The Future of AI-Driven Data Centres: Navigating Unprecedented Demand and Challenges

Introduction

Everybody is talking about AI; many already use AI: the Data Centre (DC) Industry is feeling the pain to make it happen!

Artificial Intelligence (AI) in its various shapes and forms is revolutionising industries at an unprecedented pace, driving demand to new heights. The future demand for AI capabilities is expected to multiply, presenting unique challenges and opportunities. This paper explores the implications of this surge in demand, particularly for data centres, and the strategies required to meet these challenges.

The Surge in AI Demand

AI’s rapid adoption is creating an additional layer of a demand on top of existing ‘traditional’ requirements which are putting the system through a severe stress test for years already. This additional surge is not just a temporary trend but a long-term shift that will define the industry for years to come. The current challenge lies in ramping up the rate of bringing megawatts (MW) of computing power online to generate revenue, while transitioning from traditional data centres to those designed specifically for AI.

Challenges in Transitioning to AI Data Centres

1)?? Revisiting Existing Projects:

The race to bring AI DC capacities online forces the industry to assess existing projects for their AI capabilities. The further the project has progressed, the higher the impact on cost, time and permitting challenges will be. The formula seems to be simple and straightforward and creates opportunities when projects are still in the design or even feasibility phase. However, in any case the projects must undergo a comprehensive re-assessment to evaluate the impact of AI on all aspects of their design and permits. This time-sensitive exercise aims to identify any feasible modifications that can be implemented to meet AI requirements. For projects in the design and permitting phase, a thorough review is essential to integrate AI considerations, which may necessitate significant changes to both design and permits. Even in the feasibility stage, where the urgency is less pronounced, it remains crucial to revisit land use and other criteria to ensure alignment with AI demands. The most important external factor is the availability of power in time for operation. AI DC’s need more power to the site to fully utilize the zoning plan allowances which creates an intricate interplay between many interdependent aspects which again impact design, cost and time. Ultimately the industry defines the availability of power as the limiting factor overall.

2)?? Understanding AI in Data Centres:

The industry currently lacks comprehensive information on what “AI in data centres” truly entails. Key questions include the performance specifications of new chips, the demand characteristics, and the necessary infrastructure changes to support AI operations.

These not yet fully defined requirements are currently being developed while they are simultaneously being compared to the design templates and engineering standards to produce a clearer picture of what this design for AI entails. It will be a multi-layered riddle for some time until new templates and approaches will become standards.

The questions that form this riddle are what are the key demand characteristics driving the need for AI-driven data centres? What performance specifications do the latest AI chips offer? How should data centre infrastructure be designed to support the operational requirements of AI operations? Does the shift to AI imply that higher processing speeds will necessitate increased power consumption, enhanced cooling solutions and optimized space utilization, all while adhering to stringent sustainability regulations?

3)?? Focussing on purpose-built AI Data Centres:

New technologies can disrupt industries. They tend to drive a ‘think from scratch’ approach which get implemented on a ‘green fiel’. While AI is a new technology in itself, its impact on the data centre market is that of a demand driver. It will nudge the industry to improve in terms of how to build the necessary infrastructure, some technologies will become viable now, as they have been too expensive in the past. This approach recombines currently available technology in a better or more practicable way, but it is not the disrupting killer application yet. Can we see these new technologies beyond the horizon and how quickly can they be implemented at scale?

Turner & Townsend’s Approach

Turner & Townsend are leading the way in assisting clients to navigate these challenges and find solutions to the dilemma in all the above scenarios. This involves addressing the primary question: “What changes are necessary, and what uncertainties exist compared to our current understanding of data centre design without AI?”

Turner & Townsend are currently referencing existing methodologies while anticipating future technical approaches. We believe that the evolution will be gradual rather than revolutionary, with incremental changes shaping the industry. One notable trend is the shift in spatial requirements, where more infrastructural support area (‘grey space’) is needed to support increased density in the data hall (‘white space’).

The technology of liquid cooling is currently at the forefront of discussions, with various approaches being proposed to manage heat at its source and transfer it elsewhere. We are actively assisting clients in understanding the potential impacts of these changes on projects in construction and projects in the design phase. This process often involves making the seemingly impossible possible, as we work with limited information. Part of our work is to explore and discover the answers to our questions collaborating with our clients, designers, and engineers around a table, preferably in person. Please refer to Jesus Aguilera Avila White Paper on this topic: https://www.dhirubhai.net/pulse/impact-liquid-to-liquid-technology-dc-road-market-aguilera-avila-jqxpe/?trackingId=TwigPblqRV2CCCYohlNxIw%3D%3D .

Our proactive quantity surveying approach involves intensive sprint workshops where all disciplines collaborate over a series of meetings. While not all answers can be found immediately, agreed-upon assumptions help bridge gaps and are revisited later. This iterative method is essential for progressing with the best possible scenario at any given moment. Given the interdependence of multiple aspects, changing one part often affects other parts of the building or technical systems.

In addition to considering the design of the building and the engineering of all systems within the data centre, it is crucial to address the site itself and ensure the provision of sufficient power and data connections as part of an integrated ecosystem. What are the limiting factors in this context? Achieving the energy emission reduction goal and reducing the carbon footprint will require significantly more to be accomplished with less power and fewer emissions. Please refer to the interview with Keith Skinner on this topic: https://www.dhirubhai.net/pulse/cost-control-challenges-faced-data-centre-sector-keith-skinner-p7o7e/?trackingId=rzGFH6pJTy6%2Bh%2BQ9C3kgYw%3D%3D ).

Conclusion

The speed at which AI is transforming the data centre industry is breathtaking. Synchronizing this speed with the growth of utilities is crucial. Turner & Townsend are committed to helping clients develop AI design strategies that go beyond obvious approaches, ensuring a seamless transition to AI-driven data centres.

The industry is at a crossroad where we need to build upon the known in an industry which requires fail-safe redundancies, but let’s pioneer into the unknown to get ahead faster. By addressing these challenges proactively and collaboratively, we can make the seemingly impossible possible, paving the way for a future where AI defines the data centre industry.