AI Gridlock: How Hidden Infrastructure Problems Risk Derailing the UK's AI Ambitions?

Vast sums are being invested into AI infrastructure by big tech. In 2025 alone, the largest US companies have spent more on AI than the US government has on education, training, employment and social services. The bulk of this Capex investment has been spent on data centres and expensive semiconductor chips made primarily by Nvidia.

Data centres consume huge amounts of power and (sometimes) water, creating upstream infrastructure problems. The tech giants are taking care to ensure they will have enough energy. As well as hybrid partnerships with energy companies, tech firms are investing in new nuclear facilities and renewable off-grid sources.

The UK has entered a similar AI race, just will less zeros at the end. Ironically, it is also being led by US companies, although there are some homegrown firms in the mix too. This national boom will have huge implications for a known bottleneck in the system, the National Grid.

Electricity

The data centres being planned in the UK today are on a much larger scale than we have ever seen before. Decades ago, large data centres used 2-4MW. Today, the UK’s biggest runs at 120MW.

For context, a town or city the size of Cambridge might consume about 80MW. Planned AI data centres will start at 500MW and could reach GW scale. Unfortunately, the National Grid is creaking at the seams; this is particularly evident in London. Currently, there are very few places in the UK where you can deliver that much power, and you could be waiting many years for new grid infrastructure to be made available.

The UK has another big challenge: the price of electricity. Even at wholesale rates, our electricity is probably the most expensive in Western Europe. We generate some of the cheapest renewable electricity in the world, but national pricing is skewed by gas. This makes it difficult for data centre operators to compete with other European locations which can supply large amounts of cheap, renewable energy - the Nordic countries in particular are considered as prime locations for AI data centres.

There are however a slew of government initiatives and policies intended to update the grid and solve these infrastructure barriers.

AI Growth Zones

A centrepiece of the UK’s strategy is the creation of ‘AI Growth Zones’ - special zones with favourable conditions for AI infrastructure. Announced in the AI Opportunities Action Plan (Jan 2025), these zones are intended to accelerate data centre build-out. Regions across the country have been invited to bid to host AI Growth Zone.

Each zone will offer streamlined planning approval (fast-tracking permits) and guaranteed power availability. In fact, the government has pledged to work with network operators to rapidly provide 500 MW+ of power per zone. While this is promising, it should be noted that the huge data centres in these regions will take at least five years to build, and this leads us to another problem: rack density.

Rack Density

The rate at which we can build data centres is out of pace with the acceleration of AI technology. Nvidia, the leading supplier of AI processors, is constantly upgrading its products. This rapid development means data centres designed today may be outdated by the time they are built.

The key issue here is what we call ‘rack density’. Data centres house computers in racks or cabinets, usually about 1m wide, a bit over 1m deep, and 2m tall. Pre-AI, the global average ‘density’ for a data centre rack was 8kW - in other words, each rack could accommodate up to 8kW of electricity consumption (and cooling).

Nowadays, some AI data centres are running at 50kW per rack. The current top-end Nvidia systems need upwards of 120kW/rack. They have announced that the next generation (expected in 2026) will run at 600kW/rack. This presents a huge challenge, as operators must feed 600kW into something the size of a cupboard.

Summary

The UK’s strategy for electricity supply is to plan ahead and invest: by channelling data centre growth to places with ready power, by massively upgrading grid infrastructure, and by reforming regulations to connect new sites faster.

Huge challenges remain - for instance, Microsoft has warned that without swift grid upgrades some of its UK data centres might wait until the mid-2030s for full power connection. Experts warn of a disconnect between UK AI ambitions and timely power access.

However, the ongoing measures - from 500MW growth zones and new substations to national grid investment plans - indicate a concerted effort to ensure the lights stay on for Britain’s AI revolution. The government’s Plan for Change aims to tackle these constraints so new AI centres can be plugged in, enabling the UK to realise its AI ambitions.

If you would like to discuss these topics further with Duncan Clubb, Senior Partner for Data Centres, Edge & Cloud at Cambridge Management Consulting, you can connect with him here: https://www.cambridgemc.com/people/duncan-clubb