AI is forcing a re-evaluation of the network foundations that have underpinned enterprise IT for decades, according to a recent briefing from Tata Communications.
Legacy networks can’t keep up with AI traffic
Continuous inference, agent-to-agent messaging and real-time data pipelines generate traffic that is both constant and unpredictable. Traditional architectures, built for static workloads, lack the flexibility to handle these demands. A study cited in the briefing notes that 80% of executives believe their company’s competitive survival hinges on agentic AI, while consumer use of AI tools continues to accelerate.
Kapil, Vice President of Global Network Services at Tata Communications, says the network has become a critical control layer. It’s a completely different performance paradigm that breaks traditional network design assumptions, where such extreme low latency was never a primary consideration.
Legacy systems typically tolerate latency of 100–500 ms, but mission-critical AI workloads now require under 10 ms. The gap between what old infrastructure can deliver and what AI needs creates a direct cost and reliability risk.
Related: E-hailing Drivers Face New Local Regulations
Performance and cost become inseparable
When a network is treated as a best-effort transport, organizations may only discover performance shortfalls after a rollout. A model for real-time fraud detection, for example, becomes ineffective if data is delayed by network congestion. Kapil warns that each millisecond of delay can translate into a tangible financial impact.
Public internet links often appear adequate within a single country, but crossing borders or reaching international cloud platforms introduces latency and control issues. As AI components spread across cloud, edge and on-premises environments, high-frequency east‑west traffic between GPUs can become a bottleneck.
AI-driven malicious bots now account for roughly 37% of online traffic, complicating the task of distinguishing legitimate users from automated threats. Enterprises that rely on fragmented security tools often face inconsistent protection and limited visibility, making it harder to ensure security check legitimacy.
In this new environment, the network must shift from a passive conduit to an active, intelligent platform. That change means providing real-time observability of AI traffic and the ability to steer workloads along the most efficient, secure paths. The result is a move from vague “high performance” goals to concrete service‑level commitments, such as keeping latency below 10 ms for a specific workload 99.999% of the time.
