Green Computing: Cutting the Carbon Cost of Software

Computing's environmental footprint used to be an abstract concern. Then AI arrived, and data-centre electricity demand — flat for a decade thanks to efficiency gains — started climbing steeply. Data centres consumed roughly 1.5% of global electricity in 2024 and the International Energy Agency projects that could double by 2030, with AI the main driver. Green computing is the practice of getting the same work done with less energy, less carbon and less waste. It has become a mainstream engineering concern, not a niche one.

Where the energy goes

A data centre's power splits between the IT equipment (servers, storage, networking) and the overhead needed to run it — chiefly cooling, plus power conversion and lighting. The ratio is PUE (power usage effectiveness): total facility power divided by IT power. A decade ago, 2.0 was typical; hyperscale facilities now run at 1.1–1.2, meaning almost all the electricity reaches the computers. That's why moving a workload from a corporate server room to a major cloud provider usually reduces its footprint even before considering the provider's energy mix.

Inside the IT load, the story has changed. CPUs and storage were once dominant; now GPU and accelerator clusters for AI training and inference are the fastest-growing component, with individual racks drawing 100 kW or more — an order of magnitude above traditional racks, forcing a shift from air to liquid cooling.

Three kinds of footprint

• Operational carbon from the electricity used to run equipment. Depends on both efficiency and the carbon intensity of the grid supplying it.
• Embodied carbon from manufacturing the hardware — chip fabrication is energy- and chemical-intensive. For a laptop or phone, manufacturing is the majority of lifetime emissions; for a server, it's significant and often ignored.
• Water. Evaporative cooling consumes large volumes; a single large facility can use millions of litres a day. Water stress has become a siting constraint and a community issue.

E-waste is the end-of-life version of embodied carbon: over 60 million tonnes a year globally, less than a quarter formally recycled.

Levers at the infrastructure level

• Efficient facilities: hot/cold aisle containment, higher operating temperatures, free-air and liquid cooling, and heat reuse — several European facilities now feed district heating networks.
• Clean supply: the hyperscalers are the largest corporate buyers of renewable power and are signing long-term contracts for nuclear and geothermal capacity to get around-the-clock clean energy rather than annual offsets. The distinction between "matched annually" and "matched hourly" matters and is increasingly reported.
• Carbon-aware scheduling: running flexible workloads — batch jobs, model training, backups — when and where the grid is cleanest. Google shifts compute across time and regions this way; the tooling to do it yourself exists.
• Longer hardware life: extending server refresh cycles from three years to five or six cuts embodied carbon substantially with modest performance cost.

Levers at the software level

Software decides how much hardware is needed. The gains here are often larger and cheaper than facility improvements:

• Right-size and switch off. Idle and over-provisioned cloud resources are pure waste; the same FinOps practices that cut cost cut carbon.
• Efficient AI. Choose the smallest model that meets the requirement; quantise and distil for inference; cache repeated prompts; batch requests. Inference now dominates AI energy use, so per-query efficiency matters more than training-run headlines.
• Efficient code and data. Language and runtime choices, avoiding unnecessary computation, and not moving or storing data nobody reads. Data-transfer and storage growth are quiet contributors.
• Measure. Cloud providers now expose carbon dashboards per account; open tools like Cloud Carbon Footprint and CodeCarbon estimate emissions per workload or per training run. What isn't measured doesn't get optimised.

Policy and pressure

The EU's Energy Efficiency Directive now requires data centres above a size threshold to report energy, water and heat-reuse metrics. Corporate sustainability reporting rules (CSRD in Europe, similar frameworks elsewhere) push emissions accounting — including cloud usage in scope 3 — into audited disclosures. Meanwhile grid operators in Ireland, the Netherlands, Virginia and elsewhere have paused or constrained new data-centre connections because supply can't keep up. Power availability, not chips, has become the binding constraint on AI expansion — which makes efficiency a competitive advantage, not just a virtue.