Analyst: Alastair Cooke
Publication Date: May 27, 2025
What is Covered in this Article:
- Juniper QFX and PTX switches provide high-density 800GB Ethernet, which can be optimized for demanding AI workloads.
- Apstra’s intent-based network management software delivers controlled deployment and operation of large Ethernet networks.
- Validated Designs and templates speed deployment and minimize purchasing risk for an AI datacenter network project.
The Event – Major Themes & Vendor Moves: AI Infrastructure Field Day is a semi-annual, invite-only event in Santa Clara organized by Tech Field Day. Independent industry experts join presenting companies to learn about product innovations. The event is live-streamed, and videos are published on YouTube afterward.
Make an AI-Ready Data Center With Help From Juniper
Analyst Take: Juniper Networks demonstrated a strong capability for enabling customers to design and deploy the networks required for generative AI workloads, from model training to ongoing inference. The primary pillars in the AI-ready data center are Juniper’s Ethernet switches and Apstra management software. The AI Infrastructure Field Day presentations demonstrated expertise that extends beyond Ethernet switches to the scale of AI infrastructure and the day-to-day operational tasks involved when AI becomes part of a production application.
AI Data Center Ethernet Focus
Juniper has a validated Design for building 800 GB Ethernet networks to interconnect GPUs at a massive scale, providing low-latency, low-jitter, lossless connectivity. The back-end GPU network is a critical infrastructure component for AI model training and inference, particularly with huge models. This network enables GPUs to exchange large amounts of data rapidly, directly, and without being limited by the CPU’s processing capabilities. The efficiency of the GPU network directly impacts the completion time of AI model training tasks. Juniper has a specialized, RDMA-aware load-balancing algorithm designed to avoid congestion during these large transfers.
AI servers, equipped with GPUs, may have fourteen network connections, each with a bandwidth of 100GB to 800GB, compared to the two 25 GB Ethernet connections in a typical server. High-density Ethernet switches in a non-blocking fabric are crucial for supporting the high port density of AI servers and the bursty nature of the GPU network. The optics (transceivers) for each AI server can consume over 150 Watts, in addition to the rest of the server’s power demands. To reduce power demands for network transceivers, Juniper offers low-power optics that consume 60% less power than standard transceivers. High-density switches from the Juniper QFX range enable the networking of large numbers of GPU-equipped servers using low-power optics. A “Rail Optimized” design connects GPUs with up to 32 thousand GPUs in a single cluster using the PTX switch range. These large networks can utilize Juniper’s RDMA-aware load balancing to avoid congestion on switch-to-switch links in the multi-layer network.
Apstra Network Automation
Deploying and operating large Ethernet networks can be complex and time-consuming when you build an AI-ready data center. Juniper’s Apstra network management platform simplifies the design, deployment, and validation of these networks, then enables automation in the ongoing operation of the network. Juniper’s presenters highlighted the incidence of mis-cabling when new networks are provisioned. Even if only 1% of the cables are patched to the wrong place, on average, every seventh AI server will have a patching error and require rework after initial installation. Apstra builds a cable map from the designed state and identifies when the deployed cabling does not match, allowing for resolution before the network is released for use, rather than requiring troubleshooting of any resulting network issues after production. This mechanism of describing a desired state, which Apstra then models and compares to the current state, is vital even after deployment, for ongoing operations and troubleshooting. The Apstra web console is commonly used for deployment and troubleshooting. There is also an API and integrations for automation tools such as Python, Terraform, and Ansible. Apstra allows networking changes to be integrated into an infrastructure as code practice and even a CI/CD pipeline to enable automated changes to be deployed to virtual networks.
What to Watch:
- HPE’s bid to acquire Juniper, along with the current Department of Justice investigation into the acquisition, may raise concerns among Juniper customers.
- Juniper’s major competitors, Cisco and Arista, have solutions for building large AI networks with Ethernet.
- NVIDIA also has designs utilizing their InfiniBand network fabric, although the company has been increasingly embracing Ethernet for AI networks.
- Recent studies have highlighted staffing issues as a barrier to the success of AI projects in enterprise organizations. While not directly related to Juniper, it is vital to the success of your AI project that you secure and grow AI expertise within your organization.
The Juniper presentations from AI Infrastructure Field Day are now available to view, and all past Juniper presentations can be found on their page on the Tech Field Day website. You can watch all the presentations from the four days of AI Infrastructure Field Day on the Tech Field Day website.
Disclosure: Futurum is a research and advisory firm that engages or has engaged in research, analysis, and advisory services with many technology companies, including those mentioned in this article. The author does not hold any equity positions with any company mentioned in this article.
Analysis and opinions expressed herein are specific to the analyst individually and data and other information that might have been provided for validation, not those of Futurum as a whole.
Other insights from Futurum:
DOJ Blocks HPE’s $14 billion Juniper Deal – A Competitive Setback or Market Win?
Juniper 800GB Ethernet for AI Training
Juniper AI-Native Networking Platform: Ready to Transform AI
Author Information
Alastair has made a twenty-year career out of helping people understand complex IT infrastructure and how to build solutions that fulfil business needs. Much of his career has included teaching official training courses for vendors, including HPE, VMware, and AWS. Alastair has written hundreds of analyst articles and papers exploring products and topics around on-premises infrastructure and virtualization and getting the most out of public cloud and hybrid infrastructure. Alastair has also been involved in community-driven, practitioner-led education through the vBrownBag podcast and the vBrownBag TechTalks.
