The Race for Power - Meeting the global energy demands of AI data centres

The merits of these structures have been influenced by the scale, power requirements and tenant profile as well as the teams bringing them to market. There is significant interest in developing additional data centres, in particular those driven by the growth in AI. As the scale of development increases, it will become increasingly important to ensure that data centre projects are designed with the likely source of capital and finance structure in mind.

Artificial intelligence is advancing rapidly, reshaping entire industries (including the energy industry, in a virtuous cycle) and accelerating the digitalization and optimization of global economies. Supporting this transformation is an increasingly diverse ecosystem of data centre architectures, including cloud, edge, colocation and the rapidly growing class of AI-optimized data centres.

• Cloud facilities deliver carefully designed environments within which cloud providers can provide scalable compute for everything from enterprise software to global content platforms
• Edge data centres sit close to end-users or devices, enabling ultra-low-latency applications such as EVs, IoT, autonomous transport and industrial automation
• Colocation facilities, meanwhile, offer all forms of enterprise and other organizations a way to deploy their own hardware in shared, professionally managed environments, balancing performance with operational efficiency

Within this landscape, AI data centres stand distinctly apart, and developing the power part of the “powered shell” for AI forms the focus of this paper. Their architectures are built around extremely dense clusters of GPUs, TPUs, and other accelerators designed to train and infer state-of-the-art machine learning models. Unlike cloud or colocation workloads, AI training runs at sustained, exceptionally high utilization, often for days or weeks, creating continuous, large-scale energy demand and extreme rack-level heat densities. They also feature rapid and unpredictable demand peaks that need to be met, which can cause significant strain on local grids and may require on-site mitigations to smooth peak demand. This is driving some projects to seek out on-site BESS solutions to address such technical concerns of
network operators.

The scale of consumption is already reshaping the global electricity system. According to the International Energy Agency’s 2025 report, “Energy and AI”, data centres consumed approximately 415 TWh in 2024, representing around 1.5% of global electricity demand, its base case projects that this will more than double to around 945 TWh by 2030, driven significantly by AI and the public cloud on which AI relies. The IEA’s 2026 follow-up, “Key Questions on Energy and AI”, reports that data centre electricity demand increased by 17% in 2025 and that AI-focused data centres grew faster, in its base case, AI-focused consumption is projected to more than triple to around 465 TWh by 2030, while total data centre demand doubles. Goldman Sachs Research’s October 2025 update projects global data centre power demand (AI and non-AI) to increase by 175% between 2023 and 2030, revising its earlier 165% estimate.

These trends reveal why AI data centres face unique and unprecedented energy challenges. Power demands are no longer measured in tens of megawatts but often in hundreds or even thousands of megawatts, placing strain on grids that were not designed for such concentrated loads. In some cases, the energy requirements for connecting a new AI data centre is comparable to connecting an entire city. Utilities therefore increasingly treat large AI projects not as ordinary commercial connections but as major infrastructure events, requiring long-lead transmission upgrades, transformer installations and grid-stability planning.

The focus of this paper is on how the two worlds of AI infrastructure and power infrastructure meet in a way that is not only financeable for both, but also economically deliverable. This is partly about delivering absolute value through lower cost, faster deployment, and international scalability and standardization. It is also about delivering predictability of costs, risks and revenues so that the optimal capital stack can be deployed. Achieving this is no mean feat, but AI is not only a source of unprecedented power demand (a problem to be addressed), it is also a key driver of innovation and acceleration of the energy transition (an opportunity to be grasped).

Request the full paper