Key Facts

• Headline: Network Evolution for the Agentic AI Era.
• AI compute receives most attention, but connectivity is a critical enabler and potential bottleneck.
• Agentic AI shifts traffic from predictable busy-hour peaks to always-on, continuous demand.
• AI agents request data, trigger actions, and collaborate across distributed multi-cloud environments.
• Traditional IP networks were built for voice, video, and general internet traffic, not machine-speed AI workloads.
• Real-time telemetry is required for visibility, automated intervention, and proactive operations.
• Segment routing and EVPN replace bloated, rigid architectures with convergence and precise path control.
• FlexAlgo calculates optimal paths for different traffic types, such as latency, bandwidth, resiliency, or sovereignty.
• MACsec security helps enforce policy, sovereignty, and SLA requirements.
• Enterprises in healthcare and finance are modernizing to support mixed AI and traditional workloads.
• Deployment options include self-managed IP networks over leased optical services or fully managed network services.
• Organizations that modernize can unlock AI-driven revenue; those that delay risk losing competitive relevance.

The Connectivity Gap in the AI Race

For years, the artificial intelligence conversation has centered on compute: GPUs, accelerators, memory bandwidth, power, cooling, and data center capacity. That focus is understandable. Training and inference require enormous processing power. Yet as AI systems move from experimentation to production, a less visible constraint is emerging: the network. Connectivity is no longer a passive backdrop. It is the circulatory system that determines whether AI applications can access data, coordinate across locations, and respond quickly enough to be useful.

Agentic AI raises the stakes. Unlike a simple chatbot that waits for a human prompt, an AI agent can autonomously pursue goals, call tools, query databases, invoke APIs, and interact with other agents. These actions can happen in rapid succession, across public clouds, private data centers, edge locations, and partner environments. The result is a traffic profile that looks nothing like the internet of the past. The old model of a busy hour, with predictable peaks and valleys, is giving way to always-on demand. Agents do not sleep. They do not wait for business hours. They hit the network around the clock, making decisions in microseconds.

Traditional networks were designed for voice, video, and general internet traffic. They were built to be reliable and scalable for human usage patterns. They were not built for machine-speed coordination, dynamic service chaining, or continuous policy enforcement. That mismatch creates risk. If the network cannot adapt, AI initiatives stall. If it can, new revenue opportunities emerge.

Why Agentic AI Changes Network Requirements

Agentic AI is not just more traffic. It is different traffic. An AI agent may need to retrieve a customer record, check a fraud model, call a payment API, and notify a human operator, all within a single workflow. Each step may involve a different service, a different cloud, and a different set of performance and compliance requirements. The network must provide consistent connectivity while respecting policy, security, and data residency rules.

At the same time, the pace of change is accelerating. In the past, network architects often had weeks to plan and implement changes for new demands. Today, network conditions may need to change within seconds to meet the requirements of AI agents. A path that was optimal a moment ago may become congested, degraded, or non-compliant. Static configurations and manual troubleshooting cannot keep up. The network must become dynamic, observable, and automated.

This shift has implications for every layer of the infrastructure. Compute and storage get the headlines, but the network determines how quickly data moves, how reliably services interact, and how well policies are enforced. Without a modern network, AI applications may be constrained by latency, packet loss, security gaps, or operational complexity. With a modern network, organizations can turn connectivity into a competitive advantage.

Real-Time Telemetry: Seeing the Network at Machine Speed

The first requirement is visibility. AI workloads depend on real-time telemetry to help operators understand traffic patterns and support automated intervention. Without this real-time information, operators are left trying to support AI workloads through reactive manual troubleshooting, relying on static reports that are often outdated by the time they are used. That approach may have been acceptable when traffic patterns were predictable and changes were infrequent. It is not acceptable when AI agents are making decisions in microseconds and demanding continuous service.

Real-time telemetry provides a live view of network conditions. It can reveal congestion, latency, jitter, packet loss, and path changes as they happen. It can also feed automation systems that adjust routing, prioritize traffic, or isolate problems before they affect users. In an agentic AI environment, telemetry is not just an operational convenience. It is a foundational input for policy enforcement, service assurance, and dynamic optimization.

Telemetry also supports a shift from reactive to proactive operations. Instead of waiting for an alert or a customer complaint, network teams can use streaming data and analytics to anticipate issues. They can identify trends, simulate changes, and apply remedies automatically. This reduces downtime, improves performance, and frees human experts to focus on higher-value tasks.

Segment Routing and EVPN: A Modern Foundation

The second requirement is a modern architecture. Evolving from a bloated, rigid, and complex IP architecture to one based on segment routing and EVPN is necessary to provide a foundation for convergence and precise path control. This enables dynamic traffic routing as AI agents' connectivity needs change. Legacy IP networks and traditional protocols served enterprises well throughout earlier eras of VPN and internet connectivity. However, they are often too rigid and too complex for dynamic AI demands.

Segment routing offers a different approach. It leverages existing network investments while creating an evolutionary path to the flexibility needed for AI workloads. Instead of relying on extensive hop-by-hop state or complex tunnel meshes, segment routing can encode a path in the packet header. This simplifies traffic engineering and makes the network more programmable. EVPN complements segment routing by providing a scalable, standards-based control plane for Ethernet services, including multi-tenancy, mobility, and interconnectivity across data centers and clouds.

Together, segment routing and EVPN can support convergence: the ability to carry multiple services and traffic types over a common infrastructure. They can also provide precise path control, allowing the network to steer traffic according to business intent. For AI agents that need to reach specific data sources, avoid certain regions, or meet strict latency targets, this level of control is essential. It transforms the network from a static pipe into a dynamic platform.

FlexAlgo: Matching Paths to Performance Objectives

The third requirement is FlexAlgo. Short for flexible algorithm, this feature lets the network calculate optimal paths for different traffic types. For example, one class of traffic might be optimized for latency, another for available bandwidth, another for resiliency, and another to satisfy data sovereignty requirements, depending on the needs of specific workloads. Instead of forcing all traffic through the same rigid rules, FlexAlgo allows the network to match paths to performance objectives.

In many ways, FlexAlgo delivers the traffic-engineering benefits that operators once sought with RSVP-TE, but without the massive complexity. RSVP-TE relied on manually engineered tunnels and extensive state management. FlexAlgo allows operators to define performance objectives and constraints, then lets the network automatically compute and maintain the appropriate paths. This reduces operational overhead and improves scalability. As networks increasingly support different SLAs for different AI agents and workloads, FlexAlgo ensures traffic is matched to performance requirements rather than constrained by static, one-size-fits-all rules.

For example, an AI agent performing real-time fraud detection may require ultra-low latency and high resiliency. A batch analytics job may prioritize bandwidth and cost efficiency. A healthcare application may need to keep data within a specific jurisdiction. FlexAlgo can help the network satisfy these diverse requirements simultaneously, without requiring separate physical networks or manual tunnel engineering.

Security, Policy, and Data Sovereignty

Security is another critical dimension. AI workloads often involve sensitive data, including personal information, financial records, and health data. Organizations must ensure that traffic adheres to strict policy, sovereignty, and SLA requirements. MACsec security can help protect data in transit at the link layer, providing encryption and integrity verification. When combined with segment routing, EVPN, and FlexAlgo, MACsec can support a zero-trust approach that extends across the network.

Policy enforcement must also be automated. Manual configuration of access controls, segmentation, and routing policies cannot keep pace with agentic AI. The network should be able to enforce business policies and performance objectives continuously. This includes ensuring that certain traffic never leaves a jurisdiction, that high-priority traffic receives preferential treatment, and that suspicious activity is isolated. Automation reduces the risk of human error and helps maintain compliance as AI adoption scales.

Healthcare, Finance, and Mixed Workloads

Recent engagements in critical sectors such as healthcare and finance illustrate the challenge. These organizations need to support a mix of AI and traditional workloads while ensuring that traffic adheres to strict policy, sovereignty, and SLA requirements. They cannot simply rip and replace their existing infrastructure. They need an evolutionary path that protects current investments while enabling new capabilities.

In healthcare, AI may be used for diagnostic imaging, patient triage, or clinical decision support. These applications may involve sensitive patient data and strict privacy regulations. In finance, AI may power fraud detection, risk modeling, algorithmic trading, and customer service. These workloads demand low latency, high availability, and strong security. Both sectors require networks that can segment traffic, prioritize critical flows, and provide auditability.

The ability to support mixed workloads is essential. Most organizations will not move everything to AI at once. They will run traditional applications alongside new AI services for years. The network must provide a common foundation that can handle both, without compromising performance or compliance. This is where convergence, segmentation, and automation become practical necessities.

Deployment Models and Monetization

Depending on their operational model, organizations can deploy and manage their own IP networks over leased optical services from providers. Alternatively, they can consume the same capabilities through a fully managed network service. This creates new opportunities for providers to deliver differentiated, value-added services. Service providers can offer network slices, guaranteed SLAs, and policy-based connectivity tailored to AI workloads. Enterprises can choose the model that best fits their skills, resources, and strategic priorities.

For service providers, the opportunity is significant. AI-driven services can generate new revenue streams, but only if the underlying network can support them. Providers that modernize their IP networks can offer premium connectivity, dynamic bandwidth, and assured performance. They can also help enterprises meet sovereignty and compliance requirements through managed services. Those that delay may find themselves unable to compete for AI workloads.

Competitive Urgency

AI creates both an opportunity and a challenge for service providers and large enterprises. If they modernize their IP networks, they can monetize the next wave of AI services. But if they stand still, they risk being run over by competitors who embrace network evolution. The network is no longer a commodity background utility. It is a strategic asset that determines how quickly AI can be adopted, how safely it can operate, and how much value it can create.

The result is a network that automatically enforces business policies and performance objectives, preventing connectivity bottlenecks and maintaining service assurance as AI adoption and digital transformation efforts continue to scale. Organizations that treat connectivity as a first-class part of their AI strategy will be better positioned to capture the benefits of agentic AI.

Source:Network World News