Network Monitoring for the AI Era

AI workloads are exposing the network’s limits

AI didn’t just add more traffic. It added a different kind of traffic—one with almost no tolerance for failure. Neo-cloud providers running GPU clusters are now selling 100% SLA commitments to customers, because every millisecond of unavailability translates directly into lost compute revenue. Autonomous vehicles, AI-assisted manufacturing, and other real-time applications raise the stakes even higher. Capacity, reliability, and ultra-low latency—historically a “pick two” proposition—are now being demanded simultaneously, with no room for compromise.

From passive monitoring to network intelligence

As workloads grow more demanding, so does the complexity of managing them. Gilmore argues that the industry’s old model—observing a network through dashboards and alerts—is no longer sufficient. What’s needed instead is genuine network intelligence: synthesizing telemetry, traffic patterns, and historical data into insights that are actually actionable.

The analogy she reaches for is business intelligence. Companies have had CRMs and BI tools for decades to make sense of customer data. Networks, despite generating enormous volumes of their own data, have historically lacked the equivalent. A port going down in Kuala Lumpur that quietly cascades into congestion at a Singapore peering point is exactly the kind of multi-variable problem a human NOC engineer can’t diagnose fast enough—but an intelligent system can. As experienced network engineers retire and newer operators rely more heavily on tooling, that gap becomes even more critical to close.

The economics are flipping

One of the episode’s more surprising observations is what’s happening to pricing. For the entire modern era of networking, prices moved in one direction: down. Gilmore says that’s starting to reverse. In routes and regions where AI demand is outpacing available capacity, prices are actually going up, and operators who locked in contracts at old rates are already regretting it. The fundamental supply-and-demand dynamics of the industry are shifting in ways few predicted even two or three years ago.

Physics is the ceiling—optimization is the floor

No amount of software can change the speed of light. As data centers get pushed further from population centers—forced there by power availability and community resistance—the distance increases between where AI processing happens and where inference is needed. The network has to absorb that gap.

Gilmore’s conclusion is that this constraint ultimately points toward a single destination: the fully autonomous, self-healing network. One that detects problems, reroutes traffic, and provisions capacity without waiting for a human decision. We’re not there yet—autonomous networks require visibility not just into your own infrastructure, but into the surrounding networks you interact with. But that’s the direction everything is pointing, and network intelligence is the foundation it has to be built on.