Executive Summary

• Ozgur Duzgunoglu, Senior Director of Design & Engineering at Telehouse Europe, shows us how AI and high-density compute have reset data centre expectations, reshaping where new facilities are located and how supporting infrastructure is developed.
• Cooling will radically shape data centre infrastructure innovation as liquid cooling continues to move from specialist applications towards wider adoption where rack density requires it, meaning hybrid environments will remain common for the foreseeable future.
• Site selection is a key factor in resetting expectations; while hyperscale AI training environments generally still gravitate towards large sites outside major city centres, many organisations need smaller, high-density environments close to key connectivity hubs and the end users.

Rack density expectations are rising far faster than many data centre infrastructure roadmaps were designed to accommodate. Even as recently as three years ago, many infrastructure plans were built around 4 to 8 kW per rack. Within a year, requests moved past 10 kW, and today, 15 kW has become the common starting point for high-density deployments.

This rapid shift is already reshaping future capacity planning and influencing how new sites are designed and built. AI is changing the profile of demand, with emerging AI businesses requiring high-density compute alongside fast, reliable connectivity. Gartner’s November 2025 CIO survey found that by 2030, all IT work is expected to involve AI, with 75% performed by people augmented by AI and 25% by AI alone.

AI is clearly a major accelerator, but it is not the only driver. Storage intensity is rising across workloads, alongside demand for additional services such as high-speed interconnection to carriers, cloud providers, partners, and internal sites. Data centres must therefore prepare for a steady spread of higher-density requirements, rather than treating them as isolated specialist deployments.

Flexibility to expand

The rise in rack density is also changing what customers expect from their data centre agreements. Scalability is now being written into customer agreements from day one, with organisations assessing not only what a provider can support today, but how it can accommodate their growth over the next three to five years. For example, this might mean supporting pockets of 25-30kW racks inside larger deployments.

Data centres must be planned and built with that progression in mind. Operators need to assess grid capacity before committing to designs that may be ready years before the power is available. This requires earlier coordination with utilities, grid managers, energy developers and public authorities to align new demand with regional priorities. Modular construction can reduce stranded investment, but only when power and cooling are phased against realistic demand and utility delivery schedules.

This coordination should also inform where new facilities are located and how supporting infrastructure is developed. Renewable generation, storage, and flexible onsite energy assets can strengthen the wider energy system, while non latency-sensitive workloads may be scheduled around periods of lower grid demand where operational requirements allow.

Rising power requirements also bring cooling strategy to the forefront. Facilities must be able to manage the additional heat generated by higher-density racks without compromising resilience or limiting their ability to expand.

Cooling is central to future capacity

Higher-density racks generate more heat, which needs to be directed away from the facilities reliably. The most appropriate cooling strategy will vary by location. In cooler regions, such as the UK and Northern Europe, lower ambient temperatures can extend the periods when air-side or other economiser cooling systems operate efficiently, reducing reliance on mechanical refrigeration. According to the Uptime Institute Cooling Systems Survey 2025, traditional air cooling solutions are used by the majority of data centre operators.

For the densest AI and high-performance computing workloads, direct liquid cooling is already becoming a core infrastructure requirement. While 22% of operators currently use it, the focus is shifting from whether it will be adopted, to how facilities can support it reliably and at scale. Pipework, controls, heat exchangers, water quality management, leak detection, maintenance skills, and commercial terms all need to be established.

Air cooling will not disappear as liquid cooling adoption increases. Liquid cooling is continuing to move from specialist applications towards wider adoption where rack density requires it, meaning hybrid environments will remain common for the foreseeable future. Operators need to run air and liquid cooled systems side by side without weakening resilience or creating stranded capacity. This requires an architecture that can expand in stages while maintaining clear responsibility for the boundary between facility and customer equipment.

Resource use and location are key priorities

Power, water and wider resource use are becoming central to how new data centre capacity is planned and designed. Gartner projects that the global data centre electricity demand will double by 2030.

Buyers with higher density demands are still expecting their infrastructure stack to have high standards to help meet their sustainability goals, including responsible resource usage. The industry therefore faces constant tension between rising utility demand and sustainability commitments.

This pressure should drive more rigorous measurement and better design choices. For example, closed-loop cooling systems can tackle water usage constraints, while heat reuse can offer heating solutions value where genuine local demand exists.

In addition site selection is a key factor resetting data centre expectations alongside power, connectivity, and workload characteristics. While hyperscale AI training environments generally still gravitate towards large sites outside major city centres, many organisations need smaller, high-density environments close to key connectivity hubs and the end users.

Credible options depend on decisions made now

High-density capacity is todays’ challenge for tomorrow. It is a very imminent hurdle for data centres long term planning strategy, needing to take into consideration energy infrastructure, sustainability, and customer architecture. The operators best placed for the future will be those that make disciplined decisions now: build only what demand justifies, reserve what future growth will require, and prove that power, cooling, and connectivity can be delivered within defined costs and timescales.