Scott Harrell, President and Chief Executive Officer, Infoblox.getty​Every CEO I speak with today is racing to move from AI experiments to real, scaled impact. Pilots are moving to production, customer experiences are being redesigned, and entire workflows are being reimagined and automated with AI. Yet an uncomfortable pattern is emerging: many of the most ambitious AI programs are being built on infrastructure that was never designed for AI. Legacy infrastructure lacks the real-time telemetry, scale, adaptability and robustness that new AI deployments require. This creates massive business performance and continuity risk as AI becomes increasingly mission-critical, and it can erode employee, partner and customer trust in AI-powered transformations even when the underlying AI technology and ideas are sound. The culprit? A legacy networking foundation that is often ignored as part of most AI transformation initiatives. When AI fails, it’s rarely the AI.There's a common thread in some of the most publicized digital outages of the last few years. Whether an AI platform goes dark or a cloud region stumbles, the root cause analysis almost always points back to the same place: the shared, central services that connect everything together. A prominent example occurred in October 2025, when a 15-hour outage in Amazon Web Services' (AWS) primary U.S. region took down a significant portion of the internet. The incident was triggered by a conflict between two automated systems that simultaneously attempted to update the same network configuration, resulting in the accidental deletion of critical routing information for DynamoDB, one of AWS's core database services. Because this service underpins hundreds of major applications and platforms, the error cascaded rapidly and blocked traffic across a wide swath of the internet. The business impact was substantial. Several high-profile AI platforms were knocked offline, including Perplexity AI (down for about 4 hours), CharacterAI (roughly 3 hours) and OpenAI's ChatGPT. Consumer platforms including Snapchat and Reddit also experienced disruptions. The incident underscored how deeply concentrated the internet's infrastructure dependencies have become. A single automated error in one provider's system was enough to disrupt services used by hundreds of millions of users worldwide. It's easy for the internal narrative to become “AI doesn’t work here," but in reality, the weak link is often the basic roads and addresses of the digital environment: how systems are named, how they are located and how they connect. Why AI Breaks Old Assumptions For decades, the core network services that handle these basics quietly scaled alongside applications and devices. Servers lived for months and traffic followed human business hours. If something went wrong, people noticed, worked around it and raised a ticket. AI changes those rules. Workloads are short‑lived and constantly moving across data centers, clouds and regions. A single AI interaction may touch many different internal and external services in milliseconds, multiplying the number of lookups happening behind the scenes. According to new reporting by Cisco, AI agents generate up to 450% more traffic per task, compared to a human performing the same task. Agents and automated workflows don't shrug and hit refresh when something fails; they retry at machine speed, which can turn a small configuration mistake into a cascading failure. Under those conditions, the quiet foundation becomes a real business risk. Slightly wrong or stale information about where services live no longer causes just a hiccup. It can cause intermittent slowdowns and timeouts, silent misrouting where systems appear to be up but make the wrong decisions, and sudden loss of trust in high-profile AI initiatives from customers, employees and regulators. What Leaders Need To Know For CEOs and boards, the implication is clear: AI resilience is now a strategic infrastructure issue, not just a data science or application issue. You can invest heavily in models, data governance and compute power, and still fail in production if the basic digital fabric that connects those pieces is fragmented, manual or treated as an afterthought. To reduce risk, make sure you take these three key steps:1. Transform your network to accelerate AI: Modernizing your network needs to be a part of your AI planning. When the work is done by AI, it’s inherently dependent on the network. If the network doesn't perform, AI won’t either. Increased levels of resilience, performance, automation and observability are no longer just nice-to-haves. 2. Demand pervasive automation: AI allows enterprises to move amazingly fast, yet many enterprises still rely on manual tickets and change windows to update the network infrastructure underneath them. That mismatch is unsustainable and will create bottlenecks in AI transformation efforts if not proactively addressed. 3. Use the foundation as a control point, not just a utility: In an AI world, the same basic services that connect your systems are also one of the few places you can see and shape how AI is actually being used. Forward-looking organizations are starting to treat this layer as a strategic control point, using it as a lever for governance and risk. As teams rush to achieve machine speed, they risk large-scale failure by leaving networking foundations out of their planning; without those solid foundations, they may as well be building their new programs on sand.Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?