AI isn’t just eating software — it’s rewriting the infrastructure underneath it.
At Cisco Live EMEA in Amsterdam, Cisco rolled out a big slate of updates aimed at one reality: AI workloads are shifting from “experiments” to always-on, real-time, agentic systems — and that breaks old assumptions about networks, operations, and security.
Cisco’s pitch is simple: if your AI strategy is serious, you need AI-grade networking, AI-native operations, and security that understands AI traffic (not just ports and packets).
Here’s what they announced — and why it matters.
1) New Silicon One G300: “Gigawatt-scale” AI networking (and a blunt GPU message)
Cisco’s headline hardware move is Silicon One G300 — new switch silicon designed for the kind of AI clusters that are now measured in power plants, not server rooms.
What Cisco is really saying between the lines: your GPUs are expensive, and your network is wasting them.
They’re touting “collective” networking optimizations that improve network utilization and cut job completion time — because in AI training and inference, the network is often the silent bottleneck that turns premium GPUs into idle heaters.
Cisco is pairing the chip with G300-powered systems in its Nexus and Cisco 8000 families, aimed at everyone building large AI fabrics: hyperscalers, neoclouds, sovereign deployments, service providers, and enterprises.
Translation: this isn’t just “enterprise switching.” This is Cisco stepping into the data-center AI arms race with a performance story built around GPU efficiency.
2) Nexus One: one management plane across on-prem + cloud data centers
The other half of “AI infrastructure” isn’t speeds and feeds — it’s operations.
Cisco introduced Nexus One as a unified management plane that spans on-premises and cloud-based data center environments. The bet here is that AI networking is becoming too complex to operate as a set of disconnected tools.
Translation: fewer silos, fewer dashboards, less “tribal knowledge,” and more centralized control over how AI fabrics run.
3) AgenticOps: Cisco wants to run your IT stack like an AI system
Cisco is pushing a new operational story called AgenticOps — essentially using system-wide telemetry and automation to reduce complexity across:
- networking
- security
- observability
The key point: Cisco is framing the next era of IT as agents managing agents — and it’s tying this to telemetry drawn across multiple Cisco domains (including Splunk).
Translation: the old model of “humans click around in tools” doesn’t scale when your environment is full of autonomous systems and constantly shifting AI traffic patterns.
4) AI Defense gets its biggest update: governance + runtime protection for AI tool use
This is the part that’ll make CISOs sit up.
Cisco says it’s delivering the biggest updates yet to AI Defense, with emphasis on:
- AI supply chain governance (what models/tools are being used, where they come from, how they’re controlled)
- runtime protections for agentic tool use (so agents don’t get manipulated, tricked, or redirected into doing the wrong thing)
This signals where security is going: not just protecting data — protecting agent behavior.
Because if an AI agent can trigger actions (tool calls, requests, workflows), the attack surface isn’t just “access.” It’s intent.
5) SASE upgrades + “AI traffic optimization”: keeping agent workflows safe and reliable
Cisco also leaned into SASE updates — and what stands out is the idea of inspecting agentic AI interactions in a more intent-aware way.
In plain English: security systems will have to understand what an agent is trying to do and how it’s doing it — not just whether traffic looks suspicious in a legacy sense.
Cisco’s message: agentic traffic is different, and the security fabric has to evolve to keep it both safe and reliable.
6) Sovereignty isn’t a footnote — Cisco is productizing support for it
Another underrated part of the announcement: Cisco is turning “sovereignty” into an operational offering, not a marketing line.
They emphasized support options for:
- air-gapped environments
- on-prem setups
- hybrid deployments
And they highlighted Cisco Critical National Services Centers (CNSCs) across Europe, designed for organizations with strict requirements and controlled support channels.
Translation: AI is pushing more governments and regulated industries to demand tighter boundaries — and Cisco wants to be the vendor that can say “yes” without improvising.
The big takeaway
Cisco is positioning itself as a unified platform for the AI era, built around three pressures every serious enterprise is feeling:
- AI clusters scale fast — and networks become the limiter
- Ops complexity explodes — and humans can’t babysit everything
- Agentic AI changes security — because behavior becomes the risk
This launch is Cisco saying: AI isn’t a feature you bolt on. It’s a new operating environment.
What IT leaders should do next
If you’re a CIO/CTO/CISO reading this, here’s the practical checklist:
- Map where AI training/inference bottlenecks are happening (network is often guilty)
- Consolidate observability and telemetry (fragmentation kills speed)
- Treat “AI tool use” as a security domain (governance + runtime controls)
- Review SASE posture for agent-heavy workflows (policy needs to evolve)
- If you have sovereignty constraints, pressure-test vendor support models early (not after you deploy)
AI is moving faster than enterprise change management. Cisco is betting that the winners will be the ones who upgrade the plumbing first — and make it secure enough to trust autonomous systems at scale.
