In my previous HPE Discover recap, I spent much of my time talking to HPE’s moves in management and automation. Compute Ops Management (COM), Morpheus, OpsRamp, GreenLake Intelligence. This is partly because server and infrastructure management are near and dear to my heart. And partly because it is an area of growing importance as increasingly complex AI infrastructure is deployed across the enterprise.

This week is the networking story. And, frankly, it’s the biggest story coming out of Discover. For years, HPE was a server and storage company with a networking line attached. After this Discover, a strong case can be made that the framing is inverted. HPE now presents as a networking company that also sells compute and storage. Antonio Neri said as much from the stage, and he named Cisco as the competitor he measures the company against. A few years ago, that would have sounded like bravado. This year, it sounded like a plan.

But networking is the lens here, not the whole picture. The deeper move is HPE shifting the competition from any single infrastructure domain to a layer that runs the network, servers, storage, and AI agents as one system. Networking is the catalyst and the loudest proof. But the operating layer underneath is the actual bet, and it’s a thread worth following through.

Here is what was actually announced, why the Mist and Compute Ops Management tie-in matters, what GreenLake Intelligence has become, where Juniper and Aruba really stand, and how all of it reads competitively.

The Self-Driving Network, From Slideware to Shipping

One of the major themes of Discover was the self-driving network. It describes the network as self-healing, self-protecting, and self-optimizing. True autonomy through an operations layer that finds problems, root causes,  explains the cause, and fixes it without a human in the loop. We’ve heard this kind of story and vision for some time. Frankly, we’ve been talking about it for decades. What changed at Discover is that HPE put real product behind it.

At the center of all this sits Marvis, the AI operations engine that came over with Juniper and Mist. Marvis now runs inside Aruba Central, not just Mist. Its automated remediation, the part that takes corrective action rather than simply identifying a problem, now extends across wired, wireless, and SD-WAN through Aruba Central.

That emphasis on closed-loop automation is one of the more interesting competitive developments. Cisco has invested heavily in AI-assisted operations, with tools that help administrators identify issues, understand root cause, and accelerate remediation. HPE is signaling a willingness to push further toward autonomous execution. Whether customers are comfortable giving AI that level of operational control, and whether it performs reliably enough to earn that trust, will ultimately determine how meaningful that distinction becomes.

I think any vendor pursuing autonomous operations has to move at the pace of its customers because every enterprise has a different appetite for automation. The right approach is giving IT organizations the ability to decide where AI recommends, where it assists, and where it acts autonomously.

The integration works both ways. Aruba CX campus switches can now be managed from Mist, giving long-time Aruba customers access to Mist’s operational model while allowing existing Mist customers to deploy Aruba switching without introducing a separate management platform. That’s more meaningful than simply combining two product portfolios. If HPE executes well, it reduces one of the biggest risks that follows any large acquisition: forcing customers to choose between two management models. Instead, HPE is beginning to make the combined portfolio feel like a single networking platform rather than two products sharing the same logo.

Networks for AI, Not Just AI for Networks

HPE split its story into two halves. AI for networks uses models to run the network. Networks for AI is building the network that AI infrastructure needs. And this second half is where the Juniper data center gear shows up.

Two new switches anchor this push. The QFX5140 targets inference clusters and edge AI, where I believe much of the enterprise AI opportunity is headed. But the QFX5252 is perhaps the more interesting announcement. It’s a switch tray designed for AMD’s Helios rack architecture that supports UALink over Ethernet, an open accelerator interconnect backed by a broad industry consortium as an alternative to NVIDIA’s NVLink.

More than anything, the announcement reflects where parts of the industry are headed. AMD is betting that open accelerator interconnects gain meaningful adoption as AI infrastructure matures. By supporting Helios and UALink, HPE is ensuring that customers who choose that architecture have an integrated networking option. Whether open accelerator fabrics become commonplace or proprietary approaches continue to dominate remains an open question, but HPE is positioning itself to support either outcome.

The rest fills in the picture. The QFX portfolio becomes part of HPE’s AI datacenter reference architecture under the renamed Apstra, now Networking Data Center Director. At the same time, Juniper’s MX and PTX routing platforms support the multi-site connectivity described in NVIDIA’s AI factory blueprint.

So HPE is no longer just selling the servers that go into an AI factory. It’s selling the fabric too. That puts it directly against Cisco and Arista in the datacenter spine, a fight HPE wasn’t really in before Juniper.

Why the Mist and COM Tie-in Matters

This announcement connects to my previous field note, and it is the one enterprise should watch. Mist Data Center Assurance now integrates with COM and into GreenLake more broadly. In plain terms, the network operations view and the server operations view now sit in the same place. One console. One integrated view.

This may not sound like a big deal, but it is. In most datacenters the network and compute teams work in silos with separate tools and telemetry. When something slows down, the first hour goes to arguing about whether it’s the network or the server (as a server guy, I naturally blame the network). Putting network assurance and compute management under one roof lets a team see across both and lets the system correlate a network event with a server event without a person carrying the context between two screens.

For enterprises, I see this doing three things. First, it cuts tool sprawl, which is a real line item and a real source of operational drag. Second, it enables a smaller team to run a larger estate. And it’s the first concrete step toward what HPE calls the self-driving datacenter.

GreenLake Intelligence: the Management Story I Did Not Expect to Lead With

I think what impressed me most was GreenLake Intelligence. HPE teased this last year, when it mainly framed around full-stack observability and AI-assisted operations. Think of that as a foundation.

This year, HPE built on it and gave it shape. It is now the operational layer that the rest of the portfolio reports into, and it is the most convincing version of the control-plane story HPE has told.

Enterprise IT is about to inherit a management problem it hasn’t had to solve before. AI agents won’t just consume infrastructure. They’ll request resources, access data, call other agents, and increasingly make decisions on behalf of users and applications. And as organizations move from a handful of agents to hundreds or even tens of thousands, they need to know what’s running, who owns it, what it’s allowed to do, and what it’s costing. I don’t think today’s tools were built for that.

And this is the backdrop for GreenLake Intelligence. It’s not another operations dashboard. HPE’s trying to build an operating framework for AI-driven infrastructure. At its core are three pieces: an agent registry, orchestration, and governance.

To me, the registry is the most interesting piece because it’s trying to become the system of record for enterprise AI agents. And every organization that’s serious about agentic AI will need one. Does GreenLake become that system of record? Time will tell – but HPE is definitely signaling where the industry is heading.

The bigger idea is to pull networking, compute, storage, security, and AI operations into a common operational model rather than managing each as a separate discipline. That’s where GreenLake Intelligence becomes most interesting. If HPE can make that work across its own portfolio and the mix-and-match infrastructure enterprises run, it’ll have something much more meaningful than another management console. That’s also the hard part. Building the framework is one thing. Earning enough trust for enterprises to let it coordinate increasingly autonomous infrastructure is something else entirely.

And here’s why I keep coming back to this. The networking news is louder and incredibly strategic. But this is the part that is hard to copy. A networking-only vendor can match a switch or an AIOps engine. Far fewer can sit above the network, servers, storage, and agents and run them as one big thing. That is the actual bet HPE seems to be making, and I think it’s the right one. The model is not where enterprise AI is won. The operating layer is.

The honest caution is that much of this rolls out across 2026 and 2027 (and likely 2028), so today it is more architecture than finished product. But the direction is correct, and the progress from last year to this year is real, not a rename.

Where Juniper and Aruba Actually Stand

One decision stood out to me. HPE isn’t trying to merge Mist and Aruba Central into a single management platform, at least not today. Instead, it’s keeping both and integrating the underlying AI capabilities. Features developed for one platform can be brought to the other without forcing customers to switch to a new console.

I actually think that’s the right near-term decision. Bigger networking customers aren’t going to replace management platforms overnight just because two companies merge. Preserving both consoles reduces disruption for existing Aruba and Juniper customers while giving HPE time to integrate the technology underneath. That’s more practical than chasing a rushed “single pane of glass” story.

But there is a tradeoff. Organizations running both Aruba and Juniper still have two management planes. HPE will say a common AI foundation prevents those platforms from working against each other. But operational consistency isn’t the same thing as operational simplicity. At some point, customers will want a clearer answer on which management experience will become the long-term destination. I don’t think HPE has to answer that question today, but it will over the next year or two.

How This Positions HPE Competitively

Discover kind of felt like HPE leaning more into the Cisco competitive dynamic more than Dell. But this could just be my view. HPE claims it now holds one of the two most complete networking portfolios in the market, the other being Cisco. And it pointed to an eleven-billion-dollar networking business as the goal.

The contrast with Cisco is sharper than the overlap. Cisco is merging Catalyst and Meraki into a single management plane. HPE is doing the opposite (for now). Both are betting that AI-run operations are the future of networking, so the destination is the same.

In the AI data center, Juniper’s switching portfolio gives HPE a credible position in the market for AI Ethernet fabrics alongside Cisco and Arista. On security, the new SASE and firewall offerings broaden HPE’s competitive reach against Cisco and Palo Alto while reinforcing the company’s strategy of tying networking, security, and operations more closely together.

The real separation for HPE  is not going to be a switch or a console. It’s the stack: Cisco, Arista, and Extreme sell networking (Cisco also has compute). HPE sells networking, compute, storage plus a hybrid cloud operating model. The Mist and COM link, and GreenLake Intelligence above it, are the first visible payoff of that breadth.

If HPE can make one operating model genuinely run the network, the servers, and the cloud together, that is ground a networking-only vendor cannot easily take. Dell is the mirror image and the reason the stack argument holds. It has the compute and the partnerships, but not a networking layer of this depth. And I think this is exactly the gap HPE will want to exploit.

The Next Twelve Months

HPE walked into Discover with the most coherent networking and operations strategy I’ve seen from the company. More importantly, it recognized where the market is heading. Most organizations can stand up an AI pilot. Running AI reliably in production is a very different challenge, and it’s increasingly an operational rather than an infrastructure problem.

Now comes the hard part. Strategy is easier than execution. The measure of success won’t be how many AI features HPE announced or how many products it integrated. It’ll be whether enterprise IT teams spend less time managing infrastructure and more time delivering outcomes.

I think HPE is moving in the right direction. But over the next year, customers should watch for evidence that the operational model is becoming simpler, not just more capable. When networking, compute, storage, and security begin to act as parts of a coordinated system rather than as adjacent products, HPE will have achieved success.

Until then, the strategy deserves attention. But execution is what will determine whether it changes the market.