The UK’s ambition to become an AI superpower isn’t just about policy papers and investment announcements. It requires physical infrastructure capable of handling the intense demands of modern, GPU accelerated, high-density AI workloads – and right now, there isn’t nearly enough of it.

At Kao Data, we’re working to change that. But, creating genuine AI-ready data centre space isn’t simply about installing more powerful servers in a legacy data hall. It requires new space, built for purpose, supported by rigorous preparation, testing and validation, long before a single GPU goes live. 

This is where our critical partnerships with specialists like Gratte Brothers and Heatload become absolutely vital.

Why AI Demands a Different Approach

Traditional enterprise workloads and AI training environments operate in fundamentally different ways. AI servers generate significantly more heat, require greater cooling capacity, and demand faster system response times when failures occur. A data hall that works perfectly well for standard enterprise applications can quickly become a costly liability when AI workloads doing life-saving research are deployed.

As we highlighted in our recent “AI Taker not Maker” campaign, the UK faces a genuine risk of falling behind in the global AI race. One of the most pressing – yet often overlooked – challenges is the shortage of suitably commissioned AI-optimised data centre capacity. The Government’s compute roadmap is ambitious, but ambition means nothing without the physical infrastructure correctly put in place to support it.

The Critical Role of Commissioning

This is where Gratte Brothers’ expertise becomes invaluable. As principal contractors with deep mechanical and electrical (M&E) capabilities, they understand that commissioning is about more than simply checking boxes on a compliance form. It’s about verifying that what has been designed and installed actually meets real-world performance requirements under stress.

Gratte Brothers’ approach involves staged load testing at 25%, 50%, and 75% capacity, deliberately simulating failures in power and cooling systems, and trending heat levels to ensure the data hall can maintain the precise temperature ranges and humidity levels that AI compute demands.

The electrical systems (UPS, generators, power distribution) are rigorously tested to ensure they can achieve the expected 15-second turnaround from power failure to generator backup. But, it’s the mechanical systems where the complexity really becomes apparent. 

When a chiller is deliberately shut down during testing, the entire cooling infrastructure must demonstrate it can recover quickly and restore stable conditions. For AI workloads, particularly for hyperscale customers, where thermal consistency is paramount, these recovery times can make the difference between a resilient facility and a catastrophic failure.

Testing What Can’t Be Seen on Paper

Theoretical designs, however sophisticated they may be, cannot predict every real-world interaction across complex, interconnected systems. As Gratte Brothers have learned through projects including our KLON-06 development, commissioning regularly uncovers issues that would never be evident on paper – often around equipment control settings and interface points between systems that only become apparent when everything is stress-tested together.

This is where Heatload’s specialised equipment becomes essential. Before expensive IT hardware is installed, their emulators accurately mimic the power consumption and heat rejection characteristics of the AI servers that will eventually occupy the space. These aren’t simple space heaters. They’re sophisticated systems designed with an “IT first” philosophy, sized to fit within racks or replace them entirely, ensuring heat is distributed exactly as it will be under live conditions.

Crucially, Heatload’s equipment connects to the final power outlets, testing systems end-to-end, whilst the temperature rise from inlet to exhaust is carefully balanced to adequately stress the cooling infrastructure. Their rental-based approach is also more sustainable for data centre operators like Kao Data, as the equipment is re-used across multiple projects rather than manufactured once and discarded.

Building for Future Demands

The partnership among Kao Data, Gratte Brothers, and Heatload on projects like KLON-06 exemplifies the detailed, carefully thought-out approach required to build hyperscale-ready AI capacity. By the time AI workloads go live, every system has been proven under conditions that greatly exceed expected operational loads. Sensors placed throughout the data hall provide comprehensive monitoring data, which is fed back to consultants and clients and sometimes requires design modifications before final sign-off.

This level of meticulous preparation is precisely what the UK needs more of. In our “AI Taker not Maker” whitepaper, we identified energy pricing, AI copyright law, and infrastructure readiness as critical barriers to the UK’s AI ambitions. 

The infrastructure challenge is the most tangible and the most immediately solvable. Working with expert partners like Gratte Brothers and Heatload helps Kao Data at the bleeding edge of data centres engineered for AI.