Cisco and NVIDIA Redefine AI Security with BlueField DPUs and Hybrid Mesh Firewall

Cisco Extends Hybrid Mesh Firewall to NVIDIA BlueField DPUs for AI Security

Cisco and NVIDIA have announced a partnership to address critical AI security challenges by integrating Cisco Hybrid Mesh Firewall with NVIDIA BlueField DPUs. This collaboration brings stateful segmentation directly into AI servers connected to Cisco Nexus One AI front-end fabrics, creating a hardware-accelerated security architecture that protects AI infrastructure without performance tradeoffs 1

1

. The solution tackles a fundamental problem: traditional centralized firewall appliances create traffic bottlenecks that don't scale to AI-level throughput, while host-based software agents tax CPU resources needed for AI processing 1

1

.

By running the firewall on an NVIDIA BlueField DPU rather than the host CPU, organizations reduce the risk of tenant tampering and preserve CPU and GPU cycles for AI workloads. The deployment enables 400G line-rate stateful inspection with fine-grained, flow-level observability, addressing the reality that in multi-tenant AI environments, nearly all traffic requires inspection 1

1

. This server-level enforcement model represents a significant shift from traditional approaches that struggle with the massive data throughput demands of modern AI training clusters.

Addressing the Security Tax in Enterprise AI Deployments

As AI workloads shift from experimental to mission-critical, the front-end network becomes the most important security boundary where users submit inference and training requests, storage systems exchange datasets, and multi-tenant workloads share servers 1

1

. Kevin Wollenweber, Cisco's senior vice president and general manager of data center and internet infrastructure, emphasized this transformation: "The network has gone from just plumbing and infrastructure to really a critical component to what enables these models to learn and think" 3

3

.

The Cisco Secure AI Factory with NVIDIA provides a uniquely integrated, security-first foundation that fuses AI security and observability directly into the network and every layer of the stack 2

2

. Unlike solutions that treat security as an afterthought, this approach ensures production-ready innovation is inherently protected from core to edge. Cisco Nexus One simplifies how network policy is built and deployed by discovering Kubernetes workload metadata and sharing that context with Cisco Hybrid Mesh Firewall, which translates it into application-aware, stateful segmentation rules 1

1

.

Industrializing the Token Economy with Validated AI Factory Stack

The partnership extends beyond security to address what Wollenweber calls "tokenomics" – the efficiency of token generation that will define competitive advantage in enterprise AI 3

3

. "The competitiveness for a lot of our customers is going to be around: how do we drive efficient token generation?" he explained, noting that organizations face a "complexity tax" when attempting DIY AI infrastructure approaches 3

3

.

Cisco has expanded customer choice by supporting Red Hat AI Factory software as an option within the Cisco Secure AI Factory with NVIDIA solution. This combination of NVIDIA AI Enterprise and Red Hat AI Enterprise software simplifies the development of next-generation agentic workloads by providing a single, integrated AI factory stack 2

2

. The engineering alignment between Cisco, Red Hat, and NVIDIA tackles the most persistent roadblock to enterprise AI adoption: production delivery friction. By upstreaming proprietary drivers and GPU operators directly into the RHEL ecosystem, the partners have eliminated the need for manual remediation 2

2

.

Preparing for the Agentic AI Era and Network Evolution

The shift to agentic AI – where autonomous agents communicate with other agents to execute complex workflows – creates new security challenges that the Cisco-NVIDIA partnership addresses 3

3

. By extending their Hybrid Mesh Firewall into the NVIDIA BlueField DPU ecosystem, Cisco places security controls at every GPU entrance, ensuring the "security tax" doesn't slow down token generation 3

3

.

Modern AI training clusters require unprecedented bandwidth, with some clusters now exceeding hundreds of petabits per second in total bandwidth 4

4

. The industry is quickly adopting 102.4Tbps silicon as the standard for large-scale AI factories, representing the new baseline for competitive AI clusters 4

4

. Linear-drive Pluggable Optics (LPO) is becoming more important by cutting power use by up to 50% per link, while Co-Packaged Optics (CPO) brings even bigger shifts with 30-40% less power use at the same speeds 4

4

.

The solution will be available during Controlled Availability in early Q3 CY26, with security being the first service delivered on the DPU before expanding to more AI-centric network services 1

1

. This positions organizations to scale multi-tenant AI infrastructure with confidence as they move from experimental AI to production deployments that demand industrial-grade reliability and high-performance infrastructure.