Nokia’s Latency Reduction Leap for AI Data Centres

The global race to build AI-ready data centres is accelerating and network performance has become one of the most critical competitive differentiators. 

For operators and hyperscalers, the demand for low latency, scalable infrastructure capable of supporting high-bandwidth AI workloads is reshaping how networks are designed and managed.

Within its data centre networking portfolio, Nokia offers its 7220 Interconnect Router (IXR) high-performance switches and advanced Artificial Intelligence for Operations (AIOps) features for its Event-Driven Automation (EDA) platform. 

Together, these innovations target the growing need for ultra-efficient, low latency connectivity that can handle the unprecedented data movement required by large-scale AI training and inference environments.

The AI era demands low latency infrastructure

The shift towards AI-driven workloads, particularly those powering agentic and generative AI systems, has placed data centres under immense strain. 

Processing large language models (LLMs) and other AI applications requires massive parallel computing power across thousands of GPUs or XPUs. But even as processing capability grows, network latency remains a persistent bottleneck.

Latency – the delay before data is transferred – directly impacts the speed of training and inference in AI models. Reducing it is therefore essential to achieving the performance, accuracy and responsiveness modern AI systems require.

Nokia’s 7220 IXR-H6 switches were designed specifically to meet this challenge. Capable of delivering up to 102.4TBps throughput with 800 Gigabit Ethernet (GE) and 1.6 Terabit Ethernet (TE) interface speeds, they double both throughput and interface performance compared to the company’s previous models. The result is a network fabric optimised for low latency communication across large AI clusters.

These switches are compliant with Ultra Ethernet Consortium (UEC) standards, which aim to optimise Ethernet for AI and high-performance computing environments. 

The standards introduce advanced congestion control and packet management techniques to reduce latency and improve determinism – critical for synchronising workloads across massive distributed systems.

High performance with sustainable scalability

Beyond speed, scalability and energy efficiency are becoming equally important for data centre operators. Nokia’s 7220 IXR-H6 family offers both air-cooled and liquid-cooled variants, allowing operators to choose configurations best suited to their environmental and operational needs.

This flexibility is vital for hyperscalers building out AI clusters that may exceed one million XPUs. The combination of compact form factor and low-latency network architecture enables greater density within the same physical footprint – a key advantage as operators balance performance growth with sustainability goals.

The switches are also unique in supporting both Nokia’s SR Linux Network Operating System (NOS) and open-source Community SONiC, giving data centre operators freedom to choose between vendor-specific optimisation and open networking flexibility.

Tom Burke, Chief Revenue Officer at Nscale, highlighted how these innovations translate into real operational benefits: “Our partnership with Nokia continues to deliver meaningful operating advantages amid the accelerating pace of AI innovation. With Nokia’s technology, Nscale strengthens its ability to deliver for our shared customers. 

For Tom, this collaboration is “best serving the needs of customers who rely on advanced AI technology”.

AI-enabled automation for reduced latency operations

While the hardware delivers the physical foundation for low latency networking, Nokia’s automation platform addresses the equally important challenge of maintaining performance consistency in complex, fast-evolving environments.

The enhanced EDA (Event-Driven Automation) platform introduces AIOps capabilities powered by agentic AI. These features enable natural language-based interactions, intelligent diagnostics and self-healing network behaviours that identify, analyse and resolve performance issues with minimal human intervention.

EDA AIOps integrates real-time telemetry, digital twins, dry-run testing and instant roll-back functions, allowing operators to simulate changes before deployment and rapidly revert if issues arise. This combination of predictive and proactive management not only reduces operational risk but also keeps latency to an absolute minimum by preventing congestion and downtime before they occur.

According to Bell Labs Consulting and Futurum Research, data centre operators using Nokia’s EDA AIOps can achieve a 96% reduction in downtime – a crucial metric in environments where even milliseconds of delay can translate into lost efficiency and revenue.

Alan Weckel, Founder and Technology Analyst at 650 Group, underscored the significance of these capabilities: “The 1.6 TE interface speeds on Nokia’s new family of 7220 IXR switches hit a sweet spot in the market as agentic AI compels a change in data centre networking requirements. 

“Agentic AI-powered AIOps on Nokia’s EDA platform is timely, driving increased operational efficiency and reliability in a demanding network environment. Also great to see Nokia’s commitment to the UEC as 650 Group analysis projects Ethernet will be the dominant networking protocol for AI moving forward.”

Optimising AI data fabrics for low latency ecosystems

Nokia’s approach to data centre networking reflects a broader shift in how latency is being managed across the AI ecosystem. Rather than treating it as a network optimisation challenge alone, Nokia integrates low latency design principles at every layer – from physical hardware to software and automation.

Its EDA AIOps, for example, continuously analyses network behaviour in real time, detecting micro-delays that could affect training synchronisation or data flow. By dynamically rerouting traffic or rebalancing loads, it ensures that latency-sensitive applications, such as AI model training or inference, operate at peak efficiency.

The company’s alignment with the UEC and its focus on open standards further position it to shape the next generation of Ethernet-based networking for AI. As AI workloads increasingly rely on east-west data movement between accelerators, reducing microseconds of delay per hop can deliver exponential performance improvements at scale.

Vach Kompella, Senior Vice President and General Manager of Nokia’s IP Networks Division, emphasised the scale of transformation underway: “The astonishing growth in AI adoption has led to a dramatic overhaul in how data centres operate, and is driving constant evolution in hardware and operational tools. 

“We are pleased to announce our new family of high-performance 7220 IXR-H6 switches along with advances in our EDA platform, which now leverages the sophistication of agentic AI to drive highly reliable network operations at the speed required to keep up with the pace of change.” 

According to Vach, this represents “one of many milestones for Nokia’s data centre networking portfolio, and we look forward to bringing even more technology innovation to our global customer base to support the AI super-cycle”.

Supporting the AI super-cycle

As AI models continue to grow in size and complexity, the ability to maintain ultra-low latency connectivity will define the next generation of data centre infrastructure. Every layer – from fibre routing and switching to automation and software orchestration – must operate with millisecond precision to support synchronous, large-scale AI workloads.

Nokia’s new switches and automation tools are designed not only to handle this complexity but to make it manageable, predictable and energy efficient. The combination of 1.6 TE interface speeds, open NOS options and intelligent AIOps aligns with the demands of the AI super-cycle, where speed, scale and sustainability must coexist.

By blending innovation in both hardware and AI-driven software, Nokia is positioning itself as a key enabler of the low latency networks that underpin global AI growth. Its technology helps data centres move beyond conventional architectures toward intelligent, adaptive environments built for the future of AI.

As the boundaries of data processing and model training expand, low latency infrastructure will remain at the heart of high-performance computing. Nokia’s latest developments mark an important step forward, setting a new benchmark for how speed, scalability and intelligence can come together to power the next era of AI-driven digital infrastructure.