Written by: Paolo Rizzi
As part of CAT’s Responsible AI programme in collaboration with ustwo, we welcomed
Professor Chris Preist for a session exploring the environmental impact of artificial
intelligence. Chris, Professor of Sustainability and Computer Systems at the University of
Bristol, challenged us to look beyond AI’s capabilities and consider the infrastructure,
resources, and policies that will shape its future.
As AI adoption accelerates, questions about its sustainability are becoming increasingly
important. While public discussions often focus on individual interactions with AI tools, Chris
encouraged us to think about the much larger systems behind them.
Every AI interaction relies on a network of data centres, specialised hardware, cooling
systems, telecommunications infrastructure, and electricity grids. As demand for AI grows,
so does the need for the infrastructure required to support it.
One of the central messages of the session was that electricity consumption is the standout
environmental concern. Training and operating advanced AI models requires significant
computational power, which translates directly into growing energy demand. While other
impacts, including water use and hardware production, are important considerations,
electricity remains the key factor in understanding AI’s environmental footprint.
At the same time, Chris emphasised that the picture is more complex than simply “AI uses a
lot of energy”. Digital technologies have historically become more efficient over time, and AI
is no exception. Advances in hardware, model design, and computing efficiency continue to
reduce the resources required to perform a given task.
This creates an important tension. AI demand is increasing rapidly, but the technology is
also becoming more efficient. The challenge is understanding whether those efficiency gains
can keep pace with continued growth.
The session also explored where AI computation happens. Today, most AI services rely on
large, centralised data centres operated by a small number of technology companies.
However, as models become smaller and more efficient, some AI workloads may
increasingly move onto personal devices or local infrastructure.
Chris highlighted that this shift could bring both opportunities and challenges. Large
providers are often able to measure and report their environmental impacts and invest in
efficiency improvements at scale. If AI becomes more distributed across millions of devices,
tracking its overall environmental footprint may become more difficult. It could also shift
some of the costs of AI infrastructure from technology companies to consumers through
increased electricity use and hardware requirements.
A particularly interesting part of the discussion focused on the role of government. Chris
argued that the future environmental impact of AI will not be determined by technology
companies alone. Decisions about energy systems, infrastructure development, and public
policy will play a significant role in shaping outcomes.
The same AI system can have very different environmental impacts depending on how the
electricity powering it is generated. As a result, government decisions about grid
decarbonisation and infrastructure planning are likely to be just as important as technological
innovation itself.
The session ultimately offered a balanced perspective. AI has enormous potential, but
understanding its environmental implications requires us to look beyond individual tools and
consider the wider systems that support them. The future of AI will depend not only on
technological progress, but also on the choices we make about how it is built, powered, and
governed.
A recording for this talk is accessible in the #greener-ai channel in CAT Slack community.