TECHnalysis Research Blog

June 10, 2025
Cisco Highlights Promise and Potential of On-Prem Agents and AI

By Bob O'Donnell

At a big vendor event like Cisco Live, it’s always challenging to find themes that run through the enormous range of announcements that typically get made at these kinds of shows. The volume of news can quickly get overwhelming, making it difficult to summarize what the top few topics from the show were.

At this year’s show, however, one thing that became apparent is that the opportunity for running important applications—like GenAI and autonomous agents—is no longer limited to the cloud—and Cisco is here for it. To be sure, Cisco pointed out a number of important new products and services that are specifically targeted at major cloud computing providers and large ISPs. However, they also noted that interest in expanding the capabilities of on-premise datacenters is not only still alive, it’s actually getting reinvigorated by the rapid transition to AI-powered workloads.

From new AI-optimized, on-premise-focused routers and switches delivered via the company’s traditional networking hardware business, through AI-enhanced versions of their on-prem communications, collaboration and customer support software platforms, and even into their new agentic AI offerings, Cisco made it clear at this year’s Live event that it was eager and willing to meet the growing demand for capabilities that businesses can run within their own environments.

In some ways, it’s a stunning development, given the relentless focus on moving virtually everything to the cloud that the IT industry has witnessed for the last 10-15 years. And a cynic might argue it’s an attempt from an old-school tech company to return to its glory days of providing essential equipment to the enterprise. But in truth, it’s part of a much broader industry movement that many larger companies are starting to see and talk about.

Driven in part by traditional data gravity arguments that say you need to bring the workloads to the data, the fact that the most precious corporate data—and the most useful for AI model fine tuning—still often sits behind the firewall is an important factor. Combine that with the increased range of hardware and software offerings capable of running these advanced workloads that are now available for the enterprise, and it all starts to make sense. While companies will unquestionably still use the computing resources in the cloud for some of their workloads, the interest in and ability to do more internally is a real thing. To put it succinctly, the world of Hybrid AI, where companies use both public and private clouds/datacenters for AI workloads, is upon us.

Looking through that Hybrid AI lens, many of the specific announcements from Cisco Live have important relevance. In the case of hardware, for example, new additions to its C9000 series switches and 8000 series routers leverage the latest Cisco Silicon One chips to run latency-sensitive AI workloads while still integrating support for post-quantum security and zero trust networking amongst other capabilities. The 8000 series routers also integrate an enhanced version of Cisco’s SASE (Secure Access Service Edge), SD-WAN (Software Defined Wide Area Network) and next-generation firewall (NFW) functionality.

For computing, Cisco also launched their version of Nvidia’s Blackwell RTX Pro 6000 GPU-based server (or, as Cisco calls them AI Pods), which is specifically designed for companies to run within their own datacenters. In addition to this, Cisco built on its partnership with Nvidia to show off interoperable networking hardware that offers Nvidia’s Spectrum X capabilities but use Cisco’s Silicon One chips and connects with Cisco G200 switches. Bottom line is it makes it much easier for companies with existing Cisco-based networks to connect with these new AI computing resources.

On the software platform side, Cisco announced that many of the previously cloud-only AI-powered features in Webex, such as blurred backgrounds, audio denoise, meeting summarization, etc., are now coming to companies who still want to run Webex on-prem. By itself it isn’t a huge announcement, but it is indicative of the priorities Cisco clearly has and the direction they’re taking. Similarly, for the call center-based tools, Cisco also debuted the Webex AI agent, which allows pre-built templates for customer service interactions to be created much more easily. The new Cisco AI Assistant can also provide call center employees with real-time transcriptions that can compensate for fast talkers, those with accents and, most importantly, across different languages. In addition, the Assistant can provide suggested responses, which can be tremendously helpful for customer service agents with limited experience.

As expected, Cisco also announced several important AI and agentic-based capabilities at Live, including its own customized, network-focused LLM they call Deep Network Model. It’s trained on Cisco’s own documentation as well as several decades of real-world data on how to interpret and resolve network-based issues. Cisco is using this new model to power several important new capabilities, most notably what it calls AgenticOps and the AI Canvas.

AgenticOps is Cisco’s new tool for helping manage what is expected to be a deluge of new AI-powered agents traversing over its networks. The idea is to provide something that can automate and operate agents at machine-level speeds and provide real-time telemetry, but still let human-based IT teams remain in control. The AI Canvas takes the data from the AgenticOps tool and lets organizations create dashboards and other visual indicators of key network, application and agent activities.

The Deep Learning Model is also being used to power the previously mentioned network-focused chatbot called Cisco AI Assistant. It offers a great way for IT employees and network administrators to get quick, accurate answers to common issues that they may face in their day-to-day jobs. Importantly, all of these agentic capabilities are being made available for both cloud-based and on-premises-based implementations, making the two equivalent on a feature and functionality basis.

Another key area that Cisco discussed at length at Live was creating even more secure networks—again on-prem. Noting that the powerful capabilities of AI-powered tools also give bad actors more opportunities to create dangerous cyberattacks, the company discussed important new additions to their zero trust and firewall offerings to address these issues. Specifically, Cisco talked about a new Hybrid Mesh Firewall as well as a Zero Trust Network Access (ZTNA) strategy that simultaneously works to bring security protections further into the network and across a more distributed set of hardware (including from some competitors) while also simplifying the process of configuring and running these tools. One particularly relevant feature is that the ZTNA capabilities are being extended to digital agents to help ensure that they are being constantly monitored.

Of course, there were a huge range of other announcements that Cisco also made, particularly around further extensions to and connections between their Thousand Eyes monitoring and Splunk automation tools. The new single Ethernet-cable connected PTZ (Pan Tilt Zoom) camera is also a great extension of the company’s collaboration offerings.

Taken together, all the news from Cisco Live points to a vision of a company focused on not only bringing advanced AI capabilities to cloud-based offerings, but to on-premises installations as well. They clearly see a shift in the focus of many businesses around the world and they’re eager to provide solutions that help fit those evolving needs.

Bob O’Donnell is the president and chief analyst of TECHnalysis Research, LLC a market research firm that provides strategic consulting and market research services to the technology industry and professional financial community. You can follow him on LinkedIn at Bob O’Donnell or on Twitter @bobodtech.