Artificial intelligence has triggered a massive race for compute capacity, with data centers, accelerators, and model training dominating headlines. Yet the connective tissue that allows AI systems to function—the network—is often treated as an afterthought. That oversight creates a new planning challenge for organizations moving into the next phase of AI deployment. Modernizing IP networks can unlock revenue from AI-driven services, while delaying modernization risks competitive irrelevance.
Agentic AI changes the traffic profile. Instead of human-driven sessions, AI agents request data, trigger actions, and collaborate across distributed, multi-cloud environments. They operate continuously, making decisions in microseconds. The old 'busy hour' model is replaced by always-on demand. Traditional networks were built for voice, video, and general internet traffic, not for the agility, precision, and performance AI workloads require.
Connectivity: The Hidden Bottleneck in AI Deployments
Compute gets attention because it is visible and expensive. Connectivity is less visible until it becomes a constraint. When AI agents need to move data between clouds, edge locations, data centers, and users, the network determines whether those interactions feel instantaneous or sluggish. A slow path can stall inference, delay a transaction, or break an automated workflow. For organizations planning AI at scale, network modernization is not a side project. It is a prerequisite for reliable, high-performance AI services.
The problem is not simply bandwidth. It is the ability to adapt dynamically. AI agents do not follow predictable schedules. They can generate bursts of east-west traffic, query multiple data sources, and coordinate actions across regions. A network designed for steady, human-centered traffic patterns cannot keep pace with that behavior without significant operational effort.
From Busy Hour to Always-On: Agentic AI Traffic Patterns
For decades, network planners designed around the busy hour. They studied peak usage windows, provisioned capacity for the worst-case period, and optimized for predictable demand. That model worked when traffic came from voice calls, video streaming, and web browsing. Agentic AI breaks the model. AI agents are always on. They hit the network around the clock, making decisions in microseconds. There is no natural off-peak window when automated agents can wait.
This shift has profound implications. Capacity planning must account for continuous demand rather than daily peaks. Latency budgets become tighter because agents may coordinate in real time. Multi-cloud architectures add complexity because traffic crosses provider boundaries and must comply with varied policies. In this environment, static configurations and manual changes cannot deliver the responsiveness AI requires.
Real-Time Telemetry and Automated Intervention
To support AI workloads, operators need real-time telemetry. They must understand traffic patterns as they happen, not after the fact. Real-time visibility allows the network to detect congestion, path degradation, or policy violations and trigger automated intervention. Without that visibility, operators are left with reactive manual troubleshooting and static reports that are often outdated by the time they are used.
Automation is equally important. When network conditions change within seconds, human operators cannot manually reconfigure routers and policies fast enough. The network must be able to sense, decide, and act continuously. That requires a modern architecture with open interfaces, streaming telemetry, and policy-driven control. It also requires a cultural shift: network teams move from maintaining static infrastructure to operating a dynamic system that supports AI-driven business outcomes.
Segment Routing and EVPN: A Foundation for Convergence
Many legacy IP networks are bloated, rigid, and complex. They were built for earlier eras of VPN and internet connectivity, and they have served enterprises well. But they are too rigid and too complex for dynamic AI demands. Evolving to modern architectures based on segment routing and Ethernet VPN, or EVPN, provides a foundation for convergence and precise path control. This enables dynamic traffic routing as AI agents' connectivity needs change.
In the past, network architects often had weeks to make changes to support new demands. Today, network conditions must change within seconds to meet the requirements of AI agents. Segment routing helps by simplifying traffic engineering and reducing state in the core. EVPN brings scalable, flexible connectivity for data center, campus, and wide-area use cases. Together, they create a more programmable network that can support both traditional workloads and emerging AI services.
Segment routing also leverages existing network investments while creating an evolutionary path to the flexibility needed for AI workloads. Organizations do not have to rip and replace everything at once. They can introduce segment routing and EVPN in phases, protecting prior investments while building toward a more adaptive architecture.
FlexAlgo: Flexible Algorithms for Diverse SLAs
Networks also need FlexAlgo capabilities. Short for flexible algorithm, this feature lets the network calculate optimal paths for different traffic types. One class of traffic might be optimized for latency. Another might be optimized for available bandwidth. Another might prioritize resiliency. Another might need to satisfy data sovereignty requirements. FlexAlgo allows these different objectives to coexist in the same network, depending on the needs of specific workloads.
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. As networks increasingly support different service-level agreements for different AI agents and workloads, FlexAlgo ensures traffic is matched to performance requirements rather than constrained by static, one-size-fits-all rules.
This matters for AI because not all AI traffic is equal. A real-time inference request may need ultra-low latency. A batch model update may need high throughput but can tolerate some delay. A regulated data flow may need to remain within a jurisdiction. FlexAlgo gives operators a way to express those distinctions and let the network enforce them automatically.
Security, Policy, and Sovereignty: MACsec and Beyond
Security cannot be an afterthought either. AI workloads often involve sensitive data, proprietary models, and regulated information. MACsec, or Media Access Control Security, provides link-layer encryption and integrity checking that helps protect traffic as it moves across the network. When combined with policy controls and segmentation, MACsec helps ensure that AI traffic is protected without compromising performance.
Enterprises in critical sectors such as healthcare and finance have been working to incorporate these capabilities into their network architectures as part of broader digital transformation initiatives. These organizations need to support a mix of AI and traditional workloads while ensuring that traffic adheres to strict policy, sovereignty, and SLA requirements. For them, the network must be able to separate traffic types, enforce compliance boundaries, and guarantee performance for high-priority applications.
Healthcare organizations may need to protect patient data while enabling AI-assisted diagnostics. Financial institutions may need to meet regional data residency rules while supporting real-time fraud detection. In both cases, the network becomes a control point for policy, security, and performance. Modern IP architectures make that control practical rather than theoretical.
Deployment Models: Self-Managed, Leased Optical, Fully Managed
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. That flexibility creates new opportunities for providers to deliver differentiated, value-added services. Rather than selling only raw connectivity, providers can offer assured performance, automated policy enforcement, and integrated security for AI workloads.
For enterprises, the choice depends on internal skills, risk tolerance, and strategic priorities. Some want full control over routing and policy. Others prefer to focus on AI applications and let a partner operate the underlying network. Both paths are viable when the architecture is based on modern principles such as segment routing, EVPN, FlexAlgo, and MACsec.
What Modern Networks Deliver
The result is a network that automatically enforces business policies and performance objectives. It prevents connectivity bottlenecks and maintains service assurance as AI adoption and digital transformation efforts continue to scale. Instead of reacting to problems after users or agents are affected, the network can anticipate and adapt. Instead of treating all traffic the same, it can match paths to requirements. Instead of slowing down AI initiatives, it can accelerate them.
This outcome is not automatic. It requires a deliberate modernization strategy, investment in telemetry and automation, and a shift from static operations to dynamic control. But for organizations that depend on AI-driven services, the payoff can be significant: faster time to market, more reliable performance, and stronger security and compliance.
The Competitive Stakes for Service Providers and Enterprises
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. They can offer new tiers of connectivity, guaranteed SLAs for AI workloads, and managed services that simplify adoption. If they stand still, they risk being run over by competitors who embrace network evolution.
The stakes are especially high because AI adoption is accelerating across industries. Organizations that delay network modernization may find that their infrastructure limits the AI use cases they can support. They may struggle with unpredictable traffic, manual operations, and security gaps. Those that move early can build a foundation for agentic AI, multi-cloud collaboration, and real-time automation. The network becomes not just a cost center but a platform for growth.
As AI agents become more autonomous and distributed, the demand for adaptive connectivity will only increase. Networks that can change within seconds, enforce policy automatically, and optimize paths for diverse workloads will be better positioned to support the next era of digital business. The runway is being cleared now. The question is whether organizations will use it to take off or watch competitors pass them by.
Source: Network World News