<p>A few overlapping announcements this week point to a broader push toward production-ready AI infrastructure stacks - spanning networking, security, search, and orchestration.</p>
<h2>Elastic on Red Hat OpenShift with GPU-accelerated search</h2>
<p>Elastic’s GPU-accelerated vector search, built on NVIDIA cuVS, is now available on the Red Hat AI platform (OpenShift). The numbers are significant: up to 12x faster vector indexing and 7x faster force merging compared to CPU-only setups, achieved by offloading the compute-heavy work to GPUs.</p>
<p>The practical upshot is that organizations running RAG pipelines or agentic AI workflows can use Elastic Agent Builder and Elastic Workflows directly on Red Hat OpenShift AI, with real-time data retrieval and proprietary data staying within their own infrastructure. That last part matters for enterprises with sovereignty requirements - on-premises, cloud, or hybrid.</p>
<h2>Cisco embedding security into AI servers via BlueField DPUs</h2>
<p>Cisco and NVIDIA are taking a different angle on the same infrastructure problem: security. The partnership extends Cisco’s Hybrid Mesh Firewall to run on NVIDIA BlueField DPUs rather than host CPUs, which means 400G line-rate stateful segmentation without eating into the GPU or CPU resources the AI workloads actually need.</p>
<p>Cisco Nexus One handles network policy orchestration and pulls Kubernetes workload metadata to generate application-aware segmentation rules, enforced inline on the DPU. The architecture is specifically aimed at multi-tenant AI environments where you need strict inspection of front-end traffic and containment of lateral movement between workloads.</p>
<p>Controlled availability is planned for Q3 2026, with general availability in Q4 2026.</p>
<h2>The broader Cisco Secure AI Factory picture</h2>
<p>Zooming out, Cisco is framing all of this under its “Secure AI Factory” umbrella, which has expanded to cover the shift toward agentic AI and inference-heavy workloads. Beyond the DPU security work, the announcements include:</p>
<ul>
<li><strong>NVIDIA Spectrum-6 silicon</strong> support on new N9100 switches</li>
<li><strong>Edge AI</strong> support for enterprise and service provider use cases using NVIDIA RTX Pro Blackwell GPUs on Cisco UCS</li>
<li><strong>Red Hat AI Factory</strong> software integration</li>
<li><strong>Cisco Validated Designs</strong> for AI model training on FlashStack, FlexPod, and VAST Data</li>
</ul>
<p>The goal is a pre-validated architecture with security, observability, and governance built in across on-premises, edge, and cloud deployments - rather than bolted on afterward.</p>
<h2>What this adds up to</h2>
<p>These announcements are clearly coordinated. Elastic handles retrieval and search at the data layer. Cisco handles networking and security at the infrastructure layer. Red Hat provides the orchestration platform. NVIDIA supplies the silicon underneath all of it. Each piece is validated to work with the others.</p>
<p>Whether this kind of tightly integrated stack is appealing or concerning probably depends on how much you value vendor consolidation versus flexibility. The performance gains are real, but so is the lock-in.</p>


Collections

Cisco, Elastic, Red Hat, and NVIDIA have announced coordinated partnerships to build production-ready enterprise AI infrastructure. Elastic's GPU-accelerated vector search (via NVIDIA cuVS) is now available on Red Hat OpenShift, delivering up to 12x faster vector indexing for RAG and agentic AI workloads while keeping data on-premises. Cisco and NVIDIA are embedding security directly into AI servers by running Cisco's Hybrid Mesh Firewall on NVIDIA BlueField DPUs, enabling 400G line-rate stateful segmentation without consuming GPU/CPU resources. Cisco's broader 'Secure AI Factory' umbrella also adds NVIDIA Spectrum-6 switch support, edge AI with RTX Pro Blackwell GPUs, and Red Hat AI Factory integration. The combined stack covers retrieval, networking, security, and orchestration in a pre-validated architecture, though it comes with notable vendor lock-in tradeoffs.

Elastic and Cisco expand AI factory integrations with NVIDIA and Red Hat OpenShift