One of the hottest discussion points at the recent Datacloud event in Cannes was that AI is not creating a single new data centre model, but two distinct ones. The first is the hybrid facility, where conventional enterprise infrastructure must sit alongside HPC and AI workloads. The second is the build-to-suit AI factory,designed for one primary purpose: high-density, GPU-accelerated AI processing at scale.

The critical difference is not simply plant capacity, as both can be built around a similar MW-scale electrical and cooling basis, but real estate. A hybrid facility can require two or more times the footprint of a similarly powered AI factory because it must accommodate lower-density equipment, more varied layouts, and broader operational requirements.

In engineering terms, this creates a clear divergence in how commercial data centres must be planned and operated. Operators are no longer designing around a single set of assumptions, but around two materially different models: one based on hybrid coexistence, and the other on highly specialised AI processing at scale. In many cases, AI factories are pre-let and developed in close collaboration with the customer, with the design, infrastructure and operating model aligned to specific requirements from the outset.

Traditional enterprise colocation and cloud infrastructure remain a core part of the market, but the design assumptions that supported homogeneous power, space and cooling models are becoming less reliable. As customer environments become more varied, operators must plan for the coexistence of traditional enterprise, cloud, HPC and AI workloads within a broader range of densities, layouts and thermal conditions.

The engineering challenge is to design facilities that can accommodate thermally dense, power-intensive AI infrastructure while continuing to support heterogeneous compute environments reliably and efficiently over time. The issue is not simply one of higher connected load, but of different electrical behaviour, as AI workloads can produce rapid, highly synchronised demand changes that place greater stress on voltage stability, frequency control and overall power quality.

However, this level of operational integration and infrastructure flexibility is not yet consistent across the market, and many facilities remain constrained by legacy design assumptions that were not developed for mixed-density, AI-led environments.

AI Infrastructure and Data Centre Stack Coexistence

The shift from traditional rack-based servers to scale-out AI systems is changing the nature of the infrastructure stack itself. Modern AI deployments are increasingly designed as tightly coupled clusters rather than as collections of independent servers, but in practical terms these remain bounded customer environments rather than limitless fabrics. A customer deployment may, for example, comprise a nine-rack GPU cluster today and reduce to six racks as density increases, while still remaining a discrete, customer-specific infrastructure unit. This changes how operators must think about power, cooling, networking and space planning across the hall.

From an IT perspective, AI readiness requires far more complex integration across compute, storage, networking and orchestration in both new and existing environments, creating what is effectively a distributed systems challenge that sits alongside the facilities engineering challenge in the data centre.

This is reshaping traditional IT stacks and operating models. In most enterprise environments, AI is not introduced as a wholesale replacement for existing systems, but as an additional layer that must be integrated with established platforms, operational processes and governance structures.

Supporting enterprise AI environments now requires coordinated engineering across processor architectures, storage platforms, networking fabrics and the physical data centre environment. In practice, this is no longer a challenge that can be addressed separately by IT and facilities teams, but one that demands a more integrated engineering approach across both. This is likely to require a digital twin environment capable of modelling the full ecosystem, from building fabric and plant performance through to customer equipment, infrastructure behaviour and compute workloads.

Although quantum computing remains a peripheral consideration for most multi-tenant data centres today, it may begin to influence future infrastructure planning in selected environments. Where quantum systems are considered, they introduce additional engineering requirements, including cryogenic cooling, isolation from electromagnetic interference, and protection from noise and vibration.

Hardest Data Centre Problem

Designing for the coexistence of AI and traditional enterprise environments is now among the most demanding challenges in data centre engineering, planning and operations.

The challenge for data centre engineers like myself is to plan and design power and cooling topologies that preserve resilience, reliability and efficiency wherever AI clusters must coexist with existing enterprise infrastructure.

Commercial data centre operators must now support environments in which power density and cooling intensity vary materially within the same customer hall.

The power profiles of AI and HPC environments are becoming well understood. In practical terms, a single customer hall may contain 8kW enterprise racks alongside 120kW AI racks, creating significant complexity in power distribution. As rack densities continue to rise, higher-voltage DC power architectures, including 800V DC approaches, are increasingly being considered as a way to improve efficiency and simplify rack-level power delivery.

Power challenges now extend beyond simple capacity planning to include busway limits, conductor sizing, harmonics, transient load behaviour and fault coordination. Power distribution systems must also be capable of absorbing rapid GPU load spikes, synchronised compute bursts and highly dynamic workload patterns without compromising stability or resilience.

In many mixed AI environments, cooling becomes the dominant design engineering challenge. Operators may need to accommodate a combination of direct-to-chip liquid cooling, rear-door heat exchangers, and both single-phase and two-phase liquid cooling approaches, alongside emerging immersion methods and high-temperature liquid loops, often within the same broader facility design. As rack densities and heat flux continue to rise, two-phase cooling becomes increasingly relevant where the practical heat absorption limits of water-based systems are approached, and the specific heat capacity of the working fluid is no longer sufficient on its own to manage the thermal load efficiently.

In Kao Data facilities, AI racks are creating unprecedented localised loads, bringing power density engineering into much sharper focus. Design planning is increasingly centred on how to remove megawatts of heat efficiently at source, while also considering the future potential for heat reuse where this is technically and commercially viable.

Implications for AI, IT and Data Centre Team Collaboration

AI is not only changing data centre design, but also reshaping how enterprise IT functions are organised, operated and integrated with infrastructure teams.

According to Gartner, AI will touch all IT work by 2030. In a July 2025 survey of more than 700 CIOs, respondents said they expect no IT work to be carried out by humans without AI support, with 75% performed by humans augmented by AI and 25% by AI alone. Gartner’s conclusion is that organisations must balance AI readiness and human readiness if they are to sustain value from AI.

For data centre operators, this matters because the impact of AI is no longer confined to the application layer. It is reshaping the relationship between enterprise IT, infrastructure planning and operational delivery, and creating a stronger need for coordination across all three.

As AI adoption grows across the enterprise, customers increasingly need support that spans both infrastructure engineering and operational integration. At Kao Data, that means helping customer teams focus on how hybrid architectures are developed and managed across mixed processor environments, legacy and emerging storage platforms, and high-speed network fabrics.

At Kao Data, we see AI infrastructure expertise and enterprise AI delivery increasingly converging around orchestration, workload management and infrastructure control.

In practice, AI strategies now require much tighter integration between facilities engineering, IT operations, orchestration software and workload scheduling than traditional environments ever demanded.

At Kao Data, we increasingly see AI data centres as engineered production environments optimised for the efficient delivery of AI workloads. Real estate, power and connectivity remain fundamental design constraints, but the effectiveness of these environments now depends equally on how well they integrate compute density, cooling performance, orchestration and operational efficiency.