By Andrew Tipping, Account Director, Zayo Europe
AI data centres need four things: land, power, cooling and connectivity. Without the fourth, the UK’s new AI Growth Zones and other AI data centres could quickly become expensive white elephants.
This summer will see the announcement of the UK’s first AI Growth Zones – data centres designed to handle the vast data volumes and intensive compute requirements created by soaring AI use.
Within three months of requesting expressions of interest in February, the Department of Science, Innovation and Technology (DSIT) had received more than 200 responses from local authorities, keen to capitalise on the job and wealth creation opportunities of a booming industry sector.
But not every site will be suitable for an AI Growth Zone. Land availability is one factor, with sites requiring 100 acres or more to qualify for support.
AI-enabled data centres also consume huge amounts of power[1] . Ciena’s recent Wave Services report shows that an AI training data centre consumes 9x more energy than a standard data centre, for example – which is why DSIT requires evidence of at least 500MW capacity by 2030.
Water is another key consideration, as it’s needed to cool the servers without affecting local supply. Bids must show evidence of sufficient water to cool at least 500MW of AI infrastructure.
Connectivity comes a worrying fourth
Bottom of DSIT’s list of technical criteria is connectivity. Unlike land, power and water, no specific requirements are set out. However, DSIT says it will look favourably on applications with evidence of ‘strong, reliable, and readily available’ fibre and mobile connectivity.
This lack of rigour around connectivity, compared to land, power and water, is worrying. Fibre-optic cable is the only way to get the vast amounts of data required to train and operate large language models (LLMs) into and out of new-build data centres. If the right connectivity isn’t there, the data centre won’t be able to fulfil its potential as an enabler of AI-fuelled growth.
Model training used to be done offline, putting less pressure on connectivity infrastructure. But AI use cases are increasingly moving towards real-time deep learning model training, necessitating fast, continuous flow of high volumes of data into the data centre. The need for trained LLMs to deliver answers fast, often to other parts of the country, also requires continuous high-capacity, high-bandwidth connectivity.
Four connectivity considerations for AI Growth Zones (and other AI data centre builds)
For these reasons, we’d strongly encourage local authorities and landowners to consider four aspects of connectivity when preparing an AI Growth Zone bid or AI data centre build.
1. Proximity[2] [3] :The nearer the site is to a high-speed fibre backbone, the easier, quicker and less costly it will be to get it connected. With planning permissions and wayleaves often taking months or even years to negotiate, the fewer new cables that have to be laid, the better.
In recognition of these issues[FC4] , DSIT has implemented a Barriers Busting team with the remit to address some of the bureaucratic and regulatory bottlenecks in the system. However, a review conducted by Zayo and others in the industry has found that the need for AI-suitable infrastructure is moving faster than this initiative, [5] which is focused on the delivery of commercially-funded and BDUK-subsidised gigabit-capable broadband infrastructure.
With that in mind, data centre operators should be aware that the nearest AI-suitable transport network may not belong to the operator who comes to mind first, meaning it’s worth exploring alternative providers. When DataVita set up its DV1 data centre in Lanarkshire, Scotland, for example, it was pleasantly surprised to discover that Zayo Europe’s owner-operated fibre backbone passed closer to its chosen site than any other provider’s network.
2. Diversity: With data centres now designated as Critical National Infrastructure, and AI workloads becoming increasingly business- and mission-critical, data centre connectivity must be both robust and resilient. Ensuring route diversity—with multiple fibre routes into the data centre—has long been a priority, to prevent network slowdowns or downtime.
What’s changing is the number of routes now considered necessary to ensure continuous high-speed and low-latency uptime. Where once we talked about the need for biversity or triversity, the coming buzzword will likely be ‘quadversity’ – with four separate routes to mitigate risk.
It’s important to recognise that some routes are only partially diverse – for example, the fibre may be in separate cables but laid in the same poles and ducts. Adding fully-diverse routes where possible will strengthen the resilience of the network.
3. Capacity: An AI-enabled data centre will only be as successful as the speed at which it can ingest, process and distribute vast amounts of data. Network capacity is of crucial importance and demand for ever-larger wavelengths (the measure of data that can be transported in one second)is soaring.
Until very recently, our conversations with data centre operators were mostly around 100 gigabit-per-second wavelengths. With the growth of AI it’s hit a tipping point[6] and suddenly everyone’s talking about 400Gb/s.
But wavelengths are only part of the story—overall capacity per cable is vital, too. For example, here at Zayo, our newly-deployed cables have approximately 80 channels per fibre pair and typically 216 fibre pairs per cable, delivering a total capacity of around 6.9 petabits per second per cable.
Choosing a fibre provider with that kind of capacity, as well as a 400G-enabled and 800G-ready network, will enable networks to scale as AI-fuelled demand for capacity continues to surge[7] .
Practically, that means looking for a provider with a credible technical roadmap to 800G waves (and beyond), and a solid timescale for availability. Ask about pilot projects, technology enablers—for example, what kind of optical technologies will be used—and 800G-enabled route maps.
4. Mindset: Complex projects always go more smoothly when everyone feels a sense of involvement and collaboration towards a shared goal. Connectivity is such a crucial aspect of data centre success that it’s worth bringing a partner on board early, and ensuring they’re fully aligned with the project.
Part of that is choosing a partner with a collaboration mindset, the flexibility to design and deliver (and manage and evolve) the right connectivity solution for the project, and the willingness to work together to overcome challenges both foreseen and unforeseen.
Talk to Zayo Europe about AI data centre connectivity
If you’re bidding for an AI Growth Zone, if you’ve already been awarded one (congratulations!), or if you’re planning an AI data centre build or upgrade, we can help you get the right connectivity in place.
Zayo Europe and our technology partner Ciena are working with companies across Europe to ensure they have the right connectivity for their current and future AI use cases. From 400G wavelengths to private wavelength networks and dark fibre for long-haul and metro connectivity, we can support your AI plans today and into the future. If you’d like to talk, please get in touch.