Foreword

The next phase of artificial intelligence (AI) depends on where and how data centre infrastructure is built. These facilities do not operate in isolation; their long-term viability is linked to the capacity of surrounding energy, water, and infrastructure systems. Sustainability and resilience are no longer just compliance obligations but fundamental conditions for business success and competitive advantage. This playbook provides a practical framework for developers, operators, customers, investors, and policy-makers to navigate the rapidly expanding footprint of AI infrastructure.

Executive summary

AI infrastructure expansion requires unprecedented investment, with global data centre spending potentially reaching $7 trillion by 2030. Electricity demand is projected to grow by more than 20% annually through 2030. Projects poorly matched to local resource conditions face higher costs and development delays. However, adopting sustainability and resilience measures could unlock between $700 billion and $1 trillion in total planned investment by 2030 by reducing project cancellations and improving community relations. This report introduces four tools: the development pathway, universal decision principles, market constraint profiles, and stakeholder recommendations.

Why sustainability and resilience are critical in planning data centres

The shift from general-purpose cloud infrastructure to AI workloads has fundamentally changed infrastructure demands. Training new AI models requires specialized cooling and substantial electricity, with rack densities increasing from 30-100 kW to expected levels exceeding 600 kW. These requirements extend dependencies to critical minerals, semiconductor manufacturing, and skilled labour.

Six strategic constraints shape long-term viability: electricity, water, cooling, community acceptance, land, and regulatory complexity. Global data centre electricity consumption could more than double between 2025 and 2030, reaching 3% of projected global use. Water footprints are also significant; projected 2030 electricity consumption could require 9.3 trillion litres of water annually for power generation. Currently, 45% of global data centres are in water-stressed areas. Community opposition is an increasing risk, with 75 US projects representing $130 billion in investment blocked or delayed in the first quarter of 2026 alone. In South Korea, 17 of 33 projects with permits experienced delays due to community pushback in 2026.

Decision framework for sustainable and resilient data centres

Decisions affecting sustainability and resilience are often “locked in” early in the development pathway. The lifecycle of a data centre involves multiple stages: site selection, design and construction, sourcing equipment, network connectivity, operations, usage, and finally refurbishment or decommissioning. Site selection is particularly critical as it establishes the long-term environmental, infrastructure, and regulatory context. The framework emphasizes that facility-level optimisation is insufficient; stakeholders must consider how data centres interact with broader regional systems.

Universal decision principles and recommended metrics

Six universal principles should guide decision-making: evaluating sustainability over the full lifecycle, assessing future scenarios alongside current conditions, considering system and social implications, preserving optionality and adaptability, using context-aware metrics, and considering the implications of connectivity infrastructure. For example, preserving optionality includes designing facilities with reserve space and piping to allow a transition from air cooling to liquid cooling as rack densities increase.

Recommended metrics are categorised as foundational, emerging, or aspirational. Foundational metrics include Power Usage Effectiveness (PUE) and Water Usage Effectiveness (WUE). Emerging metrics include Hourly Carbon-Free Energy Share (CFE), Water Usage Impact (WUI), and whole-life carbon. Aspirational metrics focus on social and nature outcomes, such as community benefit index scores and biodiversity net gain (BNG). Currently, approximately three-quarters of these recommended metrics are not widely adopted in public reporting.

Market constraint profiles: why local context changes the answer

There is no universal blueprint for sustainable data centres. Local context determines which constraints are most material. The report identifies six market profiles: heavily constrained hub markets (e.g. Northern Virginia), policy-shaped established markets (e.g. Amsterdam, Seoul), hot resource-constrained growth markets (e.g. Texas, UAE), urban infrastructure-constrained markets (e.g. London), land and power bottleneck markets (e.g. Frankfurt, Tokyo), and grid-limited expansion zones (e.g. Nordic markets). For instance, in hot growth markets, cooling demand drives a critical trade-off between power and water use.

Stakeholder playbook and recommendations

Coordinated action across the ecosystem is essential. Developers are encouraged to compare candidate sites using holistic frameworks like the TNFD LEAP approach and engage stakeholders early. Operators should report against common metric frameworks and optimise operations for local constraints, such as shifting flexible workloads during periods of high grid stress. Enterprise customers can influence the market by incorporating sustainability criteria into procurement and avoiding over-consuming high-intensity compute. Financial institutions should link financing terms to continuous performance trajectories, while policy-makers should move towards regional resource planning and context-aware siting.

Call to action

The decisions made today will shape the environmental and social impacts of AI for decades. To future-proof infrastructure, sustainability and resilience must be embedded from the outset. Organisations that integrate these considerations into investment and design will be better positioned to navigate resource constraints, protect asset value, and meet evolving stakeholder expectations.