Cisco Secure AI Factory with NVIDIA expands with NVIDIA RTX PRO 4500 Blackwell Server Edition GPUs
Enterprise AI is no longer a side project. Teams are putting AI into customer support, software experiences, security operations, manufacturing lines, and retail floors. That shift changes what infrastructure must deliver.
It is not enough to add GPUs. The real challenge in enterprise AI is putting GPU capacity to work efficiently across teams, workloads, and locations. Enterprises need an approach that scales in three ways at once:
- Technical scale: consistent performance for real workloads, especially inference
- Operational scale: fast provisioning, standard policies, and lifecycle control
- Financial scale: higher utilization and clearer ROI as adoption spreads across teams
At GTC 2026, Cisco is expanding support and announcing orderability for the NVIDIA RTX PRO 4500 Blackwell Server Edition across the latest Cisco UCS and Cisco Unified Edge platform. This expanded support will be aligned with a new set of Cisco Validated Designs (CVDs). The objective is simple: help customers deploy accelerated infrastructure faster, with fewer surprises, and with a repeatable path to production.
Why Cisco UCS and Cisco Unified Edge with NVIDIA RTX PRO 4500 are important for enterprises
Many enterprise AI deployments start with inference and data pipelines. The first wins usually come from putting models to work in day-to-day workflows, not from building a dedicated training cluster.
The NVIDIA RTX PRO 4500 Blackwell Server Edition is built for multi-workload environments, where AI inference, AI video, data processing, and computer vision may all need acceleration.
For enterprise IT, that translates into a more straightforward adoption curve: scale inference across more applications and more locations without turning every project into a facility redesign.
Cisco UCS turns RTX PRO Blackwell capacity into usable capacity
RTX PRO Blackwell Server Edition expands where you can place GPU acceleration across Cisco UCS, but availability alone is not the outcome. The outcome is getting the right GPU capacity to the right workload, quickly, repeatedly, and without creating a new operational snowflake every time a team asks for more inference, vision AI, or pipeline acceleration.
That is where infrastructure management becomes a strategic enabler. As AI deployments scale, the bottleneck shifts from acquiring GPUs to operational friction: slow provisioning when a new team needs capacity, inconsistent configurations across clusters, one-off builds that are hard to patch or reproduce, and low utilization because GPUs can't be reassigned fast enough to match demand.
Cisco Intersight directly connects to this moment with policy-driven control across the UCS portfolio, and with a capability that matters specifically for PCIe GPU deployments: Dynamic GPU Pooling. With policy-driven GPU sharing, organizations can allocate and reallocate PCIe GPUs in real time, such as GPUs hosted on the X580p PCIe node in Cisco X-Series modular systems, to the compute nodes that need them when they need them. The result is fewer stranded GPUs, higher utilization, and faster response when inference demand spikes or new projects come online.
The practical outcomes are speed and standardization at scale: teams can provision, repurpose, and share GPU resources using repeatable policies and profiles; reduce configuration drift across deployments; and maintain consistent visibility into inventory, health, and lifecycle status. In short, Intersight helps ensure the expanded RTX PRO Blackwell footprint translates into predictable operations and sustained throughput, not a bigger fleet that is harder to run.
Enterprise AI use cases accelerated by Cisco UCS with NVIDIA RTX PRO 4500 Blackwell GPUs
- GenAI inference for enterprise applications
Enterprises are moving beyond chatbots to production-grade agentic AI and retrieval-augmented generation (RAG) pipelines. By running inference for internal copilots and customer support assistants on the RTX PRO 4500, organizations achieve faster resolution of inquiries, reduced time spent searching for information, and more consistent, grounded answers across global teams. - Vision AI and AI video pipelines
Real-time computer vision workflows in retail, manufacturing, and logistics transform raw video streams into actionable intelligence. This acceleration leads to faster detection of safety and compliance issues, improved quality control that reduces waste, and significantly better situational awareness for operations teams managing physical spaces. - Accelerated data processing for AI pipelines
High-performance data preparation and feature engineering are often the primary bottlenecks in the AI lifecycle. Accelerating these steps ensures shorter iteration cycles for analytics teams, faster time to market for deployable workflows, and more predictable pipeline performance even as enterprise datasets continue to grow exponentially. - Visual computing and design workflows
3D design, engineering, and content pipelines demand high-fidelity visualization to maintain momentum. Providing GPU-accelerated workflows enables faster design iterations, higher productivity for teams working with complex visual assets, and seamless collaboration through responsive, low-latency visualization environments. - Media and video processing
Throughput and density drive the economics of modern media processing. Leveraging the ninth-generation encoders and AV1 support results in a significantly better cost per stream, allowing organizations to increase capacity and scale media workflows across teams without expanding their data center footprint or power budget. - Acceleration at the constrained edge
Deploying acceleration in locations where space, power, and operational simplicity are non-negotiable is now a reality with Cisco Unified Edge. This delivers lower latency by processing data at the source in factories, stores, and remote sites, while simultaneously reducing expensive backhaul requirements and ensuring resilient operations even when connectivity is limited.
Cisco UCS servers and Unified Edge compute nodes that support NVIDIA RTX PRO 4500 Blackwell Server Edition GPUs
The Cisco advantage: Powering secure AI factories around the world
NVIDIA GPU provides the "acceleration" layer, while Cisco UCS provides the infrastructure platform that provisions, secures, and operates AI workloads at scale. Through the Cisco Secure AI Factory with NVIDIA, we provide a modular reference design to deploy AI PODs as the infrastructure building block, that transforms raw data into business intelligence, securely, at scale, and with clear operational control.
Maximum density across the Cisco UCS portfolio
We are integrating the NVIDIA RTX PRO 4500 Blackwell across our M8 generation of servers to provide unmatched density, flexibility, and choice:
Cisco UCS PCIe GPU rack servers:
- Cisco UCS C845a M8: Our high-density powerhouse, supporting up to 8 RTX PRO 4500 Blackwell GPUs for massive parallel processing workloads.
- Cisco UCS C240 M8: A versatile 2U workhorse supporting up to 5 GPUs, ideal for balanced AI inference and data analytics.
- Cisco UCS C245 M8: Optimized for mid-range workloads with support for up to 3 GPUs.
Cisco UCS modular servers:
- Cisco UCS X-Series with X580p GPU PCIe Node: For those demanding modularity and composability, the X580p supports 4 GPUs per node. A full X-Series chassis equipped with two X580p nodes can support a staggering 8 RTX PRO 4500 Blackwell GPUs.
Cisco Unified Edge:
- One NVIDIA RTX PRO 4500 Blackwell can be installed in a Cisco UCS XE150c M8 compute node and up to two UCS XE150c M8 compute nodes can be installed in a Cisco Unified Edge XE9305 chassis.
Cisco Validated Designs: Turning hardware choice into a repeatable blueprint
The RTX PRO Blackwell Server Edition use cases only deliver real outcomes when they run the same way every time in production. That's why Cisco Validated Designs matter: they are the blueprint that turns broad AI applications into predictable deployments, with tested designs based on best-practices guidance from Cisco and its technology partner ecosystem that reduce risk and get you to production faster. For enterprise AI, validated designs matter because they provide clear bills of materials and configuration guidance, establish proven combinations across compute, networking, storage, and software, and make it far easier to replicate deployments consistently across sites and teams.
Recently published and upcoming CVDs and design guidance turn this launch into a proven, repeatable deployment playbook.
This includes a recently published Design Guide on AI PODs focused on the AI model training and fine-tuning use cases, Deployment Guide focused on FlexPod with Cisco AI POD configuration and Cisco AI POD for Enterprise Training and Fine-Tuning with Everpure Deployment Guide, also for AI model training and fine-tuning use cases.
One additional AI POD deployment guide with VAST Data for training and fine-tuning use cases will be released in March 2026.
The takeaway
RTX PRO 4500 Blackwell Server Edition expands Cisco Secure AI Factory with NVIDIA by giving enterprises a more accessible price, performance, and power profile for scaling GPU accelerated workloads, especially inference and multi-workload deployments that have to live within real data center and edge constraints. While NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs is the choice for maximum capability and headroom, NVIDIA RTX PRO 4500 Blackwell unlocks a broader set of energy-efficient AI use cases and makes it practical to extend GPU capacity to more teams, more sites, and more edge locations without forcing a platform redesign.
Across Cisco UCS, you get platform choice in rack and modular architectures, with Cisco Intersight helping standardize and operate infrastructure at scale. Add Cisco Unified Edge into the equation, and the same Secure AI Factory approach can be applied consistently from core to edge, with common operational workflows and guardrails. With NVIDIA RTX PRO 4500 Blackwell expected to be orderable in March, and upcoming CVDs delivering tested deployment paths, customers get a clear, lower risk way to adopt and expand quickly.
If you are building enterprise inference services, scaling vision AI, accelerating data pipelines, or pushing GPU capability out to additional sites and teams, this is the deployment approach that supports growth while keeping operations from becoming the bottleneck.
Visit Cisco at NVIDIA GTC 2026 to see UCS platforms with RTX PRO 4500 Blackwell support and learn how upcoming CVDs can accelerate your path to production.
Learn more about Cisco Secure AI Factory with NVIDIA