Table of Contents
Rob Shore, head of optical networks solution marketing at Nokia Network Infrastructure, told RCRTech that power is the main physical constraint facing new AI data center development
In sum – what to know:
Power drives geography – Limited power availability is influencing where AI data centers are built, creating a more distributed infrastructure footprint and new connectivity requirements.
Networks grow outward – Distributed AI facilities require additional fiber and long-distance optical connectivity, while routing must support different requirements for training and inference.
Automation gains importance – Network automation can accelerate turn-up, optimize resources, protect SLAs, and increasingly use AI to predict and prevent disruptions.
Power availability is emerging as the primary constraint shaping the expansion of AI infrastructure, with implications that extend well beyond individual data centers and into the networks connecting them.
Rob Shore, head of optical networks solution marketing at Nokia Network Infrastructure, told RCRTech that power is the main physical constraint facing new AI data center development. As developers seek locations with sufficient power, AI infrastructure is becoming more geographically distributed, increasing the need for high-speed optical connectivity between facilities.
“The primary constraint is power access. AI DCs are extremely power intensive and securing sufficient power dominates when and where new facilities are built. The result are modular, distributed DCs being built in more locations. These need high-speed optical connectivity,” he said.
This shift is changing the relationship between data center development and network infrastructure. Rather than building AI capacity primarily where established fiber infrastructure is already available, developers may need to connect facilities in locations selected largely because power is accessible.
That creates additional requirements for fiber construction and long-distance optical transport.
Shore identified fiber availability as the next major constraint after power. As optical systems approach the capacity limits of individual fiber pairs, connecting distributed AI infrastructure can require thousands of fiber strands between locations.
“Next major constraint is fiber availability. Due to per-fiber capacity limit, connections now require thousands of fiber strands between locations. Ability to access multiple fiber pairs and rapidly turn up optical transport capacity critical to aggressive AI, cloud and hyperscaler deployment timelines,” he added.
The implications extend into network architecture and routing. Shore said routing requirements are evolving as AI workloads become more demanding. AI training requires faster, more deterministic and lossless connectivity, while inference requires secure, low-latency connectivity. That makes closer integration between IP and optical transport increasingly important.
The growth in AI-enabled data center capacity is also driving additional network investment. Shore said higher capital expenditure on AI data centers is increasing the need for network expansion and optical connectivity for both data center interconnect and scale-across architectures.
However, he said the allocation between advanced computing and connectivity remains largely consistent, with both areas benefiting from ongoing improvements in cost efficiency.
Over the next three to five years, Shore expects power to remain the primary limiting factor. While some component constraints have relatively clear paths to mitigation, he said power availability does not have an immediate or easy solution.
That constraint is likely to continue influencing the geography of AI infrastructure and, consequently, the networks required to connect it.
As the infrastructure footprint becomes more distributed, automation is also becoming increasingly important. Shore said network automation can help accelerate operational turn-up, optimize resources for peak performance, and protect service-level agreements through proactive resiliency.
Automation is particularly relevant when networks span more locations and require more connectivity resources to be managed.
The role of automation is also evolving as AI is incorporated into network operations. Shore said network automation will increasingly incorporate AI to enable intelligent and predictive outcomes while reducing waste in optical resources.
The interview with Nokia’s Rob Shore is part of a report published by RCR Wireless News and RCRTech, titled Scaling Optical Networks for the Hyperscale and AI Era, which can be accessed by clicking here.
