Every banking transaction, every app, every health record and picture you take relies on physical servers somewhere. Data centres aren’t optional – they’re critical. But public debate often fixates on scary headlines. At Asanti Data Centres, we thought it was time to share a clear, business-focused reality check.
Modern UK facilities commonly use closed-loop cooling. Once the system is filled with water and mixed with coolant, the liquid is continually recirculated rather than constantly drawn from the mains. Water use is minimised by design.
Data centres don’t set consumer tariffs. The more pressing issue is the grid constraint. The UK has been paying wind farms to curtail production because the grid can’t move all the renewable energy to where it’s needed. If private-wire connections were enabled at scale, data centres could absorb excess clean power directly – cutting waste, carbon and cost.
Context on power really matters. Yes, data centres are consumers of power, but they are not the most energy-intensive industry in the UK. Data centres currently use around 2.5% of the UK’s electricity, placing them outside the top 10, which feature industries such as food manufacturing, agriculture, construction and transport. Our energy consumption is not going to decline, so the question becomes: how do we plan to meet this demand? Key to securing our energy future is the diversification of energy services. The upgrades currently taking place to the National Grid cannot keep pace with demand, and we can’t meet today’s demand at the pace of the past. We need to look to solutions such as battery storage and diversification of energy services away from the main providers.
Digital demand keeps rising (video alone is >50% of traffic), and mission-critical services, from banking to the NHS, now depend on data centres. Sustainability is possible, and it hinges on how we build:
Yes, AI increases computer density, but not all AI is the same. Many enterprise workloads are AI-enabled, not AI-at-scale. Designing every new site only for 100–150 kW racks invites overbuild and stranded capacity. A measured, mixed-density approach avoids repeating dot-com-era inefficiencies.
What does that mean?
There are two distinct phases of AI: training, where large language models (LLMs) are being trained to develop AI, and Inference, where they become part of an application for everyday use – think Microsoft CoPilot or Alexa. The training of the LLMs is where the increase in energy comes into play. The computational power required to support the training of LLMs is significant, as is the power draw required for the mechanical infrastructure required to keep the technology cool. The training phase of LLMs is typically around 6-12 months, so while there is an increase in power requirements, it is temporary and most organisations are using deployed AI, rather than developing their own.
Once the LLMs have completed the training phase, the AI is deployed into an application, where it can run in just about any data centre.
Don’t be distracted by extremes. The opportunity is to marry demand with smarter energy policy, faster connections, and pragmatic engineering so the UK remains competitive with lower-cost European markets.
If you’d like to hear more from the people shaping the UK’s digital infrastructure, tune in to In Conversation With, our podcast series featuring candid discussions with business and technology leaders.
Find all the episodes here: https://asanti.com/resource-centre/podcasts/
For a deeper dive, explore our latest white papers – designed to help organisations make smarter, future-ready decisions about their data strategies.
Download Asanti’s Whitepapers: https://asanti.com/resource-centre/guides-white-papers/
This article is part of a three-piece series. Find out more about Data Centres here: What Is a Colocation Data Centre? And Why UK Businesses Are Re-thinking Where Their Data Lives
