HPE has rolled out a super-sized package of hardware and software aimed at helping enterprise customers build and manage large AI infrastructures from the data center to the edge. At its Discover event in Las Vegas this week, the vendor announced HPE Juniper Networking QFX switches aimed at inferencing and scale-up architectures. It also deepened integration of its Juniper Networking data center switching and operations into its Mist AI engine and launched a unified, AI-native SASE platform.
“AI requires a solid architectural foundation, and the success of agentic AI in the enterprise depends on a modern networking foundation built for autonomous workflows, where network performance, reliability, and intelligence determine the effectiveness of the entire AI architecture,” Rami Rahim, executive vice president, president and general manager of networking for HPE, told journalists and analysts in a briefing before the event. “HPE is delivering that foundation, letting enterprises deploy agentic AI with greater control, confidence, security, and operational simplicity.”
New Hardware: QFX5140 Switch
On the hardware front, the company introduced the HPE Juniper Networking QFX5140 switch, aimed at the booming demand for AI inferencing and edge AI use cases. The 1RU, 16T QFX5140 fixed-configuration data center switch is designed for AI fabric, spine, leaf, and border leaf deployments. Its port density can be configured to support 24× 400G QSFP112, 8× 800G OSFP800 and 2x SFP28 ports, and it includes support for RDMA over Converged Ethernet (RoCEv2). Additional capabilities include congestion management features such as Priority Flow Control and Explicit Congestion Notification and dynamic load balancing — all features that enable effective GPU-to-GPU communications, according to HPE CTO Fidelma Russo.
The box fills out the mid-tier of the HPE Juniper QFX family, which includes the high-end 102T QFX5240/QFX5250 and entry-level 100GBE QFX 5100. This strategic gap filling ensures that HPE can offer a complete portfolio from cost-effective edge inferencing to massive-scale AI training fabrics. By integrating RoCEv2 and advanced congestion control, the QFX5140 addresses the latency-sensitive requirements of modern AI workloads, where every microsecond of delay can impact model performance and inference accuracy. The switch also supports telemetry streaming and programmable pipeline features, enabling fine-grained monitoring and automation through tools like HPE’s Data Center Director.
Integration with HPE AI Infrastructure
HPE also announced the QFX5252 module for its 72GPU-per-rack AMD Helios turnkey package aimed at AI training and high-volume inferencing. The package combines CPUs, GPUs and open Ethernet networking technology into a unified, high-end AI platform. The QFX family is part of a significant move aimed at tightening the integration between the networking gear HPE acquired from Juniper and its existing HPE AI infrastructure offerings. Specifically, HPE’s data center management platform, Data Center Director, now includes the QFX switch portfolio. The idea is to offer data center customers a more integrated, automated, and centralized view of all their network components to improve network visibility and speed troubleshooting, according to Russo.
This integration extends beyond simple management. By combining the Juniper QFX switches with HPE’s AI factory solutions—such as the HPE Cray XD supercomputers and the HPE ProLiant server line—customers can achieve a unified operational model. For instance, the telemetry from the QFX switches can feed directly into the HPE Performance Cluster Manager, enabling automatic adjustments to network paths based on GPU utilization and job scheduling. Such synergy is crucial for organizations deploying large language models or complex simulation workloads that require deterministic low latency and zero packet loss.
HPE Mist Integration and AIOps
Continuing that integration theme, HPE said it will integrate Juniper’s natural language Mist AI into HPE Aruba Central and vice versa, all fed by its core AIOps Marvis AI engine. Marvis collects telemetry and user state data from Juniper’s routers, switches, access points, firewalls, and applications to detect and resolve a broad range of enterprise networking problems. A key part of Marvis is its AI-based Marvis Actions component, which identifies and prioritizes network problem remediation. Overall, the move integrates AIOps across wired, wireless, and SD-WAN environments to proactively resolve issues, including trusted actions such as wired port remediation, to further extend autonomous operations across the HPE networking portfolio, Rami said. He also noted that Marvis Actions will be extended to Aruba Central by the end of the year.
Further, HPE said it will be melding the HPE Aruba CX switching portfolio with HPE Mist, giving CX customers capabilities such as AI-native visibility, zero-touch provisioning, wired assurance for Layer 2 access, service-level insights, and HPE Marvis AI-driven support, Rami said. This unification means that a single Mist dashboard can now manage both legacy Aruba campus switching and the new Juniper data center switches, reducing the operational silos that often plague large enterprises. The Marvis AI engine leverages a graph-based knowledge base built from millions of support cases and device telemetry, enabling it to predict failures before they occur—a capability that HPE calls “predictable proactive maintenance.”
HPE also expanded Mist data center capabilities to include predictive analytics for proactive maintenance of network components. For example, Mist can now use AI/ML technology to predict potential optics failures that would cause network outages. Mist also can now use an advanced reasoning AI agent for high-confidence remediation, Rami said. Agentic AI is used to continuously and autonomously reason across diverse data streams, including millions of TAC cases and a contextual graph database from HPE Networking Data Center Director, to deliver precise root cause analysis inside of the data center. “So think of this as Marvis AI engine for data center operations. So, here we’re combining telemetry, application flows, operational context, historical knowledge to understand rapidly the root cause and recommend next steps,” Rami said. “So the problems that once took hours if not days to diagnose can now be resolved literally in minutes or even proactively before anybody understands that there is an issue. Together these capabilities bring the self-driving network from where it started inside the campus to now inside the data center.”
The most important takeaway is that there’s real momentum behind HPE’s acquisition strategy to leverage the core of the Mist platform by expanding Marvis as the AI engine across the portfolio, Mike Leibovitz, senior director analyst in Gartner’s business technology and innovation group, told Network World. “They are continuing to innovate while simultaneously integrating that model into Aruba networking, into the data center, and out the branch.” “Agentic NetOps is the most exciting area of innovation in enterprise networking in more than 20 years, as organizations move toward increasingly hands-off operations. HPE is well aligned with Marvis as the AI engine across its portfolio,” Leibovitz said. “Most enterprises are starting this journey with their existing infrastructure vendors like HPE, Cisco, or Arista, but those with more heterogeneous environments are also investigating a growing set of infrastructure-independent software providers. The space is moving quickly, and leadership is still very much up for grabs.”
Unified SASE Efforts
HPE is also trying to simplify WAN access to data center resources. That’s the driving idea behind a new SASE Orchestrator package. It ties together the vendor’s SD-WAN and SSE with cloud security and a unified policy engine that will use AI to manage branch, remote user, and cloud connectivity from one place. With the policy engine, customers can set security policies once, for example, and deploy them across many sites. “The Orchestrator promises simpler operations with AI operations, faster zero-trust adoption, and a better user experience through intelligent traffic steering and application awareness,” Rami said. The Orchestrator leverages Marvis for real-time visibility into application performance and security posture across the entire WAN. It also includes automated remediation for common threats, such as blocking malicious traffic instantly based on AI-driven threat intelligence feeds. This is particularly important for organizations deploying AI agents at the edge that need secure, low-latency connectivity back to training or inference clusters.
Nvidia Updates
Beyond hardware and software enhancements, HPE also tightened its integration with core partner Nvidia, via its HPE Private Cloud AI, a turnkey AI factory co-engineered with the Nvidia. HPE Private Cloud AI delivers a preconfigured hardware and software stack featuring the latest Nvidia AI Enterprise software and blueprints. The package is being enhanced to help manage AI agent by adding support for Nvidia’s Agent Toolkit software, including Nvidia’s Nemotron open models, NemoClaw, and OpenShell secure runtime, to provide an agent operating system that reasons, lets customers monitor agent behavior, enforces policies, and reduces deployment risk, according to HPE. HPE is also bringing Nvidia Confidential Computing to the HPE AI Factory through HPE Services. Nvidia Confidential Computing protects models and private data during execution for on-premises or sovereign deployments, according to HPE. This combination of agentic AI management and confidential computing addresses two major enterprise concerns: governance and data privacy. With the ability to enforce policies on AI agent actions and protect sensitive training data, organizations can confidently deploy AI in regulated industries such as healthcare, finance, and government.
HPE Zerto and HPE Morpheus Additions
A new release of HPE Zerto Software lets customers identify any rogue agent action and utilizes data protection to rewind to a clean slate. The Private Cloud AI package now also supports secure local agent registration, providing customers with the ability to approve AI models, skills, and tools while adhering to centralized governance and security policies. This ensures that even if an AI agent is compromised, the organization can quickly roll back any malicious changes and restore operations from a known good state. The integration with Zerto also extends to the HPE AI Factory, where continuous data protection can be applied to GPU server state files and model checkpoints, minimizing downtime during training or inference runs.
Changes are also in store for HPE Morpheus, which lets enterprises manage virtual machines, containers, and cloud resources across multiple environments from a single control plane. Often positioned as an alternative to Broadcom’s VMware, HPE announced if customers own or buy HPE’s VM Essentials package for a year, it won’t charge anything. “We are announcing that as a customer goes through this transformation with HPE Morpheus VM Essentials, you don’t pay for the first year of licenses. You will get Zerto migration licenses during that period to help you move. And so what this does is it helps mitigate the double-bubble cost problem that customers see as they are looking to migrate from one platform to another,” Russo said. Zero-interest financing for HPE cloud ops software over three years further supports customer migration, Russo added. “This cost mitigation aims to accelerate customer adoption and ease transitions from legacy platforms,” Russo said. This aggressive pricing strategy underscores HPE’s determination to capture market share in the virtualization and cloud management space, especially as enterprises seek alternatives to Broadcom’s higher licensing costs.
The breadth of HPE’s announcements at Discover—ranging from new switches to unified SASE to AIOps enhancements—demonstrates a coherent vision: to become the foundational provider for enterprise AI networks. By integrating Juniper’s networking assets with its own AI and edge portfolios, HPE is positioning itself as a one-stop shop for organizations that want to deploy and manage AI workloads securely and efficiently. The emphasis on agentic NetOps and predictive maintenance reflects a broader industry trend toward self-healing networks that require minimal human intervention. As AI adoption accelerates across industries, HPE’s integrated approach could give it a competitive edge against rivals like Cisco and Arista, who are also racing to deliver AI-optimized networking solutions. With the new QFX5140, deeper Mist integration, and simplified SASE, HPE is betting that enterprises will value the reduction in operational complexity as much as the raw performance of the hardware.
Source: Network World News