Cisco Bets Big On A Unified Platform For The Agentic AI Era

by Tony Bradley, via Forbes

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.

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.

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.