputs that are reproducible
