Scalability, Density, Modularity: Requirements for Tomorrow’s Fibre Optic Infrastructure

Increasing power densities, AI applications and modular expansion concepts are fundamentally changing the requirements for data centres. Attention is now turning to an area that has long been taken for granted: the physical fibre optic infrastructure. Ahead of his panel discussion at the eco Datacenter Plaza at fiberdays 26, Matthias Nitschke, Director of Data Center Solutions at Corning, talks to eco about the role fibre architectures play in scalability and future-proofing, how planning and collaboration are changing, and what aspects decision-makers should consider today in order to create sustainable data center infrastructures for the long term.

Data centres are becoming increasingly powerful and dense. What role does the physical fibre optic infrastructure play in this?

Fibre optics are now the central nervous system of every data centre and are no longer a passive accessory. There’s a real ‘space paradox’ going on in racks: ever more powerful servers and GPUs are driving up demand for power supply and cooling, while bandwidth requirements are exploding. In theory, both would require significantly more cabling – further limiting available space and critical airflow.

This is precisely where the physical fibre optic infrastructure becomes an active enabler of power density, scaling speed and investment security in modern data centres – comparable to power and cooling. Today, it needs to be rethought on several levels. At Corning, we are meeting this challenge with technological innovations at the material and system level: through multicore fibres and extremely miniaturised connectors that enable massive densification without blocking the airflow that is crucial for cooling.

Multicore fibres in particular are the next evolutionary step in counteracting the sharp increase in port and fibre numbers and the resulting cable requirements through parallelisation at the fibre level. For data centre interconnect (DCI) in metropolitan area networks, new approaches such as hollowcore fibres are opening up new perspectives with greater spatial flexibility in connection and location selection.

AI applications and high-performance computing are changing internal traffic flows in data centres. What impact does this have on the planning of cabling and fibre architectures?

AI and HPC workloads are fundamentally changing internal traffic flows in data centres. Data traffic is shifting massively from classic north-south patterns to latency-critical east-west traffic between GPUs, compute and storage clusters, which need to communicate with each other permanently and with virtually no delay. The challenge is no longer just sheer bandwidth, but increasingly the energy efficiency of data transmission.

With each generation, port density and bandwidth increase almost exponentially – and with them the energy requirements per port. Cabling and fibre architectures must therefore be consistently designed for high-density, symmetrical and cluster-optimised backend networks. The physical infrastructure thus becomes a central lever for the performance, scalability and sustainability of AI-enabled data centres.

A key starting point is to bring the optics closer to the chip. Concepts such as co-packaged optics (CPO) make it possible to drastically reduce the energy requirements for optical signal conversion while significantly increasing the number of ports. This not only reduces the energy consumption of individual servers, but also has a positive effect on the overall efficiency of the data centre – an essential step towards ensuring the long-term performance and sustainability of AI infrastructures.

Many data centres are being expanded in a modular fashion. What challenges does this pose for a consistent and future-proof fibre optic infrastructure?

Modular growth of data centres means one thing above all else: speed combined with long-term consistency. Operators cannot afford to wait months for materials when expanding or carry out time-consuming splicing work in the server room, especially since expansion phases are staggered and technological conditions are constantly evolving.

Without a standardised setup for fibre types, connector faces and patch concepts, scalability and subsequent upgrades are severely limited.

The key therefore lies in coordinated, pre-assembled plug-and-play fibre optic solutions based on universal multi-fibre connector systems and modularly scalable patch centres (ODF). They enable fast installation, reproducible quality and reduce dependence on scarce splicing resources on site – a decisive factor for time-to-market and operational reliability.

At the same time, delivery reliability has become a key success factor. With a consistent local-for-local strategy – research and development in Berlin Adlershof and fibre optic, cable and system production in Poland – Corning ensures that modular expansions can be implemented just in time, in high quality and with a minimised carbon footprint.

From a system supplier’s perspective, pre-assembled fibre optic systems are therefore a key lever for expanding modular data centres quickly, efficiently and consistently over the long term.

How is the collaboration between manufacturers, planners and operators changing as data centres are increasingly seen as long-term infrastructure projects?

Collaboration in the data centre environment is evolving away from traditional supplier relationships towards early, deeply integrated innovation ecosystems. A data centre for the year 2030 can no longer be planned in silos. System architecture, technology and space requirements; servers, switches, cooling and fibre optic infrastructure must be thought out and decided together from day one, as development cycles become shorter and technological requirements continue to increase across all trades.

Manufacturers are becoming more involved in roadmap coordination, innovation partnerships and technology forecasting, while planners are increasingly taking on the role of integrative system architects instead of implementing individual trades in isolation. In return, operators benefit from greater planning reliability, lower life-cycle costs and a more resilient, future-proof infrastructure.

At the same time, sustainability and Scope 3 emissions are becoming more of a focus. As manufacturers, we must provide transparent data on life cycles and local supply chains – for example, via a European value chain – early on in the planning phase so that operators can not only meet their ESG targets in the short term, but also secure them in the long term.

With a view to fiberdays 26 and the eco Data Center Plaza: Which aspects of the physical infrastructure should decision-makers focus on more today in order to avoid retrofitting later?

Decision-makers should focus less on existing standards and blueprints and more on anticipating the physical limitations of the coming years. Anyone building today must consider architectures for 1.6 terabits and beyond. Latency and optical attenuation will become the hardest currency. At the same time, AI and HPC workloads require clear separation and integrated planning of front-end and back-end networks: while front-end networks primarily grow through standardisation and migration capability, back-end networks are physically limited – in particular by energy and cooling concepts as well as high-density fibre optic connectivity.

Increasing rack power outputs of well over 300 CW in the future inextricably link network design, fibre optic architecture, power and cooling into an overall system that can only be planned in an integrated, future-proof manner. A modular, high-density and energy-efficient fibre optic base is therefore crucial, enabling backend scaling far beyond the 800G limit without having to retrofit the backbone – both for training data centres and for the ever-growing inference operation. Inference in particular will become increasingly important for companies in order to support applications locally or decentrally at the edge, in real time, close to data generation with trained AI solutions of various generations.

Disruptive technologies should be included in roadmaps at an early stage. These include, in particular, hollow core fibres in data centre interconnects, which open up new possibilities in terms of architecture and location selection thanks to minimal latency. By taking these physical relationships into account today, you can avoid interventions later on, reduce life cycle costs and create a robust foundation for AI-enabled data centres in the next decade.