LEE MYALL POWER, COMPUTE, AND FIBRE
THE UK’S AI AMBITIONS DEPEND ON A TRIFECTA: POWER, COMPUTE, AND FIBRE. The UK’s AI ambitions are increasingly defined by the number of data centre announcements. Media headlines emphasise megawatts, land, and GPU counts, with significant levels of investment seemingly announced every day. The subtext is clear: those that win in the next phase of international competition will be those that can build the largest compute clusters the fastest, writes Neos Networks CEO Lee Myall . A nd there is truth in that. AI does demand unprecedented compute. But there is a growing disconnect between discussions about AI infrastructure readiness Inference is even more network- dependent than many realise. Today’s AI services do not work in single steps: they collate data, assess context, and personalise responses while connecting to multiple systems in real time. That generates increasing east-west traffic, not just user- to-server but between different services, platforms and sites.
environments and is strategic for national resilience. But for the core requirements of AI-driven data centres (high-capacity long- haul transport, metro aggregation, and data centre interconnect (DCI)), fibre remains irreplaceable. Networks between data centres are not just dumb pipes, supporting AI workloads, datasets, and services to move smoothly across sites. As organisations spread AI across hybrid and multi-cloud environments, fast and reliable DCI becomes a real advantage and needs to be designed much earlier than it is today. A single fibre path into a site can easily become a bottleneck or a single point of failure. For AI workloads, resilience needs to be built in from the outset, with route diversity and proper capacity planning. THE INVESTMENT MISMATCH: FIBRE MATTERS MOST WHEN IT IS INVISIBLE Despite the evidence, fibre is still often seen as supporting infrastructure, not a gating factor that determines whether a data centre can go live at all. This is likely because data centre investment is more visible, whether it is land acquired, megawatts planned or jobs created. Fibre is less flashy and more operationally complex, which makes it harder to communicate. As a result, there is not enough awareness that fibre is the layer that turns AI ambition into deployable, scalable infrastructure. The UK is already suffering from timelines being out of sync. As data centre developers accelerate expansion plans, fibre availability (and importantly, fibre readiness in the right corridors) is not always being planned with the same urgency. Recent research from
and actual AI workloads. Despite compute being the engine, AI is ultimately a network workload. It relies on distributed datasets, distributed compute, and constant east-west traffic between systems and sites. All in all, the foundation of AI-led growth is not just power and silicon; it is a trifecta that includes high-capacity fibre connectivity. The risk for the UK is not that it underinvests in data centres. It is that it invests out of sequence, eagerly building large-scale capacity without the optical fibre networks necessary to connect, scale and operate them reliably. AI IS A NETWORK WORKLOAD To see why fibre matters so much, a clearer understanding of how AI actually works is required. Traditional workloads could tolerate a degree of network variability, largely because they were contained within a single environment, such as an office, a regional data centre, or a single cloud region. But AI changes that pattern. AI workloads are inherently distributed. Training LLMs requires moving vast datasets between storage and compute, coordinating work across clusters, and keeping hundreds (or thousands) of machines in lockstep. While compute sits in one location, the data it depends on often does not. Datasets and model checkpoints travel between enterprise environments, clouds and specialist AI infrastructure providers.
In short, AI has three important consequences for connectivity: 1. Network performance directly shapes application performance. When a network is slow or inconsistent, the AI application is too, and customers feel it immediately. 2. Resilience is no longer optional. AI services rely on multiple systems and locations combined. If any link fails, performance suffers. 3. Scaling needs to be fast and flexible. Demand can spike overnight, so connectivity needs to be easy to add, reroute and upgrade. That said, AI is creating a different kind of traffic, with far less tolerance for delay or instability. It is resetting the standard for good connectivity.
FIBRE IS THE CONNECTIVITY WORKHORSE OF AI INFRASTRUCTURE
But not all connectivity is created equal, and fibre is the only medium that offers the combination of scale, predictability and resilience that AI demands. Of course, other connectivity options also play a part. Fixed wireless access and 5G can be valuable for specific last-mile use cases and for rapid deployment, and satellite is increasingly important for remote
8
| ISSUE 43 | Q1 2026
www.opticalconnectionsnews.com
Made with FlippingBook flipbook maker