6 networking resolutions for 2026

Focus on networking for AI inference

AI will continue its rapid evolution in 2026. A year ago, it was hard to tell how much effort and energy might be spent on building specialized networks to support GPU training clusters versus the widespread use of AI across the enterprise. 

The past year has provided some clarity, however, with the bulk of training happening in the cloud and the plan for distributed inference now taking center stage. That means preparing networks for increased data collection to feed the AI engines and decreased latency to tighten analysis and action loops. For many, this will be a key reason to take refreshes and upgrades seriously this year, from the data center to the campus and branch.

Wi-Fi 7 and private 5G are poised to optimize the access layer, and core networking capacities continue to improve, augmented by a growing number of network devices that embed data processing units to facilitate distributed inference. 

Proactive networking might be real

Networking teams have long desired proactive networking, but it has been impractical for many. Machine learning operations and AIOps promised to fix this, but NetOps teams are still stuck in break-and-fix reactive modes, largely because of the unique snowflake nature of each enterprise network. 

That might change with the arrival and early successes of agentic AI, which can spend time recognizing anomalies and tracking them to determine if they are worthy of action. Even better, AI agents can look up and recommend potential corrective actions because AI can navigate knowledge bases.

In 2026, network professionals should watch what vendors plan to do with generative and agentic AI. The potential is there, and 2026 might be the time to see if these technologies can make a difference in the network.

Gauge progress on network automation

The quest to automate networks has been picking up steam over the past few years, driven by infrastructure as code and DevOps initiatives that are transforming the application development landscape. Efforts of organizations like the Network Automation Forum are helping, bringing together like-minded networking professionals to share best practices and build consensus on how to make progress.

Cloud networking is already largely automated, as are most virtual networks for supporting virtual compute and container environments. But traditional campus, branch and data center networks have been less compliant, in part due to the lack of standard network OSes and well-formed APIs.

Two evolutionary changes offer hope of overcoming the barriers:

1. Agentic AI's Model Context Protocol as a facilitated alternative to APIs.
2. The nascent migration of SONiC out of the cloud-scale operator environments and into the enterprise.

Networking professionals without that understanding or who consider the programming skills gap too big to cross should check out the many well-established, proven network automation platforms that commercial vendors have been steadily modernizing to fit into today's dynamic environments.

Stick your nose into Kubernetes

The latest iteration of virtual networking is growing fast within container orchestration environments, namely Kubernetes. This is not a new topic, however. Kubernetes-based environments are now being deployed in production and at scale.

That progression leads to the typical challenges of scaling networks, including load balancing, DNS, IP address management, security and more. The cloud-native community has made progress on addressing these challenges, but we are now seeing mainstream networking and infrastructure vendors taking notice and entering the mix.

For example, we saw the likes of Cloudflare, Fastly and HAProxy showing up at KubeCon, and Nutanix and Avi Networks -- part of VMware, owned by Broadcom -- announcing specific Kubernetes adaptations. It might be worth exploring Kubernetes and lending a hand with adaptations, as network teams might be expected to support it in the production environment.

Take another look at enterprise 5G

Much of the historical market focus around 5G has been on the consumer side. However, the advent of complete 5G service architectures, known as standalone 5G, brings more advanced capabilities that make this a viable alternative to enterprise access networking.

For example, the private networking options now available that use network slicing along with significant improvements in bandwidth and performance, as well as a growing array of embedded security features, should make 5G SA interesting.

Many mobile network operators are making steady progress in this area. Mobile network coverage and signal strength can be a limiting factor, so this makes the most sense where 5G coverage is well established, although integrated non-terrestrial (satellite-based) options are now joining the mix to extend 5G connectivity to just about every corner of the globe.

Get serious about NetSecOps

There have always been close parallels between the operational monitoring and management technologies used by NetOps and those used on the network side of SecOps. 

The first time I studied this topic, almost 15 years ago, I found high degrees of ad hoc collaboration between NetOps and SecOps were the rule rather than the exception, even back then. Such collaboration has deepened and looks to continue doing so going forward. 

A September 2025 Omdia study on networking and security convergence by myself and a colleague revealed that a third of organizations have already established a single unified team for procuring, deploying and managing networking and security tools and technologies. Another 40% were in the process of unification. In the same study, 95% of participants said converging network and security technologies and processes was a top IT priority.

Both sides of the house agree that striking an important balance between network security and network performance is necessary, and the best way to get there is to sit at the same table. If you aren't looking at this already, it's time. In fact, it's past time.

Jim Frey covers networking as principal analyst at Omdia.

Omdia is a division of Informa TechTarget. Its analysts have business relationships with technology vendors.