Do networks really need AI – or is it just marketing smoke?

Artificial intelligence has become telecom’s favorite tagline. Every conference booth, whitepaper and vendor demo promises “AI-powered” capabilities that will transform networks as we know them. Yet behind this enthusiasm lies a simple, often overlooked question: do networks actually need AI?

The truth is uncomfortable. Most don’t. What modern networks truly require are architectures that are clear, efficient and intentionally designed not systems burdened by unnecessary computational complexity. Intelligence in networking is not measured by the number of models running in the background, but by how effectively a network anticipates issues, adapts to change and sustains performance without excess overhead.

The illusion of intelligence

For years, telecom has blurred the line between automation and intelligence. A rule-based provisioning workflow is renamed “AI-driven orchestration.” A basic threshold alarm becomes “AI-powered anomaly detection.” A decision tree wrapped inside a neural network suddenly qualifies as “predictive maintenance.”

But intelligence is defined by outcomes, not labels. If a system performs the same function it always has but now requires more compute resources, more cooling infrastructure and more operational complexity what the industry calls “intelligence” is actually inefficiency rebranded as innovation.

Behind every AI claim lies a real physical footprint. Model training and inference cycles rely on GPU-heavy environments that demand substantial electricity, cooling and water resources. These hidden costs accumulate across networks running multiple overlapping AI models, many of which provide marginal or redundant value.

In this context, what is marketed as progress can easily become regression. Misapplied AI increases environmental impact, strains infrastructure and adds layers of opacity without delivering proportional benefits.

Where AI truly belongs

Despite the overuse of the acronym, there are scenarios in which AI meaningfully transforms network operations.

Large-scale networks generate immense volumes of telemetry far beyond what threshold-based systems can effectively interpret. Degradation patterns that precede failures such as fiber cuts or router instability are often subtle and non-linear. AI models can detect these signatures long before humans or traditional tools can, enabling operators to prevent outages rather than react to them.

Used intentionally, AI improves routing decisions under dynamic conditions, reallocates resources across multi-layer environments and identifies anomalies that would otherwise remain invisible. In such cases, AI becomes a quiet but powerful assistant supporting engineers, enhancing operational reliability and helping maintain service continuity.

These applications, however, represent specific, high-impact use cases, not universal requirements. AI is most effective when applied with precision, not ubiquity. Networks do not need AI everywhere; they need it in the moments and domains where complexity exceeds human or rule-based capabilities.

The sustainability paradox

Much of AI’s environmental impact remains hidden from dashboards and KPIs. Training and running large-scale models requires significant compute power, but the water footprint attached to cooling these environments is equally crucial and often overlooked.

A 2023 study titled “Making AI Less ‘Thirsty’: Uncovering and Addressing the Secret Water Footprint of AI Models” estimates that training a single large-scale AI model can consume hundreds of thousands of liters of freshwater. According to the analysis, the training of GPT-3 alone may have required roughly 700,000 liters, equivalent to the annual drinking water needs of more than a thousand people.

When industries deploy AI models without assessing actual necessity, these environmental costs multiply. Telecom positions itself as a driver of green and sustainable digital transformation, yet excessive or poorly justified AI deployments can directly contradict those goals.

Sustainability, too, is a form of intelligence: achieving better outcomes with lower resource consumption. Effective design minimizes the need for heavy computation; it does not rely on it by default.

A more intentional path forward

AI delivers value when applied with clear purpose. It can shift networks from reactive systems to proactive, self-optimizing environments. It can reduce downtime, improve customer experience and enhance resource utilization when the use case warrants it.

The challenge lies in resisting the urge to apply AI indiscriminately. Misuse leads to higher operational costs, increased system opacity, longer development cycles and amplified environmental impact. The distinction between meaningful innovation and marketing-driven adoption is determined by intent.

Telecom’s future does not depend on deploying more AI, it depends on deploying the right AI. Thoughtful design, transparent modelling, sustainability considerations and well-defined use cases must guide the next phase of network evolution.

Conclusion

The question is not whether AI belongs in network operations. It clearly does in the areas where scale, complexity and variability exceed human and deterministic capabilities. The real question is whether every process, every tool and every workflow benefits from AI. In many cases, the answer is no.

Progress in telecom will be defined not by owning the largest models or the most AI-branded features, but by using intelligence responsibly. Networks will advance fastest and most sustainably when AI is applied with intention, precision and measurable impact.