NVIDIA has announced Scale-In, a new network infrastructure designed to address the specific demands of agentic AI factories. This infrastructure aims to accelerate, secure, and unify north-south access within these environments. Scale-In integrates BlueField-4 DPUs, DOCA software, and Spectrum-X Ethernet to provide dedicated, host-independent acceleration for critical infrastructure services.
Traditional cloud infrastructure was designed for predictable, general-purpose workloads. However, agentic AI factories connect diverse users, agents, applications, data sources, and storage systems to massively accelerated compute, requiring multi-terabit bandwidth per server. This necessitates dedicated DPU processing for line-rate networking, storage, and security.
Key Points
- NVIDIA introduced Scale-In as the fifth pillar of its AI networking infrastructure.
- Scale-In leverages BlueField-4 DPUs, DOCA software, and Spectrum-X Ethernet.
- BlueField-4 provides dedicated, host-independent acceleration for policy enforcement, storage access, security, and telemetry.
- BlueField-4 supports up to 800 Gb/s throughput and offloads critical infrastructure services from host CPUs.
- The integration of DOCA microservices and Spectrum-X Ethernet enables programmable, high-performance, and policy-driven infrastructure operations.
- Scale-In evolves north-south networks into a coordinated infrastructure domain for the AI factory.
- BlueField-4 serves as the infrastructure processor for Scale-In and the data and storage processor for CMX.
Context
According to the NVIDIA Developer Blog, agentic AI factories have distinct infrastructure requirements that traditional cloud infrastructure, designed for predictable workloads, cannot fully meet. These factories bring together massive accelerated compute with a growing number of users, agents, applications, enterprise data sources, and storage systems, all interacting continuously and at scale. The infrastructure must be accelerated and co-designed as part of the AI factory, with security, multi-tenant networking, data and storage access, and infrastructure operations working as a unified domain.
Why It Matters
This development addresses the increasing demands placed on network infrastructure by agentic AI factories, which require specialized solutions for security, data movement, and operational efficiency. Builders and operators of AI factories can use Scale-In to manage the complexity and performance requirements of these advanced AI workloads, ensuring consistent access to AI services and data as demand grows.
What To Do
- Note the role of BlueField-4 in providing host-independent acceleration for infrastructure services.
- Observe how DOCA software and Spectrum-X Ethernet integrate to support programmable and high-performance operations.
- Consider how Scale-In's approach to north-south networks might impact the design of future AI factory architectures.
- Watch for further details on how BlueField-4 accelerates services across GPU servers, agentic CPU systems, AI factory storage systems, and cloud services.