Matilda Bailey has spent over two decades dissecting the plumbing of the global internet, moving from the early days of simple switching to the complex, high-velocity world of next-generation cellular and wireless solutions. As a leading specialist in enterprise networking, she has a reputation for looking past the flashy corporate earnings reports to find the structural shifts that actually dictate how businesses function. Following Cisco’s recent record-shattering fiscal year, where the company posted a massive $17.3 billion in quarterly revenue, Bailey argues that we are no longer just seeing a growth cycle, but a complete “supercycle” driven by the peculiar demands of artificial intelligence. In this conversation, she breaks down how the transition from training AI models to deploying autonomous agents is forcing a total architectural rebuild of the enterprise, from the silicon sitting in the data center to the Wi-Fi access points hanging in the office ceiling.
Agentic AI is fundamentally changing the way data moves, shifting us toward “scale-across” architectures. Can you explain why this transition causes such a massive surge in network traffic—reportedly up to fourteen times higher than traditional models—and what that means for a typical enterprise data center?
The shift we are seeing is essentially the difference between a predictable stream and a chaotic flood. For decades, we designed networks for “east-west” traffic, which was mostly about moving data between servers, or “north-south” for client-server interactions. Agentic AI completely throws that playbook out the window. Instead of a user asking a question and getting a single answer, you now have autonomous agents that are constantly chatting with each other, pinging APIs, and querying vector databases in real-time to complete complex tasks. This creates a relentless roar of multidirectional traffic that legacy networks simply weren’t built to handle. When we talk about “scale-across” architectures, we are describing a scenario where a single logical AI cluster isn’t confined to one room; it’s distributed across multiple physical data centers. When you link those GPUs together to act as one brain, the interconnect traffic explodes. Cisco’s data shows that traffic in these scale-across environments is roughly 14 times higher than what we see in traditional data center interconnects. For the IT pro on the ground, this means your annual 10% capacity upgrade is a thing of the past. If you aren’t moving toward non-blocking topologies and high-density 400G or even 800G switching right now, your AI initiatives are going to hit a wall before they even get started.
One of the standout figures from the recent fiscal report was the $4 billion in AI infrastructure orders, much of it driven by Cisco’s own Silicon One and Acacia optics. Why should an enterprise IT team care about this shift toward vertically integrated silicon rather than just buying off-the-shelf components?
It comes down to the sheer economics and efficiency of the “token.” In the AI era, every bit of power and every millisecond of latency costs money. By controlling the silicon, the optics, and the software stack internally, companies can bypass the markups of merchant silicon and, more importantly, they can bake telemetry and power controls directly into the chip. We saw $1 billion in quarterly orders for Acacia coherent optics alone, which tells you that the industry is desperate for tighter integration to save on power and space. When a vendor controls the silicon, like with the Silicon One architecture that powered 60% of those massive AI orders, they can deliver much better power efficiency per gigabit. For a network engineer, this isn’t just a hardware spec; it’s about having a unified view of the network from the physical chip all the way up to the cloud management plane. If you keep buying legacy, off-the-shelf switching components, you’re essentially opting into higher power bills and more complex troubleshooting. You want hardware that is “programmable” at the chip level so it can adapt as AI models evolve over the next three to five years.
We are seeing a massive surge in campus networking upgrades, with Wi-Fi 7 access points now making up over half of all wireless orders. Is this just a standard hardware refresh, or is there a more urgent driver behind this modernization?
It’s definitely more than just a routine refresh; it’s a strategic pivot. Campus networking grew by 20% year-over-year because the office environment is being redesigned for workspace AI and high-density IoT. Wi-Fi 7 isn’t just “faster Wi-Fi”; it’s a deterministic, high-throughput edge network that allows for local AI inference and spatial computing. But there is also a “fear factor” driving this. Many organizations are finally auditing their infrastructure for what we call LDOS—Last Day of Support gear. This is old hardware that can’t be patched against modern threats and certainly isn’t ready for post-quantum security. Chuck Robbins mentioned that customers are increasingly using tools like Cisco IQ to find these “ticking time bombs” in their racks. In today’s climate, where AI readiness is a board-level mandate, IT managers are finally getting the budget they need to retire that unpatchable technical debt and move to a modern fabric. You cannot run a cutting-edge AI assistant on a switch that was manufactured before the cloud was even a standard concept.
With the rise of “harvest now, decrypt later” threats, there is a lot of talk about post-quantum cryptography (PQC). How is the integration of security and observability changing the daily operations of “NetSecOps” teams?
The days of treating security as a separate layer—a firewall you just slap on top of the network—are officially over. We saw the security segment jump by 14%, with over 1,500 new customers flocking to architectures like Hypershield and AI Defense. This is because the attack surface has expanded beyond what humans can manage. When you have thousands of autonomous agents moving data around your network, you need security that is embedded in the actual fabric of the routers and switches. The PQC compliance we are seeing in the latest hardware is a direct response to the “harvest now, decrypt later” strategy, where bad actors steal encrypted data today hoping to decrypt it once quantum computers are viable. By building PQC-compliant routers and smart switches now, enterprises are future-proofing their data against threats that haven’t even fully materialized yet. For the operations team, this means breaking down the silos between networking and security. You need inline AI guardrails that monitor agent behavior in real-time, because a human operator simply isn’t fast enough to catch a rogue agent before the damage is done.
Cisco reported that their internal AI assistant handled 75 million prompts and that 145,000 customer cases were resolved with zero human intervention. As autonomous operations become the daily reality, what is your forecast for the future of the network engineer’s role?
I believe we are entering the era of “declarative” networking, where the engineer stops being a mechanic and starts being an architect. We recently saw a case where a network engineer spent eight grueling hours trying to figure out why video calls were dropping, only to have an AI tool like Cisco IQ or AI Canvas identify the specific faulty access point and provide a step-by-step fix in just a few minutes. That is a game-changer for quality of life. When 145,000 cases can be solved without a human ever touching a ticket, it means the “grunt work” of CLI configuration and reactive troubleshooting is dying. In the next five years, the most successful network engineers will be the ones who focus on defining intent and validating the insights that the AI provides. You won’t be logging into individual boxes to change settings; you’ll be orchestrating a massive, self-healing system. The role is shifting toward a high-level strategic position where you manage the policy and let the AI handle the thousands of micro-adjustments required to keep the “networking supercycle” moving at full speed. My forecast is that we will see the total disappearance of the traditional NOC in favor of highly automated, AI-driven command centers that focus on proactive optimization rather than putting out fires.
