The Canadian government recently launched its Responsible Data Centre Development Principles, enabling the country to set the terms for responsible growth of artificial intelligence computing power and storage.
The principles set five expectations for new data centre projects: to create lasting local benefits, protect Canadians from extra electricity costs, minimize water use and environmental impacts, be transparent about local impacts and bring strategic value to Canada.
Read more:
Data centres: Opposing them and developing them are not the only two choices
Meanwhile another part of the story remains out of view: data centres powering AI also need servers, processors, storage devices, networking equipment and cooling systems.
What will become of all this hardware? This question deserves the same scrutiny as the electricity and water consumed throughout the life of these centres.
Before a server is switched on
There is good reason for the focus on energy in the AI infrastructure debate. International Energy Agency projections show that electricity consumption by data centres will increase from about 415 terawatt-hours in 2024 to more than 945 TWh in 2030. This is a 15 per cent annual growth in demand. More powerful servers, driven by AI adoption, account for half of this increase.
Nevertheless, electricity consumption is only one part of the life cycle of computing equipment. Before a server reaches a data centre, raw materials have been extracted, processed and turned into manufactured components. At the end of its operational life, the remaining value of the server depends on the ability to repair, upgrade, reuse, disassemble or recycle it.
Generative AI, particularly large language models, could create between 1.2 million and five million tonnes of accumulated electronic waste between 2020 and 2030, according to several scenarios developed in a 2024 study published in Nature Computational Science.
(Unsplash)
The study identifies the “circular economy” as a strategic way to reduce this type of waste by 16 to 86 per cent.
Otherwise, how will this enormous waste stream be absorbed? According to the Global E-waste Monitor, 62 million tons of e-waste were generated globally in 2022 and only 22.3 per cent of this was documented, collected and recycled in an environmentally responsible manner.
The circular approach
The circular approach goes beyond recycling. At the earliest stages of the supply chain, it asks how long equipment and its components keep their useful value.
Recent data-centre equipment research shows it’s possible to develop modular designs and standardized, repairable and interchangeable parts — to refurbish or update servers and recover components for other uses rather than throwing them away.
Longer use may seem like the best environmental choice, although older servers are known to consume more electricity to perform the same computing tasks. On the other hand, replacing older servers in an effort to reduce operational energy has the potential to create new impacts through manufacturing.
Circular rebound effects
I published a conceptual study this year on circular rebound effects, examining the problem from another perspective. The rebound lens reveals how circular strategies interact with infrastructure, governance, market expansion and demand patterns, rather than focusing simply on loss of efficiency.
Counting what is reused or recycled is not enough if the broader system continues to increase demand for new resources.
(Unsplash)
A systematic review of 247 studies found that recycling, reuse and sharing often reduce demand for new resources, but rarely on a one-for-one basis. Recycling one tonne of metal, for example, does not necessarily prevent one tonne of virgin metal from being extracted. Reusing one server will reduce the impact of that one unit, but may not prevent the purchase of a new one.
This partial offsetting of the expected environmental benefit is known as a rebound effect.
For AI infrastructure, this distinction is important to grasp. If reused servers and recovered materials only supplement an expanding computing capacity, circular practices could reduce the impact of each unit while the total material demand continues to grow.
Circular practices should therefore be assessed not by how much equipment is reused or how much material is recovered, but by whether they reduce overall purchases of new servers and the extraction of virgin resources.
The responsible AI debate
Canada’s new Responsible Data Centre Development Principles provide for the monitoring, measurement and reporting of environmental impacts. Unfortunately, they do not go far enough to explicitly address equipment life-cycles, electronic waste or the fate of retired servers.
The principles recognize that data centres have consequences for electricity systems, water resources and host communities. They should also consider hardware, repair and upgrades, reuse rates, component recovery and the verified destination of equipment at the end of its life.
Rachida Bouhid does not work for, consult, own shares in or receive funding from any company or organisation that would benefit from this article, and has disclosed no relevant affiliations beyond their academic appointment.
