AI is reshaping how we course of knowledge, clear up advanced issues, and ship digital experiences. However your AI atmosphere is simply as safe because the infrastructure it runs on—and attackers know precisely the place to search for weaknesses.
As you scale AI workloads nearer to finish customers, brokers, and machines, a crucial problem emerges: you need to maximize GPU and CPU utilization whereas additionally defending towards subtle, fast-moving threats.
Conventional safety fashions battle in these environments. Centralized firewall home equipment can turn into site visitors choke factors that don’t scale to AI-level throughput. Host-based software program brokers may also tax CPU assets you want for AI processing—and, in some instances, introduce operational threat in multi-tenant environments.
To handle this, Cisco and NVIDIA are partnering to redefine AI safety. By extending Cisco Hybrid Mesh Firewall to NVIDIA BlueField knowledge processing items (DPUs), Cisco brings stateful segmentation straight into AI servers related to Cisco Nexus One AI front-end materials. The outcome is a sturdy, hardware-accelerated, server-level safety structure that helps cease threats earlier than they attain your knowledge—maximizing safety with no efficiency tradeoff.
With Cisco Hybrid Mesh Firewall, you possibly can outline coverage as soon as and implement it in all places. This unified safety mannequin spans bodily and digital firewalls, cloud environments, and now the DPUs inside your AI servers.
The front-end community: The true safety area
In AI infrastructure, a very powerful safety boundary is the front-end community, the place customers submit inference and coaching requests, storage programs trade datasets and checkpoints, and multi-tenant workloads usually share the identical servers. As a result of exterior site visitors enters right here, it’s the zone the place inspection and isolation matter most.
Entrance-end site visitors sometimes falls into two main flows:
- Consumer → Compute (inference and coaching)
- Compute ↔ Storage (knowledge ingest, dataset entry, checkpointing)
In AI environments, you possibly can’t assume solely “some” site visitors wants inspection. Almost all of it does, and multi-tenancy calls for strict segmentation. That requires segmentation that may function at full line charge throughout the front-end cloth.
Conventional centralized firewall home equipment break this mannequin. Hair-pinning site visitors to an exterior firewall will increase latency and creates bandwidth bottlenecks, successfully a choke level for all the cluster.
Bringing safety to the AI workload with DPUs
A greater mannequin is server-level enforcement utilizing DPUs. By operating the firewall on an NVIDIA BlueField DPU—not the host CPU—you cut back the chance of tenant tampering and protect CPU/GPU cycles for AI workloads.
Cisco is redefining AI workload safety by implementing unified safety coverage utilizing Hybrid Mesh Firewall on AI servers with NVIDIA BlueField DPUs. This allows:
- Air-gapped enforcement in multi-tenant and bare-metal environments
- {Hardware}-accelerated 400G line-rate stateful segmentation in DPU
- VPC-aware coverage enforcement on the community edge
- Superb-grained observability per circulate in {hardware} at scale
- Lateral motion containment, serving to block east–west assaults on the server boundary
Cisco Nexus One simplifies how community coverage is constructed, deployed, and saved aligned with workload id and context.
On every AI server, it discovers Kubernetes workload metadata and shares that context with Cisco Hybrid Mesh Firewall, which interprets it into application-aware, stateful segmentation guidelines:
- Native discovery (Nexus One): A unified administration airplane runs on every AI server to gather Kubernetes stock metadata—workload/software id, labels and annotations, namespaces, and so forth.
- Context-aware coverage (Hybrid Mesh Firewall): Makes use of the above metadata to generate application-aware, stateful segmentation insurance policies for every workload.
- DPU enforcement: Insurance policies are enforced inline on the NVIDIA BlueField DPU with out exterior brokers or software program.
- Kubernetes integrations: Optimized for the Isovalent Kubernetes suite (together with Cilium CNI and Hubble) and appropriate with normal Kubernetes environments.
“AI is reworking each trade, and the fast rise of AI factories is driving a rising want for cybersecurity at scale throughout enterprise infrastructure. By embedding Cisco’s Hybrid Mesh Firewall coverage into NVIDIA BlueField DPUs on AI servers, our joint prospects obtain high-performance, multi-tenant, intent-driven enforcement and hardware-accelerated safety, seamlessly related through Cisco Nexus One AI front-end materials.”
—Kevin Deierling, SVP of Networking, NVIDIA
Cisco Nexus One: Community coverage orchestration and visibility for AI front-end materials
Cisco Nexus One takes these capabilities additional by orchestrating advanced community insurance policies and sustaining end-to-end visibility with multisite implementations in AI front-end materials (as proven under). This simplifies operations, strengthens compliance enforcement, and gives a safety framework that scales as AI environments develop.
Constructing the safe AI manufacturing unit of the long run
AI factories succeed when safety retains tempo with AI-scale throughput. By operating Cisco Hybrid Mesh Firewall on NVIDIA BlueField DPUs, we offer distributed, in-server enforcement with 400G line-rate stateful inspection and fine-grained, flow-level observability—with out consuming CPU and GPU assets.
Paired with Cisco Nexus One for centralized community coverage and visibility, organizations can scale multi-tenant AI infrastructure with confidence, safe from the within out.
Safety is the primary service delivered on the DPU. Subsequent, we’ll broaden by including extra AI-centric community companies operating on DPUs.
Roadmap highlights
- Managed Availability: Q3 CY26
- Common Availability: This fall CY26
What’s new
- Cisco Nexus One: Community coverage and visibility
- Hybrid Mesh Firewall: Stateful segmentation on BlueField DPUs
- Splunk: Safety observability integration
To attempt the answer throughout Managed Availability in early Q3 CY26, please contact your Cisco account consultant.
