AI Will Decimate This High-Margin Networking Business
We are already seeing companies will slow-burn AI deployment as a result
This week, at HPE Discover 2025, I sat through Antonio Neri’s keynote, several networking presentations, and meetings with executives as part of HPE’s analyst track. Something that kept coming up, and I kept asking about, permeated the discussions. Why not push further in automation? Finally, I asked the blunt question and received the answer that I had guessed lay beneath the surface. AI is coming for a high-margin networking business, but not one many are focusing on.
Setting the Stage for Decimation at the Hands of AI
At HPE Discover 2025, several agentic AI workflows were demonstrated on administrative panels. Those included more than just networking. Instead, HPE also showed those workflows in the context of hybrid cloud automation, as part of GreenLake, and more.
All of these demos were touted in a similar fashion:
AI Agents are deployed to monitor infrastructure
Instead of alerts being routed to administrators, the AI Agents would take a first pass
Reasoning models and RAG are then used to identify issues, hypothesize root causes, and suggest remediation steps.
Remediation recommendations are then sent to an administrator
The administrator can then go through the chain of thought and decide whether to implement the recommendation of the AI Agent
Almost every time someone at Discover 2025 walked through that process to varying degrees of detail, there was a quick vignette added around “this is all happening while your IT staff is sleeping, eating, or on break.” The subtle nod was that work was getting done while the administrator was not working. Still, the urgency of the issue being described was such that whenever the administrator returned from lunch, a break, or woke up, it would be a great time to fix the problem.
Therein, however, lies the problem with this example.
Using AI Agents to administer networks makes a lot of sense. Often, server, storage, and networking issues are presented in logs. Parsing through large amounts of data and generating alerts is a well-implemented industry practice. The challenge is what happens to that data now.
In some cases, automated systems can repair systems. A network port can fail, and traffic can still flow through alternate network routings. For many years, SMART errors on storage disks in the data center have been able to trigger automated service calls and dispatch replacement parts. Those are just some examples.
In other cases, however, these automated systems are unable to remediate. That is where AI Agents are set to come in. At a high level, the goal of Cisco, HPE Aruba/ GreenLake, and others in the industry is to move AI Agents that can reason through problems, conduct research, decide on a path forward, and implement that fix.
We are focused on remediation, but it goes far beyond fixing problems. It can be placing workloads, provisioning, and more. Instead of just being used for break/ fix processes, many in the industry are also focused on building and provisioning infrastructure with the new tools. Baseboard management controllers, or BMCs, in devices provide remote access to servers, storage arrays, switches, and more. Over the years, they have added methods to enhance remote observability through logs and APIs. In many ways, the DPU has become the device to provide secure access to networked environments. In a world of zero-trust networking, many of the roles of traditional firewalls go away, and the DPU becomes a powerful tool for an AI Agent administering security and connectivity in a data center.
What should be scary to many in the industry is what happens when you ask the next question, as I did repeatedly this week. “Why does a human need to be in the loop?” The demos HPE showed featured an AI Agent going through the process of root cause analysis, determining the best solution, and creating implementation plans. The AI Agents this week had one thing in common: they waited for IT administrators to click “go.”
Think about that for a moment. What is being offered by the industry is to train an AI Agent. Then give that AI Agent access to loads of information and access to do research. That AI Agent is then released into the operating environment, but it cannot take action, only recommend. This is equivalent to my high-school calculus teacher needing to be consulted every time I calculated a tip on a bill before I could write an amount or hit a button. That makes absolutely no sense.
The underlying goal is really to have these AI Agents be autonomous, just as we have learned skills in school, and then we are sent into the world to utilize them.
If you believe in that statement, then the five-step process being offered by HPE this week and others in the industry makes absolutely no sense other than as a stepping stone for the future. Taking the human out of the loop should have several important impacts:
Speed, since the human that can sleep, eat, or take breaks is no longer gating progress
Accuracy, many in the networking and other IT management areas will tell you that the #1 cause of downtime is human error. Ideally, AI Agents should get better and more error-free. If you look at self-driving cars and the safety data around Waymo, Tesla, and many others, it is clear that removing human error can lead to better results. Machine errors can also be reduced quickly across fleets of AI Agents so mistakes are not repeated.
Cost, as we need fewer humans in the loop to maintain systems, and we lower downtime costs through faster remediation
Those make a lot of sense to folks. Business and financial analysts model the costs associated with not just developing AI Agents, but also deploying and operating them versus the costs of current service and support contracts sold by vendors and IT headcount. That work has largely happened. Some offer that jobs lost to AI Agents will be replaced by jobs managing AI Agents, but if that is the outcome on a 1:1 ratio, then the entire exercise is a failure.
Still, for networking vendors and many third parties associated with them, there is another key industry, with annual estimates in the tens of billions, that will be decimated.