Artificial intelligence is reshaping the enterprise, but much of the conversation has centered around computing power. Data centers are being expanded, graphics processing units are in high demand, and model training costs are climbing. Yet the network that connects all these resources is just as important. Without a modern, flexible, and responsive IP foundation, even the most powerful AI infrastructure cannot deliver on its promise. For organizations that want to benefit from agentic AI, network evolution is not optional. It is a strategic imperative.
The connectivity blind spot in AI planning
Agentic AI refers to systems that do more than generate text or classify data. These agents can request information, trigger business processes, collaborate with other agents, and make decisions in real time. That autonomy creates a very different kind of network demand. Instead of predictable traffic patterns tied to employees' working hours, AI agents generate continuous, unpredictable requests. The network must be able to handle bursts of activity at any moment and to route traffic intelligently between distributed cloud environments, data centers, and edge locations.
Legacy networks were designed for a world of static applications and human users. They were built to support video streams, web browsing, and enterprise applications such as email and customer relationship management. These workloads have predictable peak times and relatively tolerant latency requirements. AI workloads are different. They are often distributed across multiple clouds and require high-throughput, low-latency connectivity. If the network cannot adapt in seconds, AI agents will experience delays, fail to access the right data, or make decisions based on stale information. The cost of those failures is measured not just in poor user experience but in lost revenue and competitive position.
The end of busy-hour traffic models
Traditional network engineering has relied on the concept of busy hour, the period of peak demand during each day. Capacity was planned around that maximum. But AI agents do not follow a 9-to-5 schedule. They run around the clock, often with automated workflows that trigger data retrieval and model inference at all hours. Traffic profiles are becoming always-on, with continuous demand rather than predictable peaks and valleys. This is a fundamental change for network operators.
Always-on traffic means there is no quiet period for maintenance, optimization, or manual reconfiguration. Network operations must be automated and proactive. Real-time telemetry becomes essential because operators need to see what is happening across the network at any given moment. Static reports that summarize traffic over hours or days are no longer sufficient. By the time those reports are available, the conditions have already changed. Operators need streaming telemetry and automated analytics to identify congestion, latency, or path failures as they occur and to respond automatically.
Why legacy IP networks are too rigid
Traditional IP networks rely on protocols that are well understood but inflexible. Routing decisions are often based on shortest paths rather than application requirements. Complex architectures accumulate over time through overlay after overlay, making it difficult to deliver precise service levels for different types of AI traffic. When network architects need to change traffic paths, the process can take weeks. AI agents need network conditions to change in seconds.
Modern networking approaches address this rigidity through technologies such as segment routing and Ethernet Virtual Private Network (EVPN). Segment routing simplifies traffic engineering by encoding the path into the packet header at the source. It leverages existing IP investments while allowing operators to steer traffic based on business and application policies. EVPN provides a flexible foundation for virtualized services and multi-cloud connectivity, enabling consistent policies across distributed environments. Together, these technologies give organizations the control needed to support AI agents that move data and requests across many locations.
Segment routing and path control
Segment routing works by defining a list of segments, or instructions, that a packet should follow through the network. Operators can specify a path based on latency, bandwidth, avoidance of specific links, or other constraints. This is a major improvement over traditional traffic engineering approaches. In the past, technologies like Resource Reservation Protocol with Traffic Engineering (RSVP-TE) provided sophisticated path control but required extensive manual tunnel configuration and state management. Segment routing removes much of that overhead. State is held in the packet header, not in every router along the path. The result is a more scalable and easier-to-operate network.
For AI workloads, segment routing allows the network to adapt quickly when an AI agent needs to access a particular database or send results to another cloud. The network can calculate the best path in real time, taking into account current conditions and the specific requirements of the application. If a link becomes congested, traffic can be redirected automatically without waiting for a network engineer to reconfigure devices.
FlexAlgo for differentiated service levels
Another important capability is FlexAlgo, short for flexible algorithm. FlexAlgo allows the network to compute multiple optimal paths for different traffic classes based on constraints defined by the operator. For example, one traffic class might prioritize low latency, while another prioritizes available bandwidth. Another might optimize for resiliency, and yet another could be constrained to respect data sovereignty requirements. This is particularly valuable for AI because different AI agents and workloads often have different service-level agreements.
FlexAlgo delivers many of the traffic-engineering benefits that operators once sought with RSVP-TE, but with far less complexity. Operators define performance objectives and constraints, then the network automatically computes and maintains the appropriate paths. This enables a network to offer differentiated services on a shared infrastructure. An autonomous vehicle system might need low-latency data feeds, while a batch analytics job might care more about throughput. FlexAlgo ensures that each traffic type is matched to the path that satisfies its requirements, avoiding the limitations of static, one-size-fits-all rules.
The role of automation and telemetry
Real-time telemetry is a critical enabler of network automation. Modern routers can stream high-resolution data about traffic, queue depth, latency, and packet loss to centralized controllers or analytics platforms. Machine learning tools can detect anomalies and predict congestion before it affects AI workloads. Automated policy engines can then make adjustments, changing routing, applying quality-of-service rules, or diverting traffic to alternate paths.
This shift from reactive manual troubleshooting to proactive automated intervention is necessary for AI-scale operations. Human operators cannot react quickly enough to the scale and speed of agentic AI traffic. They need systems that monitor, analyze, and act in real time. The network becomes an active participant in ensuring AI performance, not a passive transport medium.
Security with MACsec
As networks evolve to support AI, security must evolve as well. MACsec, short for Media Access Control security, provides encryption and integrity protection at the data link layer. It ensures that traffic remains secure between network devices, protecting against eavesdropping, tampering, and man-in-the-middle attacks. In environments where AI agents exchange sensitive data across distributed networks, MACsec is an important layer of defense.
Enterprises in healthcare and finance are especially aware of these requirements. They must protect patient records, financial transactions, and proprietary models while complying with strict regulations. MACsec helps ensure that data is protected as it moves between data centers, clouds, and endpoints. When combined with segment routing and FlexAlgo, it allows organizations to build networks that are both agile and secure.
Early adoption in critical sectors
Leading organizations in healthcare and finance are already beginning to incorporate these advanced networking capabilities into their architectures. They view network modernization as part of broader digital transformation programs. These enterprises need to support a mix of AI and traditional workloads, ensuring that traffic adheres to strict policy, sovereignty, and service-level requirements. They also need to do so without disrupting existing operations.
The path to modernization can vary. Some organizations choose to deploy and manage their own IP networks over leased optical services from connectivity providers. Others prefer to consume the same capabilities through fully managed network services. Both models are viable. The choice depends on the organization's in-house expertise, operational preferences, and regulatory constraints.
Opportunities for service providers
The rise of agentic AI also creates new opportunities for service providers. Operators that can offer modern, AI-ready network services will be well positioned to deliver differentiated value. Managed segment routing, FlexAlgo, telemetry, and MACsec can be bundled into enterprise-grade connectivity offerings. Providers can offer service-level guarantees that go beyond simple bandwidth and uptime, addressing latency, path control, and security.
This is a shift from selling raw connectivity to selling outcomes. Enterprises want networks that can support AI workloads reliably and securely, without requiring them to become networking experts. Service providers that can deliver that outcome will capture new revenue. Those that continue to offer only static, high-capacity pipes risk commoditization.
Network evolution as a competitive advantage
AI presents both an opportunity and a challenge. The opportunity is to monetize the next wave of AI services by building networks that can support intelligent, autonomous applications. The challenge is that traditional networks are not up to the task. Enterprises and service providers must modernize their IP networks with segment routing, EVPN, FlexAlgo, real-time telemetry, and MACsec security. Those that do will be able to prevent connectivity bottlenecks, maintain service assurance, and scale AI adoption with confidence. Those that delay will find themselves at a disadvantage as competitors embrace network evolution and capture the benefits of the AI era.
Source: Network World News