- HPE used Discover 2026 to expand its full-stack AI pitch
- The company debuted new networking, management, security and observability tools aimed at supporting agentic enterprise AI
- The message is competitive positioning: HPE wants enterprises to see it as a one-stop shop for AI
HPE isn’t just playing an AI hardware game against the likes of Cisco, Arista, Ubiquiti and Celestica. It’s peddling full stack enterprise optionality, and it just added a few more weapons to its AI arsenal at HPE Discover this week.
“The market is moving so fast and many projects will just fail if they’re not built with the right foundation,” HPE EVP, President and GM for Networking Rami Rahim warned on a call with press ahead of the conference.
According to Rahim, “The starting point for the foundation of what we call this agentic enterprise starts with a self-driving network, because agentic AI needs secure, adaptive connectivity across users, data, applications, clouds and, of course, data centers.” Additional pillars include infrastructure – everything from compute and data platforms to sovereign capabilities – and intelligent operations spanning observability and agent-driven ops, Rahim said.
To enable all of this, HPE is rolling out new management features, data center switches and more aimed at reducing bottlenecks and streamlining management of AI infrastructure, along with fresh security, governance and observability capabilities designed to help enterprises keep tighter control over autonomous AI agents.
Rollouts include new data center features aimed at enabling predictive analytics and agentic root-cause analysis, as well as new Juniper QFX switching options for AI environments, including products aimed at inference clusters and edge AI. HPE’s newly announced QFX5140 Switch is specifically designed for the next wave of AI infrastructure which will cater to inferencing at the edge, Rahim said. Already, HPE is seeing “explosive growth and demand” for inferencing capabilities, he added.
“We’re addressing networks for AI from pretty much every angle: scale up, scale out, scale across, edge on-ramps. And I believe we’re the only company doing it with an open standards-based Ethernet approach to every single layer,” Rahim said.
The timing of HPE's launch is notable as vendors across the infrastructure market race to prove they can support not just AI training clusters, but the emerging wave of inference-heavy, agent-based enterprise workloads.
A March 2026 Deloitte survey focused on the enterprise AI landscape found 64% of respondents have already started limited or at-scale deployments of AI factories. By 2028, 88% said they expect to have the same, while 73% think they’ll deliver rollouts fully at scale by then. Similarly, 36% of respondents said they have scaled AI at the edge, with 72% expecting to do the same by 2028.
“Cloud-managed edge, which relies on cloud orchestration with computing resources occurring locally on edge hardware, is the most common hardware strategy (reported by 68%). This may mean that many leaders want edge benefits without building an entirely separate operating model,” Deloitte wrote.
Building a bigger AI toolkit
HPE certainly appears to be trying to deliver a full-stack, hybrid platform to provide these capabilities.
The company integrated data center assurance with HPE Compute Ops Management, allowing customers to manage servers and networks fabrics using a single interface, and brought HPE Juniper Data Center Networking into its GreenLake cloud platform to allow for more unified and consistent management of compute, storage and networking. It also added support for HPE Networking CX switches in the HPE Mist platform, brought its AI-enabled Marvis Actions automation tooling to Aruba Central, and introduced SASE Orchestrator, a console unifying SD-WAN, SSE and cloud security controls.
Elsewhere, HPE expanded its AI factory portfolio, adding Nvidia reference designs for confidential computing, including country-specific options to ensure regulatory compliance for enterprises operating in different regions. It also upgraded its data platform and storage options for its HPE Private Cloud AI offering and introduced new GreenLake Intelligence copilots for orchestration and operations.
“Enterprises are not all just running AI. They are running traditional workloads as well,” Fidelma Russo said. She added that what HPE is doing is extending its GreenLake platform to enable it to be the operating platform for both kinds of workloads.
Taken together, the announcements show HPE pushing a full-stack argument against rivals in the crowded enterprise AI market. Rather than selling standalone servers or networking gear, the company is trying to package compute, storage, networking, security and management into a single AI factory story — one that leans heavily on tighter Juniper integration and its Nvidia partnership.
ZK Research's Zeus Kerravala told Fierce that "integrating the Juniper QFX switches into their HPE AI Factories and adding confidential computing capabilities to their Nvidia reference designs gives them a turnkey, highly secure full-stack offering that is competitive in the market."
But Kerravala said while HPE's full-stack agentic narrative is "compelling, there are a couple of gaps HPE still needs to address." These include "true platform convergence" between Aruba Central and Mist – which remain separate despite feature cross-pollination – and what Kerravala called "the multi-agent chaos problem."
"HPE is introducing a variety of specific copilots and automated agents (e.g., the Morpheus Orchestration Copilot and the OpsRamp Operations Copilot) that are supposed to talk to each other via the GreenLake Intelligence Mesh. While they claim the single Marvis engine prevents conflict, managing a multi-vendor, multi-agent enterprise ecosystem in the real world is incredibly messy," he said. "They are relying heavily on their new ServiceNow partnership to bridge this gap at the IT Service Management (ITSM) layer, but the proof will be in how smoothly those autonomous agents cross-communicate without creating conflicting infrastructure commands."