Cisco Bets Big on a Unified Platform for the Agentic AI Era

Every vendor in the world is talking about agentic AI right now. Most of them are still figuring out what it actually means for their products. Cisco, on the other hand, just showed up in Amsterdam at Cisco Live EMEA and dropped a stack of announcements that suggest they’ve already moved past the “what does this mean” phase and into the “here’s what we built” phase. New silicon, expanded security, autonomous IT operations, sovereign infrastructure—all at once. It’s a lot. And it reflects a bet that the network is about to become the most critical layer in the AI stack.

The Silicon at the Center

The centerpiece is the Cisco Silicon One G300—a 102.4 terabits-per-second switching chip designed for AI cluster networks. In simulations, it delivered 33% higher network utilization and 28% faster job completion times compared to non-optimized infrastructure. Those are solid numbers, but the real story is what they mean in dollar terms. GPUs are absurdly expensive, and every second one sits idle waiting on data is money wasted. The G300 is basically Cisco’s answer to the question: how do you make sure the network never becomes the reason your GPU investment underperforms?

Nick Kucharewski, who leads Cisco’s silicon efforts, framed the shift in a blog post accompanying the announcement. “AI infrastructure is at a turning point,” he wrote. “While multibillion-dollar buildouts show no sign of slowing down, the focus is shifting from rapid deployment—a ‘just build’ approach—to energy efficiency, operating costs, and total return on invested capital.”

I think that’s an important pivot. We’ve spent a couple of years watching companies throw money at AI infrastructure as fast as they can. Cisco is signaling that the next phase is about making that infrastructure actually pay for itself.

Matt Eastwood, senior vice president of IDC’s Enterprise Infrastructure and Datacenter Group, sees the same dynamics. “As AI adoption moves beyond hyperscalers and scales across enterprises, neoclouds, and sovereign environments, network architecture is becoming a defining constraint on performance, cost, and sustainability,” he noted, adding that Cisco’s approach of combining silicon, liquid-cooled systems, advanced optics, and integrated operations “speaks directly to the next phase of AI infrastructure.”

Full Stack, Not Just a Chip

I sat down with Jeetu Patel, Cisco’s president and chief product officer, ahead of the announcements, and what struck me was how emphatic he was that this isn’t just a chip launch. Patel and I talk on a fairly regular basis—he’s easily one of the executives I speak with most often—and he was more animated about this than usual. The pitch is that Cisco is delivering the full stack: silicon, systems, software, optics, and security, all designed to work together.

He walked me through the distinction between hyperscalers, who mostly just want components, and everybody else—neo clouds, sovereign clouds, service providers, enterprises—who need the whole package. Cisco’s argument is that the G300-powered N9000 and 8000 series switches bring hyperscaler-class infrastructure to every class of customer, with support for air-cooled and liquid-cooled designs, 1.6 terabit optics that improve energy efficiency by 70%, and built-in programmability so you don’t have to wait 18 months for a new tape-out every time requirements shift.

There’s also a shift in how Cisco talks about what the network actually needs to handle now. This isn’t just about training runs anymore. Patel made the point that agentic workflows—agents running autonomously around the clock—demand a fundamentally different level of network bandwidth. As he put it, “You can’t make up the deficiency in bandwidth by overworking compute. It doesn’t work.”

I’ve been writing about the gap between AI hype and AI reality for a while now, and that’s one of the more practical observations I’ve heard. It doesn’t matter how powerful your GPUs are if the network can’t keep up.

Neil Anderson, VP and CTO of cloud, infrastructure, and AI Solutions at WWT, seemed to agree. “This is the fastest we’ve seen Cisco move, and it’s exactly what our clients need to accelerate their AI journeys,” he said, pointing to the G300-powered N9000 and Nexus One as the combination that extends Cisco’s trusted data center networking into AI workloads.

AI That Runs the AI

Beyond silicon, Cisco is expanding what it calls AgenticOps—using AI to manage the infrastructure that runs AI. I know that sounds a bit like inception, but it actually makes sense when you think about scale. The traditional IT operations model—file a ticket, wait for a human to diagnose it, escalate, resolve—just doesn’t work when systems are operating at machine speed. I’ve talked with enough SOC analysts and IT operations folks over the years to know that the manual approach was already breaking down before agentic AI entered the picture. AgenticOps promises autonomous troubleshooting, continuous optimization, and risk assessment before changes go live. Humans stay in control of outcomes. Machines handle the volume and speed.

Security for an Agentic World

Security is the other major thread running through these announcements, and honestly it’s the one I’m watching most closely. Cisco is rolling out its biggest update yet to AI Defense, adding supply chain governance and runtime protections for agentic tool use. The company’s SASE portfolio now includes intent-aware inspection—looking at not just what AI agents are doing, but why.

That’s a meaningful distinction. An agent making an API call to pull sales data for a report looks identical to one making the same call as part of a data exfiltration attempt—at least on the surface. Understanding intent is where the real security value is. There’s also a post-quantum cryptography angle, with Cisco IOS XE 26 introducing full-stack post-quantum protections across its secure routers and smart switches. Quantum threats may still be years out, but the data being transmitted today could be harvested now and cracked later. Getting ahead of that is smart.

Sovereignty and the Hybrid Future

For regulated industries and governments—particularly in Europe—Cisco announced that its Critical National Services Centers are now operational across France, Germany, the United Kingdom, and Spain, with Italy under development. These run under strict sovereign controls: dedicated facilities, segregated processes, cleared personnel. Patel told me he expects sovereign clouds to become increasingly important as countries get more nationalistic about where their data lives and where tokens get generated. I asked him about the broader trend of enterprises wanting to keep more infrastructure local rather than shipping everything to hyperscalers, and his take was nuanced—hyperscalers will keep growing fast, but the world will always be hybrid, and sovereign requirements are only going to accelerate that.

Agents as Teammates

What stuck with me most from my conversation with Patel was how he framed the broader trajectory. He sees every job and every workflow getting completely rethought. He pointed to Cisco’s own experience—AI Defense, the product they launched last year, is now being developed with 100% AI-generated code. Not as a gimmick, but because they fundamentally changed their development process. They moved from agile to what Patel called spec-driven development, where developers create markdown files that become context for agents to write the code. The result is better-documented, better-structured code. And the bottleneck flipped—writing code is no longer the constraint. Reviewing it is.

Patel also gave me a perspective on what an AI-ready enterprise looks like in two years. “I think you have to think about agents as teammates, not as tools,” he told me. Agents will handle the work that humans either aren’t good at, don’t want to do, or simply can’t do. And every person, in his view, essentially becomes a manager of agents. That tracks with conversations I’ve been having across the industry. Tim Ferriss wrote about outsourcing tasks to streamline productivity almost 20 years ago in his book “4-Hour Work Week.” This is very similar logic, just with AI agents instead of Upwork freelancers.

The Bigger Bet

Taken together, the Amsterdam announcements paint a picture of a company positioning itself as the connective tissue for the agentic AI era. Cisco has always been a networking company at its core. But the argument it’s making now is that networking, security, observability, and sovereignty aren’t separate things anymore. They’re facets of one platform problem.

Whether that unified vision holds up in practice remains to be seen. I’ve covered this industry long enough to know that “single platform” is a promise that gets made more often than it gets kept. But Cisco has the installed base, the silicon, and—with agentic AI introducing entirely new categories of network traffic and security risk—a stronger case for integration than it’s had in a long time.

The AI era is still early. Cisco is making a big bet that the answer starts with the network. Given how central connectivity is to everything agents will need to do, it’s not a bad bet to make.