How IT can get more sustainable quicker: notes from Green IO

At Green IO this week, I heard about offshore wind, sustainability data, and cloud certification. This post has some notes I took. They all in some way relate to what can get us to more sustainable IT quicker.

Better sustainability data

It's really hard to reason about your organisation's impacts without data. Hannah Smith (virtual) and Oliver Cronk (in-person) talked us through a new Green Web Foundation community project that helps with better data: Digital Impacts Schema and Taxonomy (“DIST”).

DIST has three parts:

• a taxonomy (ensures you can compare “apples to apples” by categorising impacts in upstream impacts like software and hardware, operational impacts like employee devices and cloud services, and downstream impacts like end-user devices),
• methodologies to measure, calculate, and estimate impacts, tagged with DIST categories,
• a structured data format to exchange the collected data in a standardised and machine-readable manner.

So it answers what data to collect, how to collect it, and how to describe it once you've got it, all in one system.

Offshore wind

In "Can offshore wind power the AI boom?", professor Susan Gourvenec of the University of Southhampton talked to us about offshore wind. She debunked some myths, like how much space they use: only a very small proportion compared to other bits. Do they kill birds? Yes, but cats are their main killer by a long mile. Is it expensive? The cost of delaying decarbonisation is much higher, both in terms of economic (productivity) and social costs. Aren't the materials environment unfriendly? Yes, but greener steel is in the works. My main takeaways: offshore wind takes relatively little space (in relation to other offshore activities), there is a lot of it possible, and it could be enough to power the AI boom.

Measuring internal AI emissions

Mark Buss of OVO Energy shared how they ended up measuring their internal AI usage. OVO give everyone who works there “enterprise level productivity tooling” and engineers access to Claude, i.e. they use a lot of it. Some of the staff started asking about the impact of that usage, people wanted to know for a myriad of reasons. So OVO ran a series of in person workshops, focused on environmenral impacts of AI, making space for both positives and negatives. There's many ways to look at it, Mark explained, saying that “you can't decarbonise a cow but you can lower emissions from data centres”. He emphasised context: when calculated all their AI use, and compared it to emissions associated with other processes at OVO, and saw it was only marginal compared to cloud cost (a bit more), business travel (much more), and the fossil part of the energy they sell (magnitudes more).

Standardising energy use of streaming

Marisol Palmero of IEEE got us up to speed with the standardisation of energy-aware streaming. She explained that only in 2025, AI energy usage overtook that of streaming services like Netflix and Disney+. But they are quite different: most energy use of streaming associated with the user's devices (battery needed to power graphics on screens etc), whereas most energy use of AI is associated with data centres. Marisol explained measuring emissions associated with streaming is increasingly less voluntary: there's more preasure from regulation (CSRD, SEC Climate Disclosure Rule, UK SDR) and Omnibus 1 means evidence will be required. It's tricky as standards around this aren't interoperable yet. IEEE works on Trustworthy Energy Attribution & Measurement (TEAM) to help solve for this. And such efforts can have an impact: “standards are how we move individuals to collective action”, Marisol said.

A pipeline GreenOps observability

Diana Todea of VictoriaMetrix and Julien Nioche of DigitalPebble discussed how we already have the data we need for GreenOps. We all know that cloud providers give data that isn't very useful or actionable, Julien explained. Open Source projects can help, including SPRUCE, which can come third-party data, is actively maintained, and available free. Diana showed how she used SPRUCE and a number of other tools to build a pipeline for observability of GreenOps including a cool dashboard, and in general, the importance of observability. What's promising in that regard is that the Green Software Foundation (GSF) are taking their Software Carbon Intensity (SCI) to OpenTelemetry through semantic conventions.

Certification for data centres

“Please raise your hand if you fully understand the environmental and social impact of your cloud usage and were aware at the time of procurement”… tricky questions were asked during the session by Emily Watson-Breeze and Barton Finn of TCO Certified, who built a certification for “cloud resources” over the last few years to answer questions like that. So not data centres per se, but the cloud services that run in them. It looks at requirements for physical layer (data centre facility), IT hardware layer (servers, storage, network equipment), and operations & management (certified cloud resource). And it even looks at the very common case where cloud resources are offered as a service: you're not buying a server, but paying to use it for a specific time. It's really hard because there is hidden structure, a complex value chain, and data that is hard to compare due to lack of standards and incomplete datasets. This kind of certification can help both procurers and suppliers. I cannot wait for hyperscalers to get on this, and until they are, I guess we should all be pushing for that when procuring cloud products.

Wrapping up

It wasn't because before lunch started, the host announced “the afternoon will be all about AI at scale” that I missed taking notes for a lot of the afternoon sessions: I ended up having some great hallway chats and left a little earlier to get on my train back.

I enjoyed meeting folks at Green IO and learning about a wide variety of topics. I even got to hang out a little bit with people from the ClimateAction.Tech community, which unites tech workers that try to change the system to face the climate crisis (there's a very active Slack instance you can join if you're interested!).