HPE expands self-driving networking strategy as AI moves into production

Hewlett Packard Enterprise Co. today unveiled a broad set of networking and artificial intelligence infrastructure enhancements at its Discover customer conference, promising to help enterprises deploy AI agents at scale by implementing self-driving networks and establishing AI factories as the foundation of what it calls the “agentic enterprise.”

The combined announcements underscore HPE’s effort to position itself as a full-stack supplier for enterprises seeking to operationalize AI. Rather than focusing solely on model performance, HPE is emphasizing the infrastructure, governance, networking and operational capabilities required to manage large populations of AI agents securely and reliably.

The announcements span networking, security, AI infrastructure, data management and operations software. Together they reflect HPE’s belief that enterprises are entering a new phase of AI adoption in which autonomous agents will increasingly take actions rather than simply answer questions.

“AI doesn’t just answer. Now AI is starting to act,” said Rami Rahim, executive vice president and general manager of HPE’s networking business. “AI agents are moving into workflows, connecting to applications, using more data to make decisions.”

Rahim said many enterprise AI projects fail because organizations don’t think about architecture in advance.

“We truly believe that this is not just another AI adoption curve,” he said. “It’s becoming a technology spending shift, and it requires a solid architectural foundation.”

Network central

HPE said a central element of that foundation is networking. The company announced new capabilities that extend its self-driving networking strategy across campus, branch, data center and AI infrastructure environments. The company is integrating its Juniper networking portfolio more deeply into HPE AI Data Center Solutions, adding Juniper QFX switches and networking management software to its AI Factory offerings.

Among the new products are the HPE Juniper Networking QFX5140 switch, designed for AI inference clusters and edge deployments, and a new QFX5250 switch tray optimized for Advanced Micro Devices Inc.’s Helios rack-scale AI platform. HPE said the products are intended to reduce networking bottlenecks that can leave graphic processing units waiting for data rather than processing workloads.

The company is also extending AI-driven operations across its networking portfolio. Support for HPE Networking CX switches is being added to the Mist network management platform, while Marvis, HPE’s AI-driven networking assistant, is being expanded into Aruba Central.

“A year ago, we promised our customers that we would bring together the best of Juniper and Aruba,” Rahim said. “Today, we are delivering on that promise.”

New capabilities include predictive analytics for data center operations and an agentic AI-powered root-cause analysis engine designed to identify and remediate network issues before they affect users. HPE said the system continuously analyzes telemetry across infrastructure components and leverages operational data and historical knowledge to accelerate troubleshooting.

“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,” Rahim said.

Security is another major focus. HPE introduced a unified AI-native secure access service edge platform that combines SD-WAN and cloud-delivered security management into a single console. The platform is designed to accelerate zero-trust deployments while simplifying policy management and operations.

AI Factory enhancements

Beyond networking, HPE announced several additions to its AI Factory with Nvidia portfolio. The company is adding support for Nvidia Corp.’s Agent Toolkit software, Nvidia Nemotron models, Nvidia OpenShell secure runtime and Nvidia Confidential Computing technologies. The enhancements are intended to provide governance, observability and security controls for enterprises deploying AI agents in production environments.

The enhancements are meant to address customers’ needs for infrastructure designed specifically for agentic AI workloads, said Fidelma Russo, HPE’s chief technology officer.

HPE Private Cloud AI, the company’s turnkey AI platform developed with Nvidia, is gaining new governance and data-management features. These include secure local agent registration, model governance controls, enhanced data preparation capabilities and support for up to 256 GPUs for large-scale inference workloads.

The company is also integrating its Alletra Storage MP X10000 platform into Private Cloud AI. HPE said the storage system can automatically apply metadata and governance policies to unstructured data while improving inference performance and token processing efficiency.

A notable addition is expanded support for HPE’s Data Fabric software, which now includes integrations designed to make enterprise data more accessible to AI agents while preserving identity, access and security controls.

“We have the most proven data fabric software in the market,” Russo said, “and perhaps more important as we go into this era of AI is the speed at which it reacts in real time.”

To address concerns about AI failures and unintended agent behavior, HPE is extending its Zerto data protection technology to AI environments. The software can identify problematic agent actions and roll systems back to a known good state using continuous data protection capabilities.

The announcements also expand HPE’s broader GreenLake hybrid cloud platform. New integrations connect networking, compute and operations management tools, while new Morpheus and OpsRamp copilots provide natural-language interfaces for orchestration, observability and remediation tasks. HPE also announced a partnership with ServiceNow Inc. to integrate GreenLake Intelligence and OpsRamp capabilities with IT service management workflows.

“These are the essentials of an architecture for AI that doesn’t just provide answers, but AI that actually acts,” Rahim said.

Photo: Mark Albertson/SiliconANGLE