Hewlett Packard Enterprise (HPE) has expanded its AI networking portfolio with new Juniper-based switching products, deeper integration of Juniper technology into its AI Data Center Solution, and additional automation capabilities designed to improve utilization of increasingly expensive AI infrastructure.
The announcements, unveiled at HPE Discover 2026, extend the company’s effort to integrate Juniper Networks into a unified AI infrastructure stack spanning data center networking, operations, security, and enterprise connectivity.
Networking Moves Deeper Into the AI Stack
At the core of the announcement is the expansion of the HPE AI Data Center Solution to include HPE Juniper Networking QFX switches managed through HPE Networking Data Center Director. The move pulls Juniper’s data center networking portfolio deeper into HPE’s AI infrastructure stack, which combines compute, storage, networking, software, and services.
Speaking with reporters at HPE Discover, Rami Rahim, executive vice president, president and general manager of networking at HPE, said networking has become a determining factor in the economics of AI infrastructure.
“People have come to the realization that if your network has congestion, reliability problems, what will happen is those GPUs that you spent hundreds of millions or billions on could be used at 75% utilization, 50% utilization, 25% utilization,” Rahim said. “Networking has truly become a force multiplier for massive AI data center investments.”
J.J. Kardwell, CEO of cloud provider Vultr, which recently signed a partnership with HPE, said networking has become increasingly important as AI clusters scale.
“The impact of the network is, frankly, larger than it’s ever been because of the architecture needs of those systems,” Kardwell said during a media briefing at HPE Discover.
Kardwell said the challenge extends beyond individual GPU servers to the networks connecting racks, clusters, and data centers.
“The network between those cabinets and across those clusters becomes the key factor in accomplishing these massive training and inferencing workloads,” he said.
Networking and AI Infrastructure Economics
Sameh Boujelbene, vice president at Dell’Oro Group, said networking is becoming a critical factor in AI infrastructure economics, as operators focus on turning expensive compute investments into production systems.
“AI infrastructure is no longer just a GPU race,” Boujelbene told Data Center Knowledge. “It is a systems race, and networking is becoming one of the key economic levers that determines who can turn raw compute into usable intelligence at scale and profitably.”
The new products span AI training clusters, inference environments, data center interconnects, and edge deployments, reflecting growing demand for networking architectures that support the full AI lifecycle.
HPE also introduced two new AI networking products. The HPE Juniper Networking QFX5140 switch targets inference clusters and edge AI deployments, while a new QFX5252 switch tray is designed for AMD’s Helios rack-scale AI platform. HPE said the products are intended to reduce networking delays that leave GPUs waiting for data rather than processing workloads.
Rahim said the acquisition has also enabled HPE to address both scale-out and scale-up AI networking architectures. While Juniper had already established positions in routing and scale-out networking, he said tighter integration with HPE compute systems has accelerated development of scale-up technologies such as those being deployed in AMD Helios.
Mist and Marvis Expand Across the Portfolio
HPE is also extending Juniper’s Mist platform deeper into the Aruba installed base.
The company said HPE Networking CX switches will now be supported in the Mist platform, providing customers with AI-driven visibility, automated troubleshooting, service-level insights, and Marvis AI actions. HPE is also bringing Marvis self-driving capabilities into Aruba Central, including automated remediation functions such as wired port troubleshooting.
New Mist capabilities include predictive analytics for optics and system failures, along with an AI reasoning engine that draws on operational telemetry, support cases, and network data to accelerate root-cause analysis and remediation.
Rahim said more than 80% of network incidents in self-driving deployments are now either automatically remediated or accompanied by immediate root-cause information for operators.
Kardwell said the growing complexity of AI infrastructure is increasing the value of automated operations.
“The cost of failure and downtime is unacceptable. The ability of self-healing is so important,” he said.
Security and Operations Converge
HPE also expanded integrations between networking and infrastructure management platforms. HPE Mist Networking Data Center Assurance is now integrated with HPE Compute Ops Management and GreenLake, providing operators with a more unified view of networking and compute infrastructure.
On the security side, HPE unveiled a unified secure access service edge (SASE) platform that combines SD-WAN and security service edge capabilities through a single management console. Built on HPE Networking EdgeConnect, the platform incorporates zero-trust access controls and AI-assisted operations intended to simplify security and network management.
One surprise since the deal closed, Rahim said, has been the speed of integration.
“Things that I thought would be really difficult – getting teams together, making them feel like they’re on the same team, making the tough decisions about roadmaps and so forth – have all gone easier than I expected,” he said.
