As artificial intelligence workloads scale across cloud platforms, data centres and edge environments, pressure is mounting on the optical networks that carry this traffic. Capacity demand is rising faster than traditional fibre architectures were designed to handle, forcing network operators to confront a fundamental constraint, how to move far more data without endlessly laying new cables, consuming more power, or expanding physical infrastructure.
That challenge sits behind a new trial by Colt Technology Services, which has tested Multi-Core Fibre across its London metro optical production network. The pilot positions Colt among the first network providers globally to validate the technology in a live environment and reflects a broader reassessment of how optical networks must evolve to support AI-driven growth.
Rather than increasing capacity by adding more fibres, Multi-Core Fibre embeds multiple cores within a single strand of fibre, allowing parallel transmission of multiple data streams. In practical terms, this enables a significant increase in bandwidth without the need for new duct space, additional cabling, or proportionally higher energy consumption.
Why AI is changing optical network assumptions
The growth of AI is altering network traffic patterns in ways that differ from previous cloud or streaming waves. Training large models, distributing inference workloads, and supporting highly interconnected data centre architectures generate sustained, high-volume east–west traffic rather than the more predictable north–south flows of earlier eras.
During the London trial, Colt compared Multi-Core Fibre with conventional single-mode fibre across routes of approximately 9 km and 63 km between two points of presence. Using technology supplied by Sterlite Technologies Limited, Ciena and Nokia, the network achieved an 800 Gbps line rate and successfully validated 100GE and 400GE services.
The testing programme covered chromatic dispersion, polarisation mode dispersion, crosstalk, throughput, fault analysis and optical return loss, all delivering satisfactory results. These are not marginal metrics. For AI-driven networks, predictability and signal integrity at scale are as critical as raw bandwidth, particularly as operators attempt to densify existing infrastructure rather than expand it.
Scaling capacity without scaling complexity
One of the central implications of Multi-Core Fibre is economic rather than purely technical. As AI traffic grows, the traditional response of deploying more fibre runs into physical constraints, rising maintenance costs and sustainability concerns. By increasing capacity within existing ducts, Multi-Core Fibre offers a way to extend network lifespan while limiting additional power draw and construction work.
Colt has framed the trial as part of a wider effort to deliver higher capacity without compromising performance, security or sustainability. From an operator perspective, the appeal lies in reducing total cost of ownership while preparing networks for workloads that are both data intensive and latency sensitive.
STL’s role in the trial highlights how optical innovation is moving from laboratory environments into field deployments. The company has already demonstrated Multi-Core Fibre across underground and duct networks, building out an ecosystem that includes fibres, cables and termination solutions designed for operational use rather than experimental settings.
A signal of what comes next for AI infrastructure
While the trial itself was confined to London, its implications are broader. As AI systems become more distributed and data-hungry, network bottlenecks increasingly sit outside data centres rather than inside them. Optical transport is becoming a strategic layer in AI infrastructure, not just a utility.
Multi-Core Fibre does not remove the need for continued investment in network infrastructure, but it changes the calculus. Instead of assuming linear growth in cables, power and physical footprint, operators can extract more value from what is already in the ground.
For AI-driven economies, that distinction matters. The ability to scale bandwidth without proportionally scaling emissions, energy use or construction could shape where and how AI workloads are deployed over the coming decade. Colt’s trial suggests that the next phase of AI infrastructure may depend as much on rethinking fibre itself as on building ever larger data centres.
