The AI agent your team just deployed is quietly doing something nobody planned for: it's hammering your network. Every task it performs generates 450% more network traffic than a human doing the same job. Multiply that across thousands of agents running in parallel — which is exactly where most enterprises are headed — and you have an infrastructure crisis hiding in plain sight.
A new joint study from Cisco and Foundry, "No Time to Wait: The Accelerating Impact of AI on Campus and Branch Networks," surveyed thousands of enterprise IT leaders across 30 markets and found a pattern that should concern every CIO, CTO, and CFO making AI investment decisions right now: 75% of organizations say they are more confident in their AI strategy than in their network's ability to deliver it.
That gap — between AI ambition and infrastructure reality — is about to become very expensive.
The Traffic Explosion You Didn't Model
When most enterprises model AI costs, they account for compute, licensing, and implementation services. Almost nobody budgets for the network impact of agentic AI workloads.
Here's why that's a critical oversight. Traditional enterprise applications — CRM, ERP, collaboration tools — generate predictable, human-paced traffic. A user clicks, data moves, a response appears. The network was designed for this rhythm.
AI agents work differently. They don't wait for humans. They run continuously, call APIs in parallel, query databases at machine speed, and chain actions across dozens of systems without stopping. According to the Cisco/Foundry data, a single AI agent generates up to 450% more total traffic than a human performing the same task. Approximately 70% of that traffic is AI inference — and inference paths are now mission-critical in a way legacy network architects never anticipated.
The Cisco/Foundry report found that enterprise network traffic is already up 34% as a direct result of AI workloads. That's with current adoption levels. Companies broadly deploying AI expect total network traffic to triple in the next three years.
Three years sounds comfortable. It isn't. The same research found that most enterprises will hit campus and branch network capacity limits within 24 months — not three years, two. That's the timeline for action, not for planning.
The 2-Year Capacity Wall
When I talk to infrastructure leaders, the conversation about AI capacity usually goes one of two ways. Either they say "we just upgraded last year, we should be fine," or they say "we know we need to modernize but haven't gotten budget approval yet."
Both groups are likely heading toward the same wall.
Capacity planning built on 5-year horizons is colliding with a 24-month reality. The Cisco/Foundry data is specific: 76% of respondents admit their network needs upgrades. But only 30% of aggressive AI adopters — the companies already moving fastest on AI deployment — report being fully prepared to support projected AI growth across the network.
Let that land. The companies that are most ahead on AI adoption are the most exposed to infrastructure failure. The faster you move, the faster you hit the limits.
The most stressed part of enterprise networks isn't the WAN or the data center. It's campus Wi-Fi, which saw the greatest increase in performance and capacity requirements — cited by 50% of organizations in the study. Branch locations (31%), WAN traffic (37%), and data centers (37%) are also under significant pressure.
For business leaders, this translates directly to two risks. First, AI initiatives stall or deliver inconsistent results because the network can't reliably support inference workloads. Second, AI-related downtime becomes a real operational risk. The Cisco/Foundry study included data from Splunk's Hidden Costs of Downtime 2026 report, which surveyed 2,000 executives from Global 2000 companies: every single technology leader surveyed admitted their organization has experienced some form of AI-related downtime. Not most. All of them.
That's not a future risk. It's a current one.
The Security Blindspot Nobody Is Talking About
The capacity problem is actually the easier problem to solve. You can budget for network upgrades, accelerate infrastructure modernization, and close the gap over 18-24 months with the right investment.
The security problem is harder, and it's compounding.
Cisco's President and Chief Product Officer Jeetu Patel offered a framing at Cisco Live in June that stuck with me: "Agents are like teenagers. They're supremely intelligent, but they have no fear of consequence."
That's not hyperbole. It's a precise description of how agentic AI creates attack surface. Traditional enterprise security was designed for human-paced actions. A person logs in, performs tasks, and creates an audit trail. Security tools monitor for anomalies in that predictable flow.
AI agents break every assumption in that model. They operate continuously. They call APIs and interact with external services without a human in the loop. They can be compromised through prompt injection — where malicious instructions embedded in data manipulate the agent's behavior — or data poisoning, where corrupted inputs alter decision-making. And they can execute at machine speed, meaning a compromised agent can exfiltrate data, modify records, or trigger downstream actions in seconds before any monitoring system fires an alert.
The Cisco/Foundry research quantified how exposed enterprises already are:
- 77% say AI has already increased their attack surface over the last year
- 72% say AI-related security threats are evolving faster than their ability to adapt
- 69% believe AI workloads have introduced new blind spots in network security monitoring
- 59% say they lack adequate visibility into AI-related traffic flows across their network
That last number deserves attention. If you can't see the traffic, you can't defend it. And most enterprises are flying partially blind on AI workload visibility right now.
The adversarial side is equally alarming. CrowdStrike's 2026 Global Threat Report found an 89% increase in attacks by adversaries using AI. The window between a disclosed vulnerability and a working exploit — once measured in weeks — has compressed to minutes. At Cisco Live, CEO Chuck Robbins reported that in the past eight weeks, Cisco used AI to scan 1.8 billion lines of code across 25 programming languages. That same work, pre-AI, would have taken approximately eight years.
The same capability available to defenders is available to attackers. Neither side has a structural advantage — but defenders are protecting a complex, distributed enterprise while attackers only need to find one opening.
What the Board Needs to Hear
If you're a technical leader — CIO, CTO, CISO — you already understand the infrastructure and security dimensions here. The challenge is making this legible to business stakeholders who are excited about AI ROI but not yet thinking about infrastructure debt.
Here's the framing that lands in my experience talking to enterprise leaders:
The AI initiative and the network upgrade are the same project. Separating them creates a situation where AI delivers spotty, inconsistent results — and the blame lands on the AI strategy rather than the infrastructure bottleneck. Getting aligned on this early prevents political problems later when deployments underperform.
The cost of reactive network upgrades is significantly higher than planned ones. Emergency network capacity expansions typically cost 3-5x more than planned upgrades, and they happen under pressure — increasing the risk of configuration errors and security gaps. The business case for proactive investment is straightforward.
AI-related downtime is already happening industry-wide. The fact that 100% of technology leaders in the Splunk survey reported AI-related downtime is a powerful data point for the board. This isn't a hypothetical. It's the current baseline. The question isn't whether it will happen; it's how prepared you are to minimize the impact and recover quickly.
For CFOs specifically: the network modernization cost belongs in the AI investment business case, not as a separate IT line item that competes for budget. Enterprises that model AI ROI without accounting for infrastructure investment are systematically overestimating their returns.
A Framework for Action
Based on what the data shows, here's how I'd prioritize the response:
Immediate (0-90 days): Get visibility. You can't manage what you can't see. The 59% of enterprises that lack adequate visibility into AI-related traffic flows need to close that gap before anything else. Deploy network observability tools that can differentiate AI inference traffic from traditional application traffic. Establish a baseline now, before additional agent deployments compound the problem.
Short-term (90-180 days): Run a capacity assessment against your actual AI deployment roadmap — not your current deployment. Map the traffic implications of your planned agent rollouts and identify which campus locations, branch offices, and WAN links will hit constraints first. Build the upgrade case before you need it urgently.
Medium-term (6-18 months): Implement agentic-specific security controls. Traditional perimeter security doesn't protect against prompt injection, agent identity spoofing, or data poisoning. Security architecture needs to evolve alongside agent deployments. This means agent identity and authorization frameworks (not just human identity), runtime behavior monitoring for agents, and governance policies that define what agents are authorized to do and with what data.
Structural: Establish a cross-functional AI infrastructure council that includes network engineering, security, and the business units driving AI adoption. The organizational gap between "AI strategy" and "network operations" is the root cause of the readiness problem. Closing the gap structurally prevents it from recurring with every new AI initiative.
The Gartner Signal
One data point from Gartner's Predicts 2026: AI Agents Will Transform IT Infrastructure and Operations provides useful context on the scale of what's coming. According to Gartner, approximately 70% of enterprises will deploy agentic AI in IT infrastructure operations by 2029, up from under 5% in 2025.
That's a 14x increase in four years. The traffic, security, and capacity implications of that trajectory are exactly what the Cisco/Foundry research is documenting in advance. Enterprises that act now — while that number is still in the single digits — have a meaningful window to build the foundation properly. Enterprises that wait until they're in the majority will be building under fire.
The irony is that the most aggressive AI adopters — the organizations moving fastest to capture competitive advantage — face the most acute infrastructure risk. Speed without infrastructure readiness doesn't produce AI ROI. It produces downtime, security incidents, and disappointed stakeholders.
The Bottom Line
The conversation about enterprise AI has been dominated by model capabilities, use case selection, and change management. Those are real challenges. But the infrastructure layer — the network, the security architecture, the visibility tooling — is becoming the binding constraint on AI success.
Cisco and Foundry's research makes the math unavoidable: AI agents generate 450% more traffic than humans, most enterprises will hit network capacity limits within 24 months, and 59% don't have adequate visibility into their AI traffic today. Meanwhile, the attack surface is expanding faster than most security teams can adapt.
The enterprises that figure this out now will have a durable advantage — not just in deploying AI, but in deploying it reliably, securely, and at scale. The ones that don't will find that their AI strategy collides with an infrastructure wall at exactly the moment they expect to be harvesting returns.
Network modernization isn't a cost center decision. In 2026, it's an AI strategy decision.
Sources: Cisco/Foundry "No Time to Wait: The Accelerating Impact of AI on Campus and Branch Networks" (2026); Cisco Live keynote, June 2026; CrowdStrike 2026 Global Threat Report; Gartner Predicts 2026: AI Agents Will Transform IT Infrastructure and Operations; Splunk Hidden Costs of Downtime 2026.
