After Cisco posted record third-quarter results in mid-May, I wrote that the question I was watching next was how much AI networking demand broadens beyond hyperscalers and into enterprise, sovereign, and cloud-adjacent buildouts. Last week I went looking for the answer in person. I spent three and a half days at Cisco Live 2026 in Las Vegas, the longest I’ve stayed at any event this year, sitting through both keynotes, meeting with executives across the company, and pressure-testing the announcements against what I’m hearing from buyers.
My core takeaway: Cisco is making the most credible end-to-end claim in the industry that the network is becoming the AI platform. The new Cisco Cloud Control console is the make-or-break piece, Silicon One is the company’s most underappreciated weapon, and the “networking super cycle” Cisco pitched from the main stage is both its most self-serving claim and its most correct one.
Cisco CEO and chairman Chuck Robbins opened the show with a thesis he has been sharpening for years: the network is more powerful than the node, now applied to GPUs, models, inference, and agents. He called this the most consequential Cisco Live ever, which he says every year. This time he might actually be right. Robbins has been right about the network at every major inflection, and I shared my full keynote notes in real time from the audience.
Cisco president and chief product officer Jeetu Patel delivered the line of the show, that “humans click but agents swarm”, and the announcement payload backed up the framing. Cisco Cloud Control was the centerpiece: one console managing Catalyst, Meraki, Nexus, firewall, and Webex with cross-domain correlation and purpose-built models, running natively through MCP, now in controlled availability in the United States with more than 50 partners on day one. DJ Sampath demoed it live, including OpenAI Codex embedded through Cloud Control Studio.
The rest of the payload was dense. Cisco’s largest-ever campus and branch refresh unifies Catalyst and Meraki, led by the quantum-safe 9550 core switch and Live Protect, which shields gear in runtime without a reboot. That last piece is what customers will actually pay for. On security, Cisco built on its early-May intent to acquire Astrix Security for non-human identity, folded the Galileo evaluation technology into Splunk observability, advanced the Splunk-powered Cisco Data Fabric and an agentic SOC, and launched a sovereign critical infrastructure portfolio starting in Europe. Cisco IQ, the AI-driven services experience, stood at more than 1,700 customers, 90% of them self-onboarded, as of the opening keynote, roughly five weeks after general availability.
Robbins also put the post-Mythos security reality in plain terms, repeating that these models are as bad today as they will ever be. Cisco scanned 1.8 billion lines of code across 25 languages in eight weeks, work it says would have taken roughly eight years before frontier models. That squares with the AI-finds-vulnerabilities arms race I flagged in May, and it’s why Patel’s framing of agents as teenagers, intelligent but with no fear of consequence, will travel further than most of what was said on stage.
A networking super cycle is exactly what you’d expect a networking company to predict. Here’s the thing: the evidence says it’s happening anyway. Cisco cited a study showing agents generate about 450% more network traffic than humans, and Robbins expects AI-related network traffic to triple over the next three years, with more than 90% of customers saying they must modernize and harden their networks. I agree with the direction. Look at what came out of Computex and Microsoft Build during the same week; agent traffic is about to explode.
The capital backdrop supports it. McKinsey projects $6.7 trillion in cumulative data center capital outlays by 2030, with $5.2 trillion of that for AI-capable facilities. GPUs get the headlines in that buildout, but for years I have maintained that the bottleneck in AI clusters is the interconnect, and the networking layer is the underexamined slice of that capex.
Cisco is now putting numbers behind the thesis. The company posted record revenue of $15.8 billion in its fiscal third quarter, up 12%, with networking product orders accelerating more than 50% year over year. It booked $5.3 billion in AI infrastructure orders from hyperscalers in the first three quarters and raised its full-year target to $9 billion, up from $5 billion. And here’s the early answer to my May question: $900 million in separate AI orders from neocloud, sovereign, and enterprise customers. The broadening has started. It’s also still a tenth of the full-year hyperscaler target, which tells you how early we are.
Cisco Cloud Control is the make-or-break announcement of the show. Customers have asked for one pane of glass across the Cisco portfolio for a decade, and this is the best embodiment of Cisco One I’ve seen. The part most people miss: Cisco’s smart switches have processors inside them, and Cloud Control puts code directly on those programmable platforms while agents operate natively through MCP. If you run a Cisco estate, this is the screen your team will live in. The risk is migration and execution, not vision, and embedding OpenAI Codex through Cloud Control Studio could be a real edge over networking peers.
My colleagues saw the same shift from different angles. Melody Brue, who covers collaboration and was at the show with me, observed that Cisco is no longer positioning itself as a networking vendor but as the AI-native control plane for critical enterprise infrastructure, with Webex stepping in as a first-class citizen rather than a side story. And Mike Leone called Cloud Control the most coherent platform story Cisco has assembled in years; he still thinks the company is underrated as a full-stack agentic AI player. I agree with both of them.
Silicon One is Cisco’s most underappreciated asset. I made that point at Cisco Live 2025, when I pushed the company to talk more about its silicon, and a year later the gap between what Silicon One is and what the market knows it is hasn’t closed. The company has convinced itself. It hasn’t convinced the market. That’s an awareness and familiarity problem, and it’s fixable.
The silicon itself is doing its job. The G300, which Cisco unveiled at Cisco Live EMEA in February, doubles the G200 to 102.4 Tbps and is slated to ship in the second half of this year, anchoring the scale-out story for massive AI clusters. The P200 takes the scale-across job, using deep buffering to link distant data centers into one logical computer as power limits force buildouts to distribute. That deep-buffering play is where Cisco is ahead of the market narrative. Proof is showing up in the order book, too: the third quarter included five new AI design wins, among them the first two P200-based systems.
Cisco really thinks it can give both Jericho and Tomahawk a run for their money, and that was my exact reaction from the show floor. After a week of conversations, I believe it. People I talked with are very, very convinced this silicon can take on Broadcom’s franchises, hyperscalers want silicon diversity, and Cisco has both the silicon and the capacity to deliver it.
The NVIDIA relationship deserves honest framing. Cisco deepened the partnership across Spectrum-X integration and shared silicon, and being inside Spectrum-X is a win. It’s also co-opetition in which Cisco needs NVIDIA more than the reverse, and Cisco is one of several Ethernet partners, not the only one.
Four risks deserve attention. First, Patel described the stack as tightly integrated and loosely coupled. That’s the right design goal and a brutally hard one; few companies do both well. Second, the acquisition cadence across Splunk, Galileo, and Astrix shows Cisco understands that action control and non-human identity are the real security battleground, but integration is historically challenging, and Leone’s caution applies: a chunk of the agent-security stack is still when-and-if-available, with some ambitious pieces slipping to later in 2026. He also raised the right pricing question with TechTarget, asking where the meter sits when a non-Cisco agent calls into Cisco data, since Cloud Control comes in three tiers with a la carte token packs. The open ecosystem promise gets tested fast right there.
Third, Cloud Control’s risk isn’t the vision; it’s moving an enormous installed base onto a new console without breaking operations. Fourth, the super cycle needs enterprises to deploy agents fast enough to drive the demand Cisco promised. I think they will. But I’ll own my track record here: I thought industrial AI gear would lock in more this year, and it hasn’t. I’m adjusting my timing, not my thesis.
If I were at lunch with Robbins and Patel, I’d put four things on the table. Market the silicon like it matters, because Silicon One is a strategic asset hiding in plain sight, and awareness, not capability, is the constraint. Show Cloud Control migrations at installed-base scale by early 2027, not demos. Close Astrix and ship visible Galileo-driven enhancements inside Splunk before the end of 2026, the bar Leone set in his analysis of the Galileo acquisition. And turn that $900 million of neocloud, sovereign, and enterprise AI orders into billions, because the broadening is what validates the super cycle beyond hyperscaler capex.
One more, for the long game: keep investing at the edge. I said it from the keynote audience and I’ll say it again here. Edge, campus, and branch enablement is Cisco’s biggest play, where it can win the big prize and the home run lands when industrial and physical AI deployments accelerate.
Two years ago, Cisco wasn’t even in the enterprise AI conversation. Today the stock sits just off the all-time high it set during show week, and I credit the company for earning it, as I said on this week’s Six Five Pod. It broke out its AI revenue when peers wouldn’t, built silicon that hyperscalers are actually designing in, and told the most coherent platform story in its history. Robbins is strategic and tells it like it is, and in an age of agents, trust and security mean a lot. Trust is the asset Cisco has spent four decades building.
Long term, I think Cisco does well. Near term, two proof points decide the narrative: the fourth-quarter print against that $9 billion AI order target, and the first Cloud Control migrations at scale. Cisco set the bar in Las Vegas. Now it has to clear it.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。