How to stop AI from straining networks

Can smarter protocols make networks greener? 

As AI workloads demand scale and speed, most telecom companies continue to focus only on latency and throughput, with sustainability being more of an afterthought. Increasing bandwidth alone is not enough to handle the European AI boom: greener and smarter routing protocols and standards are also needed. Without this, AI scaling will always come with avoidable energy wastage.

IPv4’s drawbacks – from workarounds such as NAT, overloaded routing tables and address sharing – have increased routing complexity which boosts processing overhead and energy use in core networks. 

IPv6 is usually looked at as an innovative way to boost internet of things (IoT) growth and solve address depletion, however, it could also be a useful tool in slashing energy wastage from inefficient routing. This could potentially be the sustainability boost that European AI needs. 

In contrast, IPv6 promotes better multicast support and flatter routing tables, which slashes processing overhead and enhances the distribution of large AI training datasets. By boosting more localised traffic flows, IPv6 could also support regional ISPs in operating more efficiently per packet. In turn, this would help avoid some of the vast energy usage of hyperscale interconnects. 

IPv6 deployment and localised routing could make regional internet service providers less energy-intensive per packet, compared to the vast bandwidth usage and heat generation of hyperscale AI training. IPv6 offers a foundation that, when combined with emerging solutions such as intent-based networking, software-defined networking (SDN) and green routing, is helping to sustain AI workloads while optimising energy efficiency. 

IPv6 should not be solely responsible for efficiency

Yet, Martin Butler, professor of digital transformation at Vlerick Business School, warns that IPv6 is not a silver bullet for AI-related energy savings, despite having energy benefits. “IPv6’s reduced reliance on NAT and its support for simpler routing and auto-configuration can offer modest efficiency gains in large-scale, distributed AI environments, particularly where software-defined networking is used,” he says. “Still, the overall energy impact, based solely on IPv6, remains limited.”

He adds that the inherent capabilities that it provides for more control and granular routing is where the potential value lies: “IPv6 is not a silver bullet for AI-related energy challenges, yet offer architectural advantages of scalability and operational ‘clarity’ which can support performance efficiency if correctly used.”

European IPv6 adoption rates remain patchy. While the Baltics and France are leading the way, the UK and Spain, among others, are still lagging behind. Dual-stack networking, which combines both IPv4 and IPv6 address usage can also spike energy wastage in the short term, mainly because of duplicate processing. 

Historically, European telcos often treated network design as mere plumping or engineering detail. However, this could now be changing. 

“Historically, emissions reduction has not been a design parameter for routing protocols; speed and uptime have taken precedence. That is now changing,” says Sriram Panchapakesan, CEO of telecommunications, media, technology, energy and utilities at digital transformation firm Sutherland. He also believes that some operators are experimenting with “carbon-aware routing”, where AI analyses carbon intensity, real-time energy metrics and traffic loads to guide routing decisions.

“Sustainability is increasingly becoming a design consideration, particularly in the field of IP and optical convergence, traffic engineering and SDN-based routing,” adds Afzar Aslam, vice-president and chief technology officer (CTO) at Nokia Europe. 

Some EU policies such as the Green Digital Coalition are attempting to focus on sustainable networks as well. “It is now really necessary to introduce mandatory requirements for the energy efficiency of routers, datacentres and interconnects – just as it has already been done for household appliances,” Alexandr Adamenko, co-founder of Winday, points out. 

How are European companies preparing their networks for AI? 

However, despite IPv6’s potential, it can only decrease emissions at scale if telecom companies are willing to sacrifice some performance for efficiency. Major UK and EU telecom companies such as Deutsche Telekom, Orange, BT and Nokia have already warned of increasing power usage, soaring costs and heat output from AI-driven traffic.

These companies, caught between net-zero commitments and soaring AI demand, are now investing significantly in efficiency. Some measures include IPv6 adoption, AI-driven routing, greener protocols and more.

“Across Europe, telecom operators are accelerating efforts to modernise their networks to support AI workloads. Prioritising energy-efficient hardware, leveraging more advanced silicon and integrating AI-native architectures are three key strategies that are working well,” Aslam says.

“As operators are moving workloads closer to the edge to reduce latency, new transmission options like activation of additional spectrum for busier hours, or automation in capacity on demand, allows operators to deactivate network components when not in use. Many are also using automation and AI-driven network optimisation to allocate resources more dynamically based on real-time demands which is helping improve efficiency without sacrificing performance.”

Lithuania has become a breakout hub for software innovation and telecom, mainly because of its smaller market size. This makes it easier to test out edgier ideas relating to routing strategies and AI workload balancing in weeks, instead of years, in bigger markets. Companies such as Telesoftas, now Helmes, are experimenting with AI-enabled monitoring and network efficiency on a national scale before expanding further in Europe.

In Germany, Deutsche Telekom is using AI on a trial basis to dynamically reroute traffic to slash energy wastage and decrease bottlenecks. Some of its main products include AI models that predict demand surges, such as during AI model training events or large streaming, and proactively shift loads accordingly. Not only is this cost-effective, but it also takes Deutsche Telekom a step closer to its 2030 carbon-reduction goals.

France’s Orange is investing in fibre as a lower-energy alternative to legacy and copper infrastructure. Compared to copper or digital subscriber line (DSL) networks, fibre networks generate considerably less kWh per gigabit transmitted.

Orange is also working on dynamic energy-saving initiatives for idle network elements. This will let them adjust more intuitively to low-traffic periods. Similarly, Nokia is pioneering green routing, which focuses on protocols that optimise for lowest energy intensity. In the UK, BT is trialling greener data centre interconnects and AI-driven energy monitoring, while also exploring renewable integration for network nodes, especially in rural areas with higher grid emissions. 

The road ahead 

Although several companies in Europe are already preparing to better equip their networks to handle AI, challenges remain. Network efficiency needs to become a core part of AI and corporate ESG strategies, failing which companies risk exposure to new EU climate reporting rules and rising operational expenses.

Emerging solutions such as liquid-cooled interconnects, greener routing algorithms and carbon-aware workload scheduling need more funding than ever, as they transition to pilot deployments from research, if companies want to optimise efficiency gains across systems. Another key challenge is Europe’s approach to digital infrastructure and the narrow framing of sustainability itself.

“Europe’s deep regulatory instinct, while rooted in legitimate goals, risks becoming counterproductive,” says Vlerick Business School’s Butler. “An overemphasis on rulemaking without sufficient space for experimentation can inhibit the very innovation needed to build next-generation, resource-efficient infrastructure.”