Earlier this year, the National Engineering Policy Centre (NEPC) – led by the Royal Academy of Engineering in partnership with the Institution of Engineering and Technology, and BCS, The Chartered Institute for IT – released a landmark policy report, Engineering responsible AI: Foundations for environmentally sustainable AI.
This piece of work addresses a critically under-recognised dimension of the artificial intelligence (AI) revolution: the resource use, environmental impact and infrastructure demands of large-scale AI systems. Its findings have direct relevance for building services engineers and infrastructure professionals, touching on data centres, demand for cooling, power supply, materials supply chains and sustainability obligations.
The explosion of AI deployment across sectors has implications that go beyond software. The NEPC report highlights that AI systems and their enabling infrastructure (for example, large language models, computing clusters and data centres) place growing demands on energy, water and ‘critical’ materials, resources that are managed (or should be managed) within the built environment.
For example: data centre cooling and compute loads drive increased energy demand; many data centres withdraw potable water for cooling; and manufacture and end of life of server hardware draw on rare materials, impact e-waste and create supply-chain risk.
From a building services viewpoint, this means mechanical, electrical and sustainability engineers must now view AI infrastructure as part of the ‘services’ system – not just for IT, but for site infrastructure, cooling, power resilience and resource management.
Key recommendations
The NEPC report formulates five ‘foundational steps’ that governments, organisations and infrastructure providers should adopt to mitigate the environmental impact of AI systems. They are: expand environmental reporting mandates; address information gaps across the value chain; set environmental standards for data centres; reconsider data collection, transmission, storage and management practices; and lead the way with government investment.
Expand environmental reporting mandates: require companies, especially those operating large-scale AI compute infrastructure, to disclose carbon emissions and energy, water and material use. Address information gaps in the value chain: ensure developers, operators and users of AI systems have access to data about environmental impact and infrastructure demands. Set sustainability requirements for data centres: embed cooling-water use, critical materials, power efficiency, e-waste recycling into design and operation. Reconsider data collection, transmission, storage and management practices: ask whether large-scale data hoarding, unnecessarily heavy models or redundant compute are justified, given their resource footprint. Lead the way with government investment: encourage low-resource AI, efficient hardware and sustainable infrastructure, and procurement decisions that embed sustainability.Five foundational steps
Implications for building services engineers and CIBSE members
When designing or specifying data centres, cloud hubs or AI-hosting infrastructure, the ‘services’ requirement must include water use (cooling), thermal management, electrical demand profiles, resilience and decommissioning/reuse of hardware.
The business case for AI infrastructure should factor in not just compute cost, but resource cost, environmental sustainability, and cost of power and water. Engineers should support metrics around these factors.
Existing certifications and standards may need revision or expansion to explicitly include metrics around AI compute loads and associated cooling/water demands.
Professional engineers should engage with the ‘digital infrastructure’ agenda as part of sustainability and net zero plans: AI-hosting is a growing part of the built environment. Education, professional development and client briefing must now incorporate an awareness of AI sustainability issues – not just ethical or algorithmic concerns, but the physical/engineering infrastructure side. The NEPC report urges that environmental sustainability should be embedded in education.
Challenges and future directions
While the report makes a strong statement about environmentally sustainable AI, implementation presents challenges. Data centres’ energy- and water-use figures are often opaque; operators may classify them as commercially sensitive. The report warns of the risk that AI growth could undermine other decarbonisation efforts if unmanaged.
A further challenge for the building services sector is that clients often focus purely on IT/compute value, rather than infrastructure resource value. Bridging that gap will require technical specification and effective stakeholder communication.
The NEPC also points beyond the near term: future AI systems may demand orders of magnitude greater computation, implying that current design practices will need to evolve. The report is the first in the Engineering Responsible AI programme and signals further work that is planned on how AI systems can deliver societal benefit while aligning with sustainability.
A blueprint and warning
The NEPC’s Engineering responsible AI report sets a timely and important agenda for how AI deployment must align with environmental sustainability.
As AI becomes more embedded in industry, research and public services, the physical infrastructure that supports it must be designed, operated and managed in a resource-efficient, sustainable way. For engineers tasked with the infrastructure behind the screens, this report offers both a warning and a blueprint.
CIBSE members and building services professionals should explore how the five foundational steps apply to their project scope, and proactively engage in the design of data centres, AI hosting and compute infrastructure systems, with sustainability at their core.