Optical fibre: the foundation of AI-ready data centres

As artificial intelligence accelerates across industries, it is fundamentally forcing business leaders to rethink how to build and operate data centres. In Australia, hyperscalers and technology giants are accelerating their expansion plans, with the local data centre market expected to double to US$40 billion within four years, according to CBRE.

Meeting AI’s rapid growth is not only about increasing capacity. It’s about solving the real-world and unique problems that the technology presents. In this new era, data centres must support massive compute power, faster builds and higher-density environments — all underpinned by infrastructure that can keep pace with AI’s growth.

And optical fibre is at the heart of that critical infrastructure.

The challenge of AI in data centres

AI is spreading at a pace similar to the early days of the internet — only much faster. What took decades to build for web-scale applications is now happening in just a few years with AI. This is creating unprecedented demands on data centre infrastructure.

To keep pace, data centre design is evolving rapidly. High-performance graphics processing units (GPUs), essential to AI workloads, require more connections between servers than ever before. At the same time, in traditional CPU-based data centre designs, power and cooling limitations often mean fewer servers or GPU nodes can be accommodated in a cabinet. The result is a sharp increase in the amount of cabling needed to connect everything — far more than in traditional data centres. Each server needs high-speed links to switches, storage systems and management tools, which puts enormous pressure on the network.

To put this into perspective, NVIDIA’s DGX SuperPOD, a leading example of AI infrastructure, contains 32 GPU servers connected to 18 switches in a single row. That setup alone requires 384 400 GbE fibre links just to handle data movement across the cluster, not including additional connections for storage and management. It’s a remarkable increase in the volume of fibre cabling required inside the data hall, and a clear example of why traditional network designs are no longer sufficient.

This increased density also demands new approaches to energy management and cooling to handle higher power consumption and heat output, while supporting continuous, heavy internal data traffic across highly interconnected systems. At the same time, labour shortages — both in large-scale data centre construction and in day-to-day operations — are slowing deployments and making it harder to maintain these increasingly complex environments.

Together, these pressures mean data centres can no longer rely on traditional design or incremental upgrades. Scaling AI effectively requires a complete rethink of how data moves, how power is delivered and how heat is managed — while ensuring speed, reliability and scalability.

This is where the network becomes critical. At the centre of that network is fibre innovation — providing the essential foundation for AI-ready data centres.

Make data centres AI-ready with fibre

For data centre operators, AI is reshaping the entire foundation of their facilities. The traditional focus on compute power is no longer enough: success depends on the network that connects everything, ensuring vast amounts of data can move quickly, reliably and without interruption.

AI workloads depend on high-speed, low latency connections between GPUs working together, as well as to scale out the network, encompassing multiple thousands of GPUs in the rapidly emerging AI factories. Without the right infrastructure design, bottlenecks slow performance, costs rise and scalability stalls.

Meeting these demands requires fibre networks to evolve. Today’s data centres need denser, higher-capacity fibre that can handle massive data flows while fitting into the same physical footprint. At the same time, with skilled labour in short supply and the pressure to accelerate deployments, fibre needs to be faster and simpler to install. That’s driving a move towards pre-terminated fibre solutions: plug-and-play systems that streamline builds and reduce downtime.

Staying ahead means understanding where the market is going. Increasingly, data centre operators deploying large AI clusters are shifting from point-to-point cabling to more scalable structured cabling systems. The choice between single mode and multimode fibre remains key, based on the size and design of the facility, while InfiniBand and Ethernet continue to play critical roles in managing different types of AI traffic.

At the same time, the industry is pushing towards even greater speeds and capacity, to support next-generation performance — moving from 400 Gb/s to 800 Gb/s, 1.6 Tb/s and beyond, using 8- and 12-fibre connectivity, and with industry developments for higher speeds pointing towards 16-fibre connectivity in the future.

Looking ahead, fibre is not just part of the solution — it is the foundation for what comes next. The data centres that lead in the AI era will be those that can move more data at higher speeds, maximise, and scale seamlessly to support future AI technologies and architectures.

As AI continues to redefine industries, a high-performance optical fibre foundation is the critical infrastructure that will ensure data centres are built not just for today’s demands, but for the future of AI-powered innovation.

Image credit: iStock.com/adventtr