What to Expect at HPE Discover 2026

What to Expect at HPE Discover 2026

9 minutes read time

The Agentic Enterprise Gets a Full Stack, and a Third Answer to the One-Platform Question

  • HPE’s Discover 2026 blueprint centers on three layers: self-driving networks, agentic infrastructure, and intelligent hybrid operations, all under the GreenLake umbrella and powered by a single Marvis AI engine.
  • The third way: After Extreme Connect and Cisco Live, HPE completes the triptych. Same agentic destination, but HPE is the only one of the three explicitly refusing to merge its management platforms.
  • The wildcard: Morpheus 9 and a first-year-free migration program make HPE the most aggressive VMware alternative on the floor, a battle Cisco and Extreme are not fighting.

HPE briefed analysts last week ahead of Discover 2026, with Rami Rahim, EVP and GM of HPE Networking, and Fidelma Russo, CTO and EVP and GM of Hybrid Cloud, walking through what the company calls “a blueprint for AI that acts.” Note that the news is staged across three embargoes tied to the keynotes: the Morpheus platform migration program on June 15, self-driving networks and agentic infrastructure with Antonio Neri’s keynote on June 16, and intelligent hybrid operations with Russo’s keynote on June 17.

The framing will sound familiar to anyone who has followed the spring event circuit. AI is moving from answering to acting; agents are entering production workflows, and the underlying infrastructure was not built for it. Rahim was honest about the stakes: the market is moving so fast that many projects will simply fail without the right foundation, controls, and operating model. HPE’s answer is a three-layer architecture: self-driving networks for secure, adaptive connectivity; agentic infrastructure for compute, data, and sovereignty; and intelligent hybrid operations to run it all, all wrapped in advisory, professional, and financial services.

Here is what to expect, and how it reads from the Maravedis perspective on managed connectivity, with the obligatory comparison to Cisco Live and Extreme Connect, both of which I covered in recent weeks.

Self-driving networks: the Juniper integration delivers its proof points

This is Rahim’s first Discover as an HPE employee, and the networking news is essentially a progress report on the promise made when the Juniper deal closed: bring together the best of Juniper and Aruba without slowing innovation. Four announcements anchor it.

First, HPE Juniper Networking switches enter the HPE AI Factory portfolio, managed through HPE Networking Data Center Director (the platform formerly known as Apstra), creating a full-stack validated solution.

Second, a new QFX5140 switch, a 1RU (one rack unit), 16 Tbps platform built on Broadcom’s Trident5 silicon, purpose-built for inference clusters at the edge, with AI load balancing and congestion control, supporting speeds from 25G to 800G (gigabits per second). It is available now. This follows the QFX5250, which Rahim noted is already generally available as the industry’s highest-performance scale-out switch with full liquid cooling, ahead of every other OEM.

Third, and most relevant to readers of this blog, Marvis Actions is coming to HPE Aruba Central before the end of the year. Mist customers consistently describe Actions as indispensable because it does not just identify problems; it resolves them proactively. Extending that to the Central installed base, across wired, wireless, and SD-WAN, is the single most consequential item for the managed connectivity channel, where Central runs an enormous share of hospitality, MDU, and mid-market deployments. In the other direction, Aruba CX switching (the AOS-CX line) integrates with the Mist platform, offering zero-trust provisioning, AI-driven visibility, and Marvis Assistant support. The data center gets predictive analytics across telemetry (power, temperature, optics, system health) and agentic root cause analysis, effectively Marvis for data center operations, turning diagnoses that took hours or days into minutes.

Fourth, security: a new SASE (secure access service edge) Orchestrator, available later this year, brings SD-WAN and SSE (security service edge) into a single management experience with a unified policy engine. Together with NAC (network access control) and firewalls, HPE now claims a comprehensive built-in, not bolted-on, security portfolio.

The platform question: HPE’s answer is one brain, two bodies

Here is where the three-way comparison gets interesting. Extreme’s answer to the agentic era is Platform ONE, a single cloud platform that has been GA since July 2025. Cisco’s Cloud Control is a unified cross-domain destination that entered controlled availability on June 2. HPE is taking a different position, and Rahim was unambiguous: Mist and Aruba Central will not converge into one platform.

The logic is deployment models. Mist was purpose-built for cloud, Central supports on-prem, VPC (virtual private cloud), and private cloud. What converges is the experience: one AIOps (AI for IT operations) engine, Marvis, one data science team, and applications that are progressively cross-pollinated until the two platforms reach feature parity over years. The customer then chooses based purely on deployment model.

I asked Rahim directly how HPE prevents conflicting recommendations and actions as agents proliferate across Wi-Fi and wireless domains. His answer was architectural: there is only one AIOps engine and one team innovating on it, so conflicts are impossible by construction. Marvis is a large-scale experience model that extends from campus and branch to the data center, the WAN and security. It is an elegant answer and a genuine differentiator compared with the multi-agent orchestration challenges that both Cisco (AI Canvas, Agent Studio) and Extreme (Agent ONE, Exchange) are taking on. The trade-off is that HPE’s approach is more closed: where Extreme exposes MCP (Model Context Protocol) servers through its Exchange marketplace and invites partners to build and monetize skills, and Cisco offers Agent Studio for third-party agents, HPE’s agent extensibility story lives further up the stack, in Morpheus and GreenLake Intelligence, not in the network layer itself.

I also asked whether HPE envisions vertical-specific agents for healthcare, education, and similar markets. Rahim’s view is that infrastructure problems are similar across verticals, varying in scale and complexity rather than by industry, so a single engine that learns from every vertical is the better architecture for now. Contrast that with Extreme, which launched Exchange with explicit healthcare, education, retail, and manufacturing skills at day one. Two defensible philosophies; the market will referee.

Exhibit: HPE Technology Stack

Source: Maravedis, based on HPE pre-Discover analyst briefing, June 2026

The exhibit maps the portfolio the way HPE presented it, and the two amber bars are the point. In the networking layer, Mist and Aruba Central remain separate platforms by design, but both run on a single Marvis AI engine, with Data Center Director extending into the data center fabric and the SASE Orchestrator extending into security. That one-engine, one-team architecture is why Rahim could tell me that conflicting agent actions are impossible by construction.

GreenLake Intelligence serves as the mirror role at the bottom of the stack, linking the Morpheus orchestration copilot and the OpsRamp operations copilot through an agent mesh that now extends into ServiceNow for ITSM (IT service management). Zerto sits in the operations row but operates across layers, protecting Morpheus virtual machines and rolling back AI factory agents to a known-good state. Read vertically, the chart is also HPE’s differentiation claim in a single image: Extreme sells the top band, Cisco sells the top band plus observability, and only HPE sells every layer of this picture under one roof.

All these pieces of the technology stack can get quite confusing, I must confess. HPE now carries Mist, Aruba Central, Marvis, Morpheus, OpsRamp, Zerto, Data Fabric, Alletra, and GreenLake Intelligence, several of them arriving through acquisitions (Juniper, Morpheus Data, OpsRamp, Zerto) and several renamed at least once, with Apstra becoming HPE Networking Data Center Director as the latest example. If you lose track of which brand is the network brain, which is the operations brain, and which is the thing being managed, you are in good company, and that confusion is itself a real adoption tax on buyers. The exhibit above is my attempt to make it stick: two AI engines, the products they drive, one GreenLake roof. When the names blur, come back to that picture.

Agentic infrastructure: where HPE plays a card the other two do not hold

Russo’s portion covers ground that neither Cisco Live nor Extreme Connect could, because neither company sells the full compute and storage stack. The AI Factory portfolio now comes in two flavors: NVIDIA reference architecture designs, which include confidential computing and country-specific compliance (GDPR for EU data privacy, plus the US government’s FIPS and STIG security standards), and a new HPE-validated design that integrates Juniper QFX switching and Apstra-based fabric management. Private Cloud AI, the turnkey inferencing appliance, scales to 256 GPUs (graphics processing units), is Vera-ready for NVIDIA’s next platform, and gains KV cache (key-value cache) acceleration, built-in data intelligence, and metadata enrichment via the Alletra MP (multi-protocol) X10000 array, which now runs object and file in parallel with no performance trade-off.

Two items deserve particular attention. Data Fabric 8.2 adds integration points for agentic workloads, enabling agents to see and access enterprise data across clouds and third-party storage with proper identity and access management, plus a pre-integrated appliance option for faster deployment. And Zerto, the protection software HPE acquired years ago, is being repositioned for a genuinely new risk category: rogue agents. If an agent fails, Zerto can roll the environment back to a known-good agent state. Nobody else on the spring circuit framed agent protection as a disaster recovery problem, and it is a clever, credible angle. In the Q&A (question-and-answer session), Rob Strechay confirmed that Zerto’s agent protection is Morpheus-only today and does not yet cover Red Hat or native Kubernetes on Private Cloud AI, a scope limitation worth tracking.

Three vendors, one destination, three vehicles

Stack the spring’s three events side by side, and the convergence is unmistakable. All three companies are selling agentic AI operations with human-in-the-loop today and autonomy tomorrow. All three insist this is not a chatbot bolted onto a dashboard. All three lean on domain models, live network context, and encoded expertise.

 

Extreme

Cisco

HPE

Platform answer

One platform (Platform ONE, GA July 2025)

One platform (Cloud Control, CA June 2026)

Two platforms, one AI engine (Marvis)

Agent brand

Agent ONE (Coworker, Operator)

AI Canvas, agents in Cloud Control

Marvis Actions, GreenLake copilots

Extensibility

Exchange marketplace, MCP, partner monetization

Agent Studio, App Builder

Morpheus MCP server, plug-in architecture

Beyond networking

No

Splunk, security, collaboration

Compute, storage, virtualization, AI factories

Agent risk answer

Governance boundaries

Agent identity, just-in-time access

Zerto rollback to known agent state

Unique flank

Fabric (SPB, Shortest Path Bridging) lock-in, channel skills

Live Protect, post-quantum

VMware migration economics

Source: Maravedis Research

The differences matter more than the similarities. Extreme is the channel-native pure play, the only one with an explicit partner monetization path for AI skills, but it is fighting an adoption battle within its own base. Cisco brings the largest estate and the deepest security and observability stack, but its story is squarely an enterprise command-center story, and the SMB (small and medium-sized business) channel questions from our research remain open. HPE is the only one selling the entire stack underneath the agents, from QFX switching to GPUs to storage to the hypervisor, and the only one weaponizing migration economics against an incumbent (Broadcom) rather than against each other.

The Maravedis read: what the managed connectivity channel should watch

Three things, in order of consequence for our readers.

First, Marvis Actions on Aruba Central is the announcement to actually test. Central is the operating system of a large slice of the managed Wi-Fi, hospitality, and MDU world. If proactive remediation lands there in a form MSPs (managed service providers) can operationalize, the truck-roll economics shift in a way that maps directly onto managed service margins. The question, as always, is packaging: which Central tiers get Actions, and at what licensing delta. HPE deferred specifics on availability to the press releases, so read the fine print on June 16.

Second, the never-converging platforms position cuts both ways for the channel. It spares Central customers a forced migration, which is exactly the anxiety Extreme is wrestling with on its Platform ONE conversion and the change-management burden the Cisco convergence implies. But it also means MSPs straddling both stacks will run two consoles indefinitely, with parity arriving over years. One brain, two bodies is a sound architecture; it is also a long-term operational commitment to dual tooling.

Third, the Morpheus migration program is a template the networking side should copy. First-year-free licensing, free migration tooling, and zero-interest financing directly address the licensing treadmill complaints that dominated our SMB study, Small Business Wi-Fi: The MSP Networking & Security Requirements, in which support quality and licensing simplicity outranked feature depth for 100 percent of surveyed MSPs. HPE has shown it knows how to remove commercial friction when it wants a market badly enough. The managed connectivity channel should ask why that creativity stops at the hypervisor.

Bottom line

Expect HPE Discover 2026 to be the most architecturally complete of the spring’s three agentic shows: networking, compute, storage, data, virtualization, and operations under one GreenLake roof, with one Marvis engine threading the network layers together and Zerto guarding against agents gone wrong. Coming after Extreme Connect and Cisco Live, it confirms that the industry has fully settled on the agentic playbook and is now differentiating on stack breadth, platform philosophy, and commercial aggression. Extreme bets on channel monetization, Cisco on enterprise gravity, HPE on owning every layer and underpricing the incumbent next door. For the managed connectivity channel, the headline is not “AI factory racks”. It is Marvis Actions arriving on the platform most of this market already runs, and a migration playbook that proves vendors can make economics, not just architecture, the reason to move.

Back to blog

Leave a comment

Please note, comments need to be approved before they are published.