If you watch the news, you’ve probably seen the growing controversy around data centres, with local communities pushing back against new developments and governments in some countries beginning to regulate the industry. Yet despite all this attention, we’re becoming better and better at making computing feel almost weightless. Calling it “the cloud” certainly hasn’t done us any favours, because it makes the whole thing sound almost ethereal, which is pretty much how we experience technology today. You click a button or ask an AI model a question, and something happens almost magically, instantly.
Of course, we know there’s a very physical system behind all of this, with machines consuming electricity, generating heat that needs to be managed, and relying on data centres, networks and hardware that require resources to build and operate. But it’s easy to forget about those consequences when we’re using technology, and perhaps even easier when we’re designing and building it. While we’re becoming more aware of the environmental footprint of digital products, the people making decisions about how those products work rarely have the information they need to understand how their choices contribute to it. And I think this is one of the most interesting challenges facing digital sustainability today.
Traditional carbon accounting helps us understand what happened by collecting data, calculating emissions and reporting on performance, often months after the decisions responsible for those emissions were created. By then, an infrastructure choice may have been locked in, an AI model deployed or an inefficient process repeated millions of times, leaving us to work out retrospectively what could have been done differently.
This is what we explored during a recent Responsible AI session organised by ClimateAction.tech in partnership with ustwo, where I was joined by James from GreenPixie and Mark from OVO. We looked at how environmental information could become more useful if it were available while decisions are being made, and how that might change the way we design, build and operate digital products.
GreenPixie: making infrastructure impacts visible
James Hall, Head of GreenOps and Tokenomics at GreenPixie, explained how they’re working to make energy consumption, carbon emissions and water use as accessible and granular as the cost information organisations already receive from cloud providers. GreenPixie helps organisations compare the impact of different infrastructure configurations, computing resources and AI models by connecting cloud usage data with environmental modelling, giving engineers the information they need to make more efficient choices.
This is an excellent example of how to make environmental performance part of the same conversations where cost, security and performance are already considered. This helps organisations move away from treating sustainability as a separate concern to be examined only once the decisions have been made. Check out Greenpixie’s work here.
OVO: turning environmental data into operational improvements
Mark Buss, Senior Digital Sustainability Manager at OVO, shared how they use carbon data to identify inefficiencies across its cloud infrastructure and better understand the footprint of its AI tools. He brought some great examples that demonstrated how greater visibility can help teams identify practical opportunities to reduce resource consumption, from unnecessary computing activity to inefficient database queries. OVO is also working with GreenPixie to estimate the impact of different AI tools and models, combining the usage information available from providers with environmental data to understand where emissions are coming from and support conversations about how to reduce them.
ustwo: bringing carbon intelligence into the developer’s workflow
At ustwo, we’ve been exploring what happens when environmental information reaches developers while they’re working. Together with computer science students and academics at the University of Bristol, we developed PRISM, an open-source extension for Visual Studio Code that estimates the energy consumption and carbon emissions associated with AI activity and makes those estimates visible directly within the development environment.
PRISM allows developers to better evaluate which models to use and for what, explore their estimated impact over time and receive immediate feedback on individual interactions, helping them recognise patterns and question whether different approaches might reduce unnecessary resource consumption.
Making sustainability a more important part of the decision making process
Sustainability reporting has been an important first step in helping us understand the carbon footprint of digital technology, but environmental data becomes far more useful when it reaches the people who can make decisions that impact that footprint at the right time.
Of course, we need to recognise that the digital product industry and individual developers can only influence part of the picture. Cloud and AI providers have a responsibility to build and operate their infrastructures responsibly and disclose more meaningful environmental data so that we can actually figure out where the impact comes from and what needs to be done to reduce it. But it’s also important that organisations building digital products consider environmental performance alongside cost, functionality and user experience.
In all of this, we need to keep in mind that some of the most consequential decisions happen long before anybody starts writing code, when we’re still deciding what to build, whether a product needs AI in the first place and if the value it creates justifies the resources required to deliver it.
End Notes
There is a real opportunity to make environmental information part of everyday decisions, allowing people to understand the consequences of different options while they still have the chance to influence them. If we can make that shift, carbon intelligence could become a standard part of how we define quality in digital products, and a practical part of how we design, build and operate them, helping us prevent unnecessary impacts.
A full recap of the presentation will be available soon.
Thanks to ustwo for their continuing support and partnership with CAT on the Responsible AI series. Keep an eye on our #announcements channel in CAT Slack and on our newsletter for further events.