I’ve been covering Cisco long enough to remember when the biggest debates centered on routing protocols, switching architectures and whether voice over IP would ever become mainstream. Through every major technology cycle since then, Cisco has remained one of the industry’s constants.

The names around Cisco have changed. Entire categories have come and gone. Companies that once looked unstoppable have faded into the background or disappeared altogether. Through it all, Cisco has maintained a seat at the table.

Part of that longevity comes from a characteristic that has occasionally frustrated customers, partners and investors alike. Cisco rarely feels compelled to be first.

The company has traditionally preferred to wait until a market is large enough, customer demand is clear enough and the opportunity is significant enough to justify a long-term commitment. That approach can make Cisco look late when viewed through the lens of quarterly headlines. Viewed across decades, it often looks more like discipline than hesitation.

That thought kept coming back to me while following the announcements coming out of Cisco Live this week.

Individually, none of the announcements seemed particularly surprising. Cisco discussed Cloud Control, AgenticOps, AI infrastructure, observability, security and ecosystem partnerships. Any one of those topics could have been viewed as another product announcement or an incremental expansion of an existing strategy.

Taken together, however, they revealed something much larger.

For several years, Cisco has been assembling pieces of what increasingly appears to be one of the industry’s most comprehensive AI infrastructure strategies. The clues have been visible in acquisitions, product launches, earnings calls and executive commentary. Silicon One continued gaining traction. The company became increasingly vocal about AI networking. Security remained central to its roadmap. The Splunk acquisition looked larger than a traditional observability play. Cisco executives spent more time discussing AI infrastructure, AI factories and the operational demands these environments would create.

At the time, those developments felt like individual data points. After Cisco Live, they look more like breadcrumbs.

The individual pieces were visible. The broader architecture connecting them was not.

This week, that architecture became much easier to see.

Looking back, Silicon One now appears less like a standalone chip initiative and more like the foundation of a larger strategy. Cisco recognized early that AI infrastructure would require moving extraordinary amounts of data with low latency, high efficiency and predictable performance. Owning key components of that infrastructure stack provides control over both economics and innovation in a market where those advantages increasingly matter.

That infrastructure foundation extends beyond silicon. Cisco’s networking portfolio and optical technologies position the company to connect AI clusters, AI factories and data centers operating at unprecedented scale. While much of the AI conversation remains focused on models and GPUs, Cisco is advancing a different argument. Intelligence only becomes valuable when it can access information, collaborate with other systems and move data efficiently. Every one of those capabilities ultimately depends on infrastructure.

The network remains foundational to Cisco’s strategy, but Cisco Live made something else clear. The network is not the destination. It is the starting point.

Above the infrastructure layer sits observability, telemetry and security. This is where the Splunk acquisition begins to look increasingly strategic. When Cisco announced the acquisition, much of the discussion centered on observability. Today that explanation feels incomplete.

AI agents need context. Operators need context. Security teams need context. That makes telemetry one of the most valuable assets in the stack.

As enterprises deploy larger AI environments, operational visibility becomes increasingly important. Security platforms require it. Automated workflows require it. Governance requires it. The value of telemetry grows alongside the complexity of the systems consuming it.

The next layer is Cloud Control.

One of the more revealing aspects of Cisco Live was the language Cisco chose to describe the platform. The company could have positioned Cloud Control as a dashboard, management console or another version of the industry’s familiar “single pane of glass.” Instead, Cisco repeatedly referred to it as an operating layer.

That distinction matters.

Operating layers occupy a unique position within technology markets. They become the place where policies are enforced, workflows are executed and operational decisions are made. Over time, they become the place where ecosystems form because customers, developers and partners all have incentives to interact through the same operational framework.

We’ve seen this movie before. Microsoft built Windows. VMware built vSphere. AWS became AWS. Nvidia turned CUDA into one of the most valuable assets in technology. In every case, the infrastructure mattered. The layer sitting above it ended up mattering more.

Cisco appears to be pursuing a similar opportunity.

That context makes the launch of Cloud Control with more than 50 partners particularly significant. The number itself is less important than what it represents. Products attract integrations. Platforms attract ecosystems. The breadth of participation suggests Cisco views Cloud Control not as a management product but as a foundation that other companies can build upon.

That ambition also reflects the reality of enterprise AI. Organizations are not going to operate environments built entirely around a single vendor. AI infrastructure will span clouds, security platforms, applications, networking technologies and operational tools from multiple providers. Any company seeking to become an operating layer must be capable of managing that complexity rather than pretending it does not exist.

The emergence of AgenticOps adds another dimension to the strategy. Much of the current AI conversation revolves around increasingly capable models and agents. Cisco seems less interested in competing in that race than in addressing what happens after those agents arrive in the enterprise.

An organization deploying hundreds or thousands of agents inherits an entirely new operational burden. Security policies must be enforced consistently. Agent behavior must be observable. Interactions between systems must be governed. Organizations need confidence that automated actions remain aligned with business objectives, regulatory requirements and operational constraints.

Those concerns may not generate the same headlines as model benchmarks. They tend to become far more important once technology moves from experimentation into production.

This is where Cisco’s strategy begins to diverge from many of the companies competing for influence in the AI market.

Salesforce wants Agentforce to become a central environment for enterprise automation. ServiceNow is embedding agents throughout workflow management. Microsoft continues expanding Copilot across its ecosystem. The hyperscalers want AI workloads running on their clouds. OpenAI and Anthropic remain focused on building increasingly capable intelligence.

Cisco is taking a different route.

Rather than focusing primarily on the intelligence itself, the company appears focused on the environment in which that intelligence will operate. That perspective naturally leads toward networking, security, observability, governance and operational controls because those are the systems enterprises already depend on to run critical infrastructure.

Viewed through that lens, Cloud Control begins to look less like a product announcement and more like the connective tissue binding Cisco’s broader AI strategy together. Silicon One provides the infrastructure foundation. Networking and optics provide connectivity. Splunk delivers visibility and telemetry. Security provides protection and policy enforcement. Cloud Control sits above those layers as a common operational framework, while AgenticOps extends that framework into a world increasingly populated by autonomous systems.

Whether Cisco can successfully execute on that vision remains an open question. Every platform strategy eventually encounters tensions between openness and control, partner participation and vendor influence, customer flexibility and platform standardization. Technology history is filled with platform ambitions that looked compelling on paper and proved difficult to execute in practice.

What feels different after Cisco Live is not that Cisco has solved those challenges. It is that the company’s direction has become considerably easier to understand.

For several years, Silicon One, Splunk, AI networking, security investments and automation initiatives appeared to be related but largely independent efforts. This week they began to resemble components of a single architecture.

Cisco has spent decades adapting to successive waves of technological change. The AI era may prove to be another one of those transitions. The company is betting that enterprises will eventually need more than powerful models and faster infrastructure. They will need a way to operate, secure, observe and govern increasingly autonomous environments.

For a few years now, we’ve been looking at individual pieces. Silicon One here. Splunk there. AI networking over here. This week those pieces finally snapped into focus.

The mountain is above the waterline now.

Cisco has shown us the strategy.

Now it has to deliver on it.