---
title: "Scaling Agentic AI Factories Through Extreme Co-Design with NVIDIA BlueField"
description: "Agentic artificial intelligence workflows transform data center infrastructure by turning context management, security enforcement, and inter-service communication into active components of the inference data path. To prevent host central processing units and graphics processing units from stalling on these tasks, NVIDIA introduces BlueField-4 data processing units, Vera BlueField-4 STX storage processors, and the DOCA software platform. The BlueField-4 processor offloads Layer 3 routing, encryption, and telemetry using a sixty-four-core Grace processor with up to eight hundred gigabits per second connectivity. Simultaneously, the Vera BlueField-4 STX powers the NVIDIA CMX context memory platform alongside DOCA Memos to store, protect, and share key-value cache across flash storage tiers. This co-designed infrastructure isolates tenant traffic and accelerates context reuse without degrading host-side agent execution."
---

# Scaling Agentic AI Factories Through Extreme Co-Design with NVIDIA BlueField

[NVIDIA Developer Blog](https://yomu.fyi/company/nvidia-developer-blog) · Michelle Horton · Jul 16, 2026

**Type:** Explainer

## Summary

Agentic artificial intelligence workflows transform data center infrastructure by turning context management, security enforcement, and inter-service communication into active components of the inference data path. To prevent host central processing units and graphics processing units from stalling on these tasks, NVIDIA introduces BlueField-4 data processing units, Vera BlueField-4 STX storage processors, and the DOCA software platform. The BlueField-4 processor offloads Layer 3 routing, encryption, and telemetry using a sixty-four-core Grace processor with up to eight hundred gigabits per second connectivity. Simultaneously, the Vera BlueField-4 STX powers the NVIDIA CMX context memory platform alongside DOCA Memos to store, protect, and share key-value cache across flash storage tiers. This co-designed infrastructure isolates tenant traffic and accelerates context reuse without degrading host-side agent execution.

## Context

Agentic AI workflows require multiple model calls, tool executions, memory lookups, and persistent context management across distributed systems. Managing dynamic context such as KV cache in traditional host CPU domains causes latency bottlenecks and stalls GPU inference during long-context workloads.

## Approach / What changed

NVIDIA uses BlueField-4 DPUs, Vera BlueField-4 STX storage processors, and the DOCA software framework to offload networking, storage, security, and KV-cache management to dedicated hardware. The architecture implements DOCA Memos and the CMX context memory storage platform to create an active, Ethernet-attached flash tier for KV cache reuse.

## Takeaways

- BlueField-4 integrates up to 800 Gb/s connectivity, a 64-core NVIDIA Grace CPU, LPDDR5X memory, and PCIe Gen6 to offload networking, storage, and security from host CPUs.
- The NVIDIA CMX platform uses Vera BlueField-4 STX storage processors and DOCA Memos to manage an Ethernet-attached flash tier optimized for sharing and retrieving KV cache.
- Moving security controls to in-silicon processing on BlueField isolates tenant policies, network enforcement, and telemetry outside the host workload domain to prevent tampering.

**Tags:** [Architecture](https://yomu.fyi/topic/architecture), [Caching](https://yomu.fyi/topic/caching), [LLMs](https://yomu.fyi/topic/llm), [Performance](https://yomu.fyi/topic/performance), [Scalability](https://yomu.fyi/topic/scalability)

- Source: [NVIDIA Developer Blog](https://developer.nvidia.com/blog/scaling-agentic-ai-factories-through-extreme-co-design-with-nvidia-bluefield)
- Source URL: https://developer.nvidia.com/blog/scaling-agentic-ai-factories-through-extreme-co-design-with-nvidia-bluefield
- Ingested by Yomu: 2026-08-27T15:01:10.846Z

[Read original post](https://developer.nvidia.com/blog/scaling-agentic-ai-factories-through-extreme-co-design-with-nvidia-bluefield)