Data centre operators have long planned around land, power and water. But as AI reshapes the scale, density and complexity of the sector, Gavin Faulds, Chief Commercial Officer at Emtelle, argues that fibre must now be treated as critical infrastructure.
The data centre industry has become rightly focused on power. But as AI accelerates demand for compute, there is another constraint we risk underestimating: the physical connectivity infrastructure that allows data centres to scale, interconnect and perform.
Across Europe, America and Asia-Pacific, developers are racing to secure land, utilities are struggling to provide enough grid capacity, and governments are increasingly viewing data centres as critical national infrastructure rather than simply commercial real estate.
Let’s have a new conversation
Naturally, much of the conversation has centred on power. Power availability has become the defining constraint for many new developments, while water availability and sustainability considerations continue to influence where facilities can be built and how they operate.
Yet amid all this discussion, another critical resource receives remarkably little attention – fibre. Not just the networking equipment inside a rack, but the physical connectivity infrastructure that allows modern data centres to function as highly distributed, AI-ready environments.
For too long, fibre has been viewed as infrastructure that follows construction. Secure the site, secure the power, design the buildings, and then connect everything together. That approach made sense when facilities were smaller, workloads were more predictable and expansion cycles were measured over many years.
In my view, this is no longer a technical afterthought. It is a strategic planning issue. Thanks to AI, today’s hyperscale facilities are no longer standalone buildings serving local demand.
Increasingly, they form part of interconnected campuses, regional ecosystems and global cloud platforms. AI training clusters span multiple data halls, and GPU workloads generate unprecedented volumes of east-west traffic, placing greater pressure on the density, resilience and speed of the underlying fibre network.
Capacity must be added rapidly without disrupting live environments. As a result, connectivity is becoming less of a supporting consideration and more of an enabling part of infrastructure planning.
Looking much earlier in the process
The challenge is that physical infrastructure decisions are often the hardest to change once a facility is operational. Operators can upgrade servers or replace switches. Cooling technologies will continue to evolve. Even power systems can be expanded over time.
But retrofitting physical connectivity infrastructure is rarely straightforward. Additional duct routes, new fibre pathways, higher-density cable management and revised access points all introduce complexity, cost and operational risk. In facilities where uptime is everything, every retrofit becomes a careful balancing act between today’s operational requirements and tomorrow’s growth ambitions.
This is why fibre deserves consideration much earlier in the planning process. As an industry, we’ve become very good at asking whether a site has sufficient power for future demand. We should be asking equally challenging questions about connectivity.
Can this campus double in size without major civil works? Has sufficient route diversity been designed into the network? Will today’s duct infrastructure, fibre pathways and high-density connectivity support the requirements of tomorrow’s AI workloads? How easily can new buildings, new halls or entirely new campuses be integrated into the network over the next decade?
This is where the detail of physical design matters. Pre-planned duct routes, spare pathway capacity, diverse entry points and high-density fibre connectivity all help determine how quickly a data centre can respond when demand changes. In high-density environments, these decisions influence deployment speed, cable management, resilience and the ability to scale fibre architecture over time.
These are no longer purely technical questions, but commercial questions too. The organisations that answer them well will be better placed to expand quickly, respond to customer demand and avoid the significant costs associated with rebuilding infrastructure that was never designed to evolve.
A crucial shift in mindset
Perhaps most importantly, they will retain optionality. No one knows exactly what AI infrastructure will look like five or ten years from now. Compute densities will change, optical technologies will evolve and new architectures will emerge. Demand for bandwidth will also continue to grow.
Planning for flexibility is therefore just as important as planning for capacity and requires a subtle but important shift in mindset. Instead of designing only for today’s requirements, operators need to consider tomorrow’s unknowns, including the routes, ducts and connection points that will allow future capacity to be added with minimal disruption.
We can no longer afford to view fibre as simply the final stage of development. Instead, it should be considered alongside the earliest planning decisions. For decades, successful data centre developments have depended on three fundamental resources: land, power and water. Those pillars remain essential.
But AI is changing what defines strategic infrastructure. The next generation of data centres won’t be constrained solely by the availability of megawatts; their ability to grow will also depend on the quality, resilience and scalability of the connectivity linking buildings, campuses, cloud platforms and customers.
Fibre is no longer just about connecting a completed facility. It is also about retaining future flexibility, reducing retrofit risk and enabling data centre operators to respond to changing capacity requirements. If land, power and water define where data centres can be built, fibre will increasingly influence how effectively they can grow.
