- Cisco has a bold plan to enable network operators to keep up with the exploding demands of AI applications and data traffic
- The company unveiled a new AI-ready network architecture and AgenticOps platform to manage surging demand from agentic AI and improve infrastructure visibility and automation
- The architecture merges Meraki and Catalyst platforms, introduces ruggedized switches for industrial use, and emphasizes post-quantum security and real-time observability
CISCO LIVE, SAN DIEGO—"Paradigm shift" is about the most ancient cliche in the technology industry. The phrase originated in 1962, coined by the American physicist and philosopher Thomas Kuhn, to describe scientific revolutions in which the underlying assumptions of normal science become incompatible with new phenomena. Examples include Darwinian evolution, general relativity and the Big Bang theory of the universe's origin.
DJ Sampath, Cisco SVP AI software and platform, invoked the ancient cliche when describing changing network requirements brought about by AI. "We're talking about a paradigm shift," Sampath said in a briefing last week for technology journalists and analysts. "It's a new way of thinking about reimagining operations."
But cliches get to be cliches because they are apt descriptions of circumstances.
And Sampath is right.
Agentic AI will explode demand for data and connectivity, Jeetu Patel, Cisco president and chief product officer, said at the same briefing. Where we now have billions of people using the internet, we will soon have tens of billions of agents straining network infrastructure, safety and security. "The classic ways that infrastructure was handled won't be able to deal with the scale and proportion we're talking about," Patel said.
That plays into Cisco's strength, with a broad portfolio and long history of building global networks.
"Think of us like the picks and shovels company during the gold rush," Patel said.
What's new
Today, Cisco unveiled a network architecture to rebuild and secure the world's networking infrastructure, unlocking the revolutionary potential of agentic artificial intelligence. The architecture meets the new demands of explosive network traffic, mission-critical uptime requirements and intensifying security threats, managing the AI-ready network securely and with simplicity, Cisco said.
Networking constrains AI potential, Patel said. GPUs are expensive, and must be used to maximum capacity. "The idle time of a GPU during a training run is like throwing away money. It's like burning cash," he said. Network operators need to ensure that packets are getting to the GPUs on time.
The need to eliminate the networking bottleneck is the subject of a free Fierce Network Research report: "AI and the Network: Optimizing Network Design and Operations to Meet AI Demands."
Safety and security are also constraints on AI. "If people don't trust the system, they're not going to use it," Patel said.
These concerns have CEOs feeling a little lost. According to a recent Cisco study, 97% of CEOs are enthusiastic about AI but only 1.7% feel prepared. They're concerned about infrastructure, safety, security and the skills gap, Patel said.
About that paradigm shift
Human operators can't keep up with the exploding demands of AI networks. But AI can't be trusted to make decisions on its own. To resolve that contradiction, Cisco introduced AgenticOps, an AI-driven approach to managing IT operations with human oversight.
AgenticOps is built on an advanced networking large language model (LLM) called the Deep Network Model. "It's trained and fine-tuned on over 40 years of expertise, on data that has been vetted for accuracy, which means the model is going to be very precise," Sampath said. "This model is continuously going to learn based on a ton of telemetry that we constantly provide. This model is constantly learning and getting better."
AgenticOps incorporates the Cisco AI Assistant, a natural language interface that identifies issues, diagnoses root causes and automates workflows.
These AgenticOps capabilities come together in AI Canvas, a generative AI interface for managing infrastructure. AI Canvas is cross-domain, unlocking data silos and connecting network silos together, with collaborative human oversight, Sampath said.
Sampath demonstrated AI Canvas remediating a simulated networking glitch. He showed a ServiceNow trouble ticket, based on a ThousandEyes alert for network timeouts and transaction failures. "What typically happens now is that you've got to go to different dashboards and swivel your chairs between different consoles to figure things out," Sampath said. Instead, network operators can manage everything from a single screen using AI Canvas.
The human operator asks an AI assistant a simple question in natural language. The AI pings Cisco Meraki for device performance metrics, generates a visualization for the entire network, troubleshoots the problem and recommends a solution. The human operator reviews the recommendation and drags and drops the solution in place. Multiple operators can work together and — importantly — revert changes if they prove disadvantageous.
"Having the human in the loop is important. If something did not work as expected, you have the opportunity to change that back," Sampath said. (We learned about the importance of having humans in the loop in 1983.)
Patel said AI Canvas is multi-user and collaborative. "The task of infrastructure management is a multiplayer sport. Multiple people get involved in troubleshooting and policy setting." Other AIops tools are single-user, not multi-user and collaborative, he said.
The user interface is text-first, readable by both humans and AI agents. The agents themselves can become collaborators in network operations, Patel said.
Picks and shovels
To meet the demands of more complex networks, Cisco is merging management of its Meraki and Catalyst devices into a single platform, combining next-generation wireless, switching, routing and industrial networks on a single platform supporting cloud, on-premises or hybrid deployment. Cisco extended ThousandEyes assurance to include mobile endpoints and industrial IoT, and integrated Splunk for real-time insights across the network and application layers.
The new, expanded Meraki provides a unified toolset for managing IT infrastructure, including offices, retail stores, and factory floors, Anurag Dhingra, Cisco SVP and general manager, enterprise connectivity and collaboration, said.
Manufacturing is pushing the limits of networking requirements. AI-enabled robotics requires high throughput, low latency and rock-solid reliability to connect robots moving around a factory floor.
To serve the new generation of manufacturing, Cisco introduced 19 new Ethernet switching products, ruggedized for industrial applications.
With robotics in mind, Cisco is converging standard Wi-Fi with Ultra Reliable Wireless Backhaul. Robots and other mobile endpoints moving across access points can automatically connect to new networks even before the existing network connection is broken.
Cisco is building security into the network from the ASIC to the application, starting with network devices built on the Silicon One architecture, to provide trustworthy, tamper-proof systems that are post-quantum-cryptographically enabled with secure boot systems, the company said. Devices can mitigate against exploits without the requirement to re-image or reboot them, maintaining security without reducing reliability, Dhingra said.
Here comes the specs
New Cisco C9350 and 9610 Smart Switches on Silicon One deliver up to 51.2Tbps throughput, below 5 microsecond latency and quantum-resistant secure networking. Cisco 8100, 8200, 8300, 8400 and 8500 Secure Routers provide native SD-WAN and Secure Access Service Edge (SASE) integration, next-generation firewall and post-quantum security in a single-box WAN, with up to three times the throughput of previous generations. Cisco is extending its Cisco Wireless 9179F Series Access Points for Wi-Fi 7 access for stadiums and other large venues.
Cisco's continued maturation
The new architecture is the latest iteration of Cisco's long evolution from a networking company to a full-stack observability and security provider, IDC President Crawford Del Prete told Fierce Network.
"The networking business is still a massive part of the company going forward, and it will continue to be," Del Prete said. "But observing what's happening within the network, middleware and application layer has become more important over time." Cisco has moved into those areas with its Splunk, AppDynamics and ThousandEyes acquisitions.
Networking providers, including Cisco, saw demand collapse immediately after COVID as operators overbuilt networks, Del Prete said. But AI infrastructure is increasing demand once again.
"People are modernizing their networks as they need to build out for an AI-enabled world," Del Prete said. Servers and storage demand alone grew 55% last year, and that demand is catching up to networking as well. "I've got this infrastructure to process the data, but now I need to move the data," Del Prete said. "And if I'm going to move the data more quickly and more efficiently, Cisco is going to benefit."
Cisco's future will depend on increasing demand beyond its traditional enterprise and telco customers to include hyperscaler cloud providers, Del Prete said. The vendor faces formidable competition from other vendors who are already entrenched in that market, such as Arista.
And Cisco needs to decide on its future in collaboration, Del Prete said. The company has a large Webex installed base, but faces intense competition from other players, notably Microsoft, which has the advantage of bundling Teams with its other products. Cisco needs to decide whether Webex should be part of its focus going forward, the analyst said.
Cisco's vision is ambitious, but it faces competition on multiple fronts. As Del Prete noted, Arista is a top competitor. That vendor is strong in high-performance data center networking and cloud networking, focusing on programmable, AI-driven network automation and telemetry, competing in hyperscale and enterprise data centers.
Juniper Networks emphasizes AI and automation with its Mist AI platform, and targets similar enterprise and service provider markets as Cisco.
Hewlett-Packard Enterprise is pursuing a $14 billion bid to acquire Juniper despite opposition from the U.S. Department of Justice. HPE and its Aruba business unit focus on AI-powered edge networking and wireless solutions, integrating AI for network management and security, particularly in enterprise and industrial IoT.
And cloud providers such as Amazon Web Services, Microsoft Azure, and Google Cloud compete with Cisco indirectly by offering integrated AI and networking services.