AI ambitions are exposing the limits of enterprise networks built for the cloud era. To support autonomous workloads at scale, enterprises must evolve from static connectivity to intelligent, self-managing agentic infrastructure that predicts, adapts and optimises in real time – writes Viswanathan Ramaswamy at Tata Communications.
Today’s enterprise leaders all have an AI strategy, but most are attempting to run it on networks built for the cloud, not optimized for AI. A recent Bloomberg report commissioned by Tata Communications revealed that 77% of leaders consider AI a board-level priority, and yet only 35% are leveraging new network infrastructures that meet AI’s performance demands.
While AI is a board-level priority, a staggering 94% of IT leaders identify the network as the biggest barrier to success, according to data from IDC. That disconnect is reflected in McKinsey’s findings that only 5% of enterprises are truly ‘AI-ready’ at scale, with the vast majority stuck between pilots and partial deployment. The enterprises using transitional or legacy networks are experiencing the ‘tech debt tension zone’.
The tech debt tension zone, the gap between how aggressively enterprises are deploying AI and the readiness of their underlying infrastructure to support it at scale, is widening because enterprise leaders are continuously putting massive AI workloads on networks that were not built for this type of traffic.
The concern is that traditional business applications could tolerate 100-500ms of latency. Mission-critical AI workloads require sub-10ms latency. This isn’t an incremental improvement; it’s a completely different performance paradigm that breaks traditional network design, and it helps explain why IDC analysts find that 88% of AI projects are scrapped before ever reaching production.
Connectivity vs capability
Decision makers have spent the last decade focused on connectivity: laying fiber and developing 6G, but the next decade is about capability: what the network can do autonomously. AI introduces a massive shift from predictable, user-to-application traffic to constant machine-to-machine communication across cloud, edge, and data centers. Tata Communications found that this east-west traffic increases network loads by 50x to 100x, demanding infrastructure that can handle continuous, high-frequency data flows.
To realize a true operational return on AI investment, the infrastructure must become ‘agentic’ or capable of independent decision-making. Traditional automation follows strict rules that use an ‘if/then’ format. Agentic infrastructure, on the other hand, uses AI to understand intent, forecast traffic patterns, and re-route capacity without human intervention.
Predicting the unpredictable
This is the difference between a simple road and a fully autonomous, intelligent transport system. Imagine a multi-lane digital highway with an AI-powered traffic control system. A mission-critical AI workload, like a real-time medical diagnostic, can be dynamically assigned its own dedicated, high-speed, low-latency lane. If the system predicts a disruption on that route, it instantly and automatically reroutes the traffic to another clear path, ensuring performance is unaffected.
Leveraging these capabilities, AI models can analyze historical data, patterns, and application usage to forecast traffic spikes. What was previously viewed simply as increased network usage during standard office hours (M-F 9-5) is now being synthesized into a more detailed timeline. For example, a user may opt to use their mornings to conduct heavy research leveraging LLMs requiring increased bandwidth and their evenings sending emails. The agentic AI anticipates this usage pattern and drives traffic to optimize the network accordingly.
These optimizations allow for dynamic provisioning, enabling the network to automatically procure capacity or shift workloads from the edge network to the core to accommodate this data use. The shift from reactive troubleshooting to proactive self-healing minimizes outages before the user even notices, optimizing the customer experience.
The human element
The use of agentic AI does not replace the need for network engineers. In fact, it elevates their role. This represents a fundamental operational shift from being reactive ‘firefighters’ to proactive ‘architects’.
Engineers will no longer be asked to manage Command Line Interface alerts. Instead of manually re-routing traffic during an outage, the team’s role is now to define the rules, policies, and business outcomes for the intelligent fabric. The network itself then executes those policies automatically and autonomously.
This upskilling will require leaders to look at the talent gap and solve for the future of enterprise technology needs. Amidst industry talent shortages, this AI orchestration helps overburdened NetOps teams punch above their weight class, allowing them to focus on high-value strategic oversight instead of manual tasks.
The ROI of an agentic future
The total economic value from generative AI is forecasted by McKinsey to be between $2.6 and $4.4 trillion annually. The companies that win the AI race won’t be the ones with the best algorithms, but the ones with the underlying infrastructure capable of unleashing them.
CIOs should reframe their thinking: the network is not a cost center, but an insurance policy for their AI investment portfolio.
An intelligent, agentic network de-risks these massive investments by eliminating the need to overprovision, enhancing security, and delivering a flexible foundation that can adapt to the next wave of AI demands without a complete architectural overhaul.
