20251121 Things I learned working in digital sustainability
For most of the last year I have been working at the University of the Arts London to develop its first digital sustainability strategy. UAL is Europe’s largest specialist art and design university. It has around 25,000 network users and a large and varied digital estate that includes laptops, AV kit, media workstations, shared graphics hardware and multiple data centres. It is a significant producer and consumer of digital media, which is why the strategic investment in developing a new net zero strategy called for a specific focus on digital emissions.
This is a departure from the original purpose of this blog. Until now I have mostly written about travel, particularly my time hiking the Pacific Crest Trail last summer. That experience was defined by self-sufficiency and an almost total absence of digital infrastructure. This new role could not have been further from that environment. The contrast has shaped the way I think about technology, sustainability and the tensions that arise when contemplating the two.
These are the main lessons I am taking forward:
1: ‘User devices’ are optimised for responsiveness, not efficiency
Most of the devices we interact with directly on a day-to-day basis are built to perform locally rather than minimise energy use. A typical laptop is specified well above the minimum required for median use. Manufacturers over provision compute and graphics performance to ensure responsiveness across diverse workloads.
In energy terms this means much of the capacity embedded in user devices is rarely used. The result is high upfront supply chain emissions and a hardware fleet that becomes obsolete long before it stops working.
2: Centralising compute improves efficiency, but at the cost of transparency and control
Efficiency improves when workloads move away from distributed high spec endpoints and into centralised compute environments. This includes on-premises servers, colocation data centres and public cloud infrastructure. Energy savings are achieved through higher utilisation rates and lower overheads for cooling and power distribution. These factors are captured in metrics like Power Usage Effectiveness (PUE), which is generally lower for professionally operated facilities than for small institutional data centres.
However, efficiency is not the only consideration.
Moving into hyperscale cloud environments reduces your ability to see how specific processes map to actual energy use. Reporting often covers whole regions rather than individual services. It is also difficult to separate the emissions associated with your own processes from the emissions growth driven by hyperscale investments in AI infrastructure. Even if an organisation is not deploying AI it is still part of the aggregate demand picture via its procurement choices.
So while centralisation increases efficiency, it also introduces methodological challenges and reduces agency over energy sources and reporting boundaries.
3: The supply chain is the main driver of emissions and lifespan extension is the strongest lever
Life cycle assessments for user devices vary, but most credible studies place the supply chain footprint for a typical modern laptop at around 200kg of CO2e. Organisations that refresh their fleets every three to four years accumulate significantly more emissions from hardware turnover alone compared to an organisation with a five year refresh policy.
Extending lifespans through refurbishment policies, modular hardware, repairability and principles-based refresh cycles can significantly reduce emissions. If a device designed for a five-year life stays in service for an additional year, the associated lifetime emissions per year fall by approximately twenty percent. Avoiding new procurement where possible and choosing to reuse/refurb where appropriate is the best thing an organisation can do in this regard (conveniently this aligns with reducing cost as well as carbon).
This links directly to #2 above. Lower local compute requirements enable hardware to stay utilised for longer because performance constraints shift upstream.
4: Behavioural barriers matter more than technical barriers
Some of the largest emissions reductions often require user-facing changes such as accepting longer processing times, reducing multi monitor setups or adopting refurbished hardware. These are visible interventions that users rightly or wrongly interpret as losses of convenience or status. The energy and carbon impacts of data centres and networks are largely invisible by comparison.
Technical feasibility is rarely the limiting factor. Success depends on organisational trust, leadership alignment, and the perception that changes are being shared fairly. People need to believe the institution is acting collectively.
5: The goal is responsible digital growth, not minimal digital activity
Digital systems provide value because the marginal cost of delivery is low. But as efficiency improves, consumption grows faster than reductions in per unit energy use. You hear this a lot in the digital sustainability discourse; it is called Jevon’s Paradox. Equipment that may once have been decommissioned can continue to be used indefinitely lower down in the organisation e.g. large format AV screens accumulate in the exhibitions or events teams. We stop asking whether the activity is 100% necessary.
A hypothetical scenario where all digital systems are shut down would not be optimal. Digital technology removes physical barriers and enables remote participation. We can reduce emissions in other areas such as travel, estate development and logistics through digital platforms. Digital sustainability is therefore more nuanced than just minimising energy for tech, but instead optimising for impact across the whole system.
Reflection
After months on the Pacific Crest Trail I rarely thought about the digital world. Yet even in that environment I relied on mapping apps, satellite messaging and online planning tools. Those technologies enabled my journey rather than detracting from it.
The same is true institutionally. If all future growth at UAL came through online only provision, emissions per student would fall significantly. Even if cloud storage and digital media volumes continued to rise, the broader system impact would outweigh it. The dominant footprint would shift from buildings and travel to compute and networks. Managing that shift responsibly requires clear priorities – and for people to be sure about what they value.
No institution can eliminate digital emissions entirely. The aim is to shape digital systems so they support wider decarbonisation goals. The most effective interventions are often invisible (e.g. extending hardware life, virtualising compute workloads, and setting expectations around performance and convenience). The task is to design for sustainable digital development rather than to treat it as an impact to be minimised.
