January 23, 2026

Building networks in readiness for AI workloads

Scott McKenzie
Solutions Architect

As outlined in our recent blog, Rethink your switch refresh, today’s switching decisions must consider more than just lifecycle and capacity. With many Cisco platforms nearing end-of-life, organisations are moving beyond identical replacements and considering infrastructure options that meet emerging demands. For those exploring or expanding on-premises AI workloads, this means building a network capable of supporting unprecedented levels of speed, throughput and data transfer. In this second blog in our series, we examine what it takes to prepare your switching environment for AI. (If you missed our first blog looking at the security scenario, you can access that here.

Why AI needs a new network

Training and running large AI models generate an entirely different class of network traffic. These workloads demand the following characteristics from the network:

  • High-throughput, low-latency connectivity between nodes and GPUs
  • Massive east-west traffic flows within the AI cluster
  • Consistent and lossless across thousands of connections
  • Scalable fabric-based designs that can evolve as models and datasets grow.

Networks designed for traditional application structures often struggle to meet the requirements that these AI workloads demand. Customers in sectors such as research, financial services and large-scale data analytics looking to deploy on-premises AI workloads are now evaluating how their switching infrastructure influences the time-to-value for AI projects.

Which Cisco switches are built for AI?

Not all switches are created equal when it comes to AI workloads. The advent of AI has also seen the introduction of different types of networks, including front-end, back-end, management, storage and inter-GPU, each with their own distinct challenges to overcome.

While Cisco Smart Switches are designed for flexible security and segmentation options, AI generally imposes higher demands, particularly in inter-GPU networks. For these purposes, Cisco and NVIDIA’s partnership extends the Spectrum-X architecture to include Cisco’s Silicon and Optics. This allows the introduction of features such as Per-Packet Dynamic Load Balancing to overcome traditional challenges, such as non-uniform utilisation of uplinks within the network.

Cisco includes the switches with the Silicon One chipset, such as the Nexus 9300-GX and 9300-SG series of switches and leverages Nexus Dashboard to monitor and manage the deployment.

Building AI-ready with the Nexus 9300-GX and 9300-SG offers the following:

  • Support for 800G and 400G port speed to support frontend, backend, and inter-GPU networks.
  • Congestion avoidance and lossless operation.
  • Spine-leaf architecture for scalable, resilient, and predictable performance.
  • AI/ML-aware networking protocols for adaptive traffic management.
  • Centralised visibility and control via Nexus Dashboard.

Providing a non-blocking, congestion-free, lossless network accelerates AI workloads and shortens completion times, enabling businesses to realise value sooner.

When is AI-ready switching the right path?

  • An AI-ready switching path is relevant if you are:
  • Running or planning to deploy on-premises AI/ML clusters
  • Focused on reducing model training or inference time
  • Scaling AI workloads across multiple GPUs or servers
  • Deploying a dedicated AI/ML environment within your data centre.

These environments require switches that are specifically designed for high-performance computing. While the broader Smart Switch portfolio plays a key role in securing and segmenting the management network, AI often demand more specialised infrastructure for the backend and inter-GPU networks.

Planning your refresh

If you’re evaluating switch options in the context of supporting AI projects and workloads, it’s essential to consider not just capacity, but how your switching infrastructure will impact your AI models. Start by mapping out the expected data flows and workloads over the next three to five years. Will your infrastructure support your business growth, or will it constrain it?

Data#3 can support this planning process with tailored infrastructure assessments. We’ll help you understand your readiness to support AI demands and, if not, the upgrade path you should consider.

To learn more, request a meeting with the Data#3 team using the form below or contact your Account Manager to schedule a straightforward, expert-led discussion designed to clarify your priorities and ensure you’re making informed decisions about your next refresh.

Contact us

Information provided within this form will be handled in accordance with our privacy statement.