Connectivity is now an AI performance issue

Most enterprises already see the signs. AI is moving into core workflows, demands are increasing and systems need to respond in real time. What is becoming clearer is that performance no longer depends only on models, data or compute. It depends just as much on how reliably everything is connected.

When connectivity fails, the impact is immediate. Transactions stop, services go offline and productivity drops. In an AI-driven environment, that effect becomes sharper and more direct, as it is not just processes that are interrupted, but the flow of decisions and actions the business relies on.

Recent data highlights how fundamental this shift is. According to Cisco’s report AI Impact on Wide Area Networks (2026), AI-driven workloads can generate up to 450 percent more traffic per task than traditional activity, with around 70 percent tied to inference. In practice, that means continuous, real-time interaction between models, data and applications becomes critical to how services perform.

At the same time, growth is accelerating quickly. Token consumption is increasing close to tenfold year over year, and while AI traffic is still a smaller share today, Cisco projects it could reach up to a quarter of all network traffic by 2035. For enterprises, this is a structural shift that directly affects how networks need to perform.

A new execution layer across distributed environments

Enterprise environments are already distributed, but AI makes that dependency far more visible. Cloud regions, private data centers and external platforms are now part of the same execution path, often within a single workflow.

AI workloads move continuously between systems, with agents interacting with models, data sources and tools across environments. Cisco describes this connection as a “spinal cord”, which captures the shift well. Connectivity becomes the layer that ties intelligence, data and execution together, and when performance drops, the impact is immediate.

Latency, congestion or packet loss no longer stay local. They propagate across tasks and show up directly in application performance. What used to be a network issue becomes a limit on how AI services perform.

Cisco estimates enterprise traffic could grow about 2.5 times without AI, but up to nine times with widespread adoption. That difference alone requires a rethink of architecture and resilience.

From redundancy on paper to resilience in practice

Many organizations still rely on an outdated view of resilience. Having multiple connections is often seen as sufficient. In practice, those connections often share the same physical routes or exposure to risk, meaning what looks robust on paper can behave like a single point of failure.

In an AI-driven environment, that gap becomes harder to ignore. Traffic is continuous and latency-sensitive, so even small disruptions can impact performance. A network that is technically available but unstable can still degrade AI-driven services.

Resilience is therefore a question of design rather than scale. Traffic needs to move across independent paths, with seamless failover and stable performance under load. It is less about adding capacity and more about predictable behavior in real conditions.

For enterprises, the implication is clear. Connectivity must be treated as core architecture, aligned with AI, cloud and security. And as AI adoption accelerates, it becomes a defining capability that enables AI to operate at full speed.

The real question is simple: is the network built for how AI behaves, or for how systems used to work?

AI will amplify every delay and weak point. Those who design for that will see it in performance. Those who do not will see it where it matters most.

Johan Ottosson
Head of Strategy