Last week I was delighted to take part in an interesting panel discussion on AI infrastructure and the considerations for data centres, hosted by our partners, Red Engineering Design. It’s always fascinating to be part of these industry meet-ups with so many views, opinions and experienced heads in the room and so firstly, hats off to RED for bringing us all together.
Most of the discussion centred around data centre design, build and implementation for AI workloads, such is the topic of the moment right now and one of the questions to the panel I found most interesting was whether the UK could develop standardised cooling requirements for AI chips to reduce energy and water demand. This has been a question that is often discussed because we all know that while AI capability and demand increases, the infrastructure supporting its rapid development must also evolve, and ensuring long-term sustainability is key.
On the face of it, with energy and water usage in AI data centres increasing, standardising cooling requirements for AI chips across the UK could help to curb these environmental impacts. So is it a realistic goal, and what are the obstacles to progress.
Firstly, as I mentioned on the panel, one of the fundamental barriers to standardising cooling for AI chips is the sheer speed at which GPU and CPU technologies are evolving – almost on a 6 monthly cycle! AI chips are not just becoming more powerful; they’re also becoming more diverse in terms of thermal output and architectural design. NVIDIA, ARM, AMD, Intel, and a growing range of custom chipmakers are releasing new generations of hardware at an accelerated pace — each with different cooling needs.
This rapid innovation cycle means that any rigid standardisation implemented today may become obsolete tomorrow. For example, a standard designed for air cooling at a certain thermal design power range might not accommodate the needs of future chips requiring direct-to-chip liquid cooling or immersion technologies – something we’re currently implementing in our new KLON-03 data centre in Harlow.
This dynamic nature of the chip landscape makes it difficult — and potentially counterproductive — to impose strict cooling standards that could restrict flexibility and innovation. Instead, the industry might need to consider adaptive frameworks rather than hard standards.
Secondly, another layer of complexity is how quickly engineering consultants like RED and the broader supply chain are adapting to these hardware changes. New cooling solutions — such as advanced coolant distribution units (CDUs), rear-door heat exchangers, or two-phase cooling systems — are being developed in tandem with emerging chip technologies.
As cooling technologies evolve, so too does their efficiency. Rather than locking the industry into a particular solution, the UK should focus on fostering a climate of innovation where engineers are incentivised to design the most efficient systems for the latest chip generations. Efficiency, in this case, becomes the outcome of smart engineering rather than regulatory compliance.
This also means that the best results might come from encouraging knowledge sharing and setting performance-based targets (e.g., power usage effectiveness (PUE), water usage effectiveness (WUE)) rather than prescribing the exact methods to be used.
The next obstacle in developing any national standard is regulation — or rather, the lack thereof in our industry. The UK data centre industry is not yet truly regulated, especially when compared to sectors like construction or energy. Introducing enforceable standards for cooling AI chips would require first establishing a regulatory framework that defines who is responsible, how compliance is measured, and what consequences exist for failing to meet requirements.
This is a significant hurdle. Currently, data centres are held to general environmental standards and planning permissions, but there is no uniform body overseeing technical operations such as cooling design or implementation. Without an independent authority or governing body, it would be difficult to align cooling standards across operators, particularly when many of them operate globally.
A more realistic starting point could be the development of industry-led voluntary codes or best-practice frameworks, similar to the EU Code of Conduct for Data Centres. These could later evolve into formal standards if sufficient buy-in and maturity are achieved.
Finally, even if standardisation were technically and administratively possible, geographical variability throws another wrench into the mix. Every data centre operates within its own unique environmental parameters: altitude, ambient temperature, humidity, access to water, and availability of renewable energy all affect what kind of cooling is most suitable.
Additionally, data centres are often constrained by local noise ordinances, planning permissions, and zoning laws. For instance, a rural site might be able to deploy large evaporative cooling towers, while an urban edge facility must rely on closed-loop systems to avoid noise and plume issues.
This site-specific variability makes it difficult to impose universal cooling standards without creating unintended inefficiencies or stifling innovation. Any future approach must be flexible enough to account for these localised differences while still driving toward broader sustainability goals.
So while the idea of standardising cooling requirements for AI chips in the UK is appealing from a sustainability perspective, the realities of rapid hardware evolution, diverse engineering solutions, regulatory immaturity, and site-specific constraints make rigid standardisation impractical — at least for now.
The UK however has a unique opportunity to lead — not by dictating specific cooling methods, but by championing outcome-based standards and transparency. Establishing national benchmarks for energy and water efficiency — such as maximum allowable WUE thresholds for AI workloads — would provide a target without limiting innovation. Operators would be free to choose the most suitable cooling method for their chipsets and site conditions, as long as they meet or exceed these benchmarks.
Furthermore, fostering collaboration between government bodies, academia, and industry through initiatives like the UK’s Digital Infrastructure Roadmap could help set a foundation for future regulation. The government could also offer incentives for facilities that meet or surpass sustainability targets, encouraging early adoption of advanced cooling methods.
Above all, the key lies not in forcing uniformity, but in guiding the industry toward smarter, more sustainable cooling — whatever shape that may take.
