Governments across the world are looking at how they can stay at the forefront of developments in artificial intelligence.

 Earlier this year, the UK Government announced the establishment of ‘AI Growth Zones’ (AIGZs), with the dual objectives of accelerating the build-out of data centres, while also helping to drive local rejuvenation. Regional authorities, local authorities and industry were invited to bid for sites that can host at least 500 MW of data‑centre capacity by 2030, with an announcement expected on the first tranche of AIGZs this summer. Submissions for future AIGZs re-opened last month.

However, hyperscale datacentres come with very specific local requirements. In this article we start to disentangle what might make a promising location, and highlight the local authorities which may be particularly attractive for datacentre investment. While the AIGZ initiative spans across the whole UK, this analysis focuses on England as some datasets do not cover all nations.

Watt’s up

A single AIGZ‑grade campus is slated to draw at least 500MW of power, comparable to an entire city. Energy is the biggest cost of running a datacentre, and investors will need to account for all sorts of complications, like the availability of cheap and clean generation, or the capacity of the network to transmit power to where it is needed.

One consideration is how the datacentre will physically ‘plug in’ to the network. Local distribution networks almost never carry that amount of spare capacity, so developers will typically need a new connection straight into the national transmission system. Sites that are near to National Grid’s lines will therefore avoid lengthy (and costly) new cables.

We calculate the share of each lower tier local authority that lies within ten kilometres of a transmission line using National Grid’s overhead line maps. While the transmission network spans across the whole country, the map below shows how outlying areas like Cornwall and Lincolnshire tend to have a smaller proportion of sites near to a transmission line.

High Fibre

Hyperscale datacentres require reliable high bandwidth connections to users and other datacentres. As a guide for where these connections might be easier, we looked at Ofcom’s analysis of the number of available providers for leased lines in each area. This allows us to calculate the share of each authority classified as having a ‘high network reach’. While datacentres themselves will typically build or lease their own dedicated connections, having a diverse set of local networks may provide additional options for redundancy.

As shown on the map below, the greatest levels of connectivity are in urban areas. These will also tend to be areas where the datacentre is in close proximity to a larger number of users, which can help to reduce latency.

Cooling the cloud 

Servers expel waste heat that must be dumped, and this generally involves the usage of water. The amount of water consumed varies depending on the size of the datacentre, the technology used, and the local climate, however one estimate suggests a typical datacentre could consume as much water per day as a town of 30,000 to 50,000 people. With parts of the country under the threat of water shortages, availability of water could limit the locations for large-scale datacentres.

Using Environment Agency data we computed the proportion of water bodies in each district rated as having water available for licensing. The map below clearly shows the relative scarcity of water in the South-East.

Room to grow

Hyperscale datacentres are huge physical installations which require large areas of land, and corresponding planning permission. While the AI Growth Zones guidance looks favourably on the use of brownfield land, there will be relatively few brownfield plots sufficiently large for the types of datacentre that are envisaged.

To give a sense of the constraints that available land presents, we measured the share of each local authority’s area which is not taken up by built-up areas, Green Belt, National Parks and Areas of Outstanding Natural Beauty. As shown in the map below, the East Midlands region is shown as having the most available space according to this metric.

Local impact

The AIGZ process looks favourable on sites which can demonstrate benefits to the local community such as job creation and investment in local infrastructure. It may be easier to make the case in areas which are relatively deprived and so may benefit more from any ‘levelling up’ that local investment could provide.

We have mapped the average Index of multiple deprivation, a composite measure which considers different facets of deprivation: Income; employment; health and disability; education, skill and training; crime; barriers to housing and services; and the living environment. The map below shows a clear divide with areas in the South-East having on average much lower levels of deprivation.

Putting it all together

The different factors rarely align in a way which makes one area more attractive to datacentre investors on all measures. For example, sparsely populated areas with lots of available land and more secure water supplies are often further from the energy and telecoms infrastructure that datacentres rely on. As a result, no local authority is in the top 25% of all areas when rated across all five factors.

While smaller individual sites within a local authority may look more attractive, this demonstrates how compromises will inevitably need to be made by investors and policymakers. Utilities will need to understand how these decisions are made in order to best predict and prepare for the demands that hyperscale datacentres will place on them.

Although no area scored at the top of every metric, we can use this data to list the ‘all-rounders’ which are in the top half of the rankings across all five factors, and so may be an attractive location for would-be investors to consider. These are shown in the map below:

The AIGZ policy is attempting to spark local rejuvenation, as well as increasing data centre capacity. Are these multiple objectives in competition? The evidence is mixed:

■        On the one hand, taking account of local economic factors will limit the number of areas deemed suitable for datacentres. If we only had only considered the four physical constraints and not the IMD, the map above would have another four local authorities highlighted – two in the South West, and two in the East Midlands.

■        However, the remaining physical constraints still conflict with each other. No area is in the top 25% of the rankings for the four remaining metrics.

There are also important questions about how AI datacentres will demonstrate a local impact. While targeting relatively deprived areas with higher unemployment might seem fairest, a datacentre needs relatively few employees. Local governments may wonder whether these sorts of benefits are worth the potential costs (such as the impact on the views of those living nearby). While the AIGZ announcement references other types of local impacts (like benefitting nearby researchers and industries using AI tools), it’s not immediately obvious that having a datacentre in close proximity will help these types of users. Investors and local authorities which want to demonstrate these sorts of impacts will need to think carefully about how to do so.