Home > Technology Articles > Green Computing: Tech's Growing Energy Problem
Green Computing: Tech's Growing Energy Problem
Every search query you run uses about 0.3 watt-hours of electricity. A single ChatGPT query uses the same. Multiply that by billions of queries per day, add the servers running them, the cooling systems keeping those servers from overheating, and the infrastructure connecting everything — and you start to get a sense of what the computing industry's energy footprint actually looks like. Global data centres consumed an estimated 460 TWh of electricity in 2025 — roughly 1.8% of total worldwide electricity demand — and the IEA projects that figure will surge past 800 TWh by 2028. That's an electricity footprint now rivalling the annual consumption of France.
Electricity demand from data centres soared by 17% in 2025, with AI-focused data centres climbing even faster — well outpacing growth in global electricity demand of 3%. Green computing exists because the computing industry recognised, earlier than most sectors, that this trajectory was unsustainable — and that the solutions had to come from within the technology itself.
What Green Computing Actually Is
Green computing — also called green IT or sustainable computing — refers to the design, manufacture, use, and disposal of computing resources in a way that minimises environmental impact. That covers a surprisingly wide range of activities: how chips are fabricated, how data centres are cooled, how old hardware is disposed of, how software is written, and how renewable energy is sourced for the systems running it all.
The term gained formal recognition in 1992, when the US Environmental Protection Agency launched the Energy Star programme — initially focused on reducing energy consumption in computer monitors and CPUs, which at the time were left running continuously, consuming power even when idle. Energy Star became the first widely adopted green computing standard and the template for efficiency labelling that spread across consumer electronics globally.
The Real Problem — Where the Energy Actually Goes
Most people picture a data centre as rows of blinking servers. What they underestimate is the cooling infrastructure behind those servers. Cooling systems consume 30–40% of total data centre power, making them the second-largest electricity consumer after the IT equipment itself. Maintaining optimal temperatures — typically between 68–77°F — requires continuous active cooling, and in early data centres, that cooling was deeply inefficient.
The metric used to measure data centre efficiency is Power Usage Effectiveness (PUE) — total energy consumed by the facility divided by the energy used by the computing equipment itself. A PUE of 1.0 would be perfect: every watt going into the building powers computation. A PUE of 2.0 means half the energy is lost to overhead, primarily cooling. Average PUE improved from 2.5 in 2007 to 1.55 in 2022 across the industry — representing genuine progress, but also demonstrating how far baseline efficiency has come.
Amazon's data centres reported an average PUE of 1.14 in 2025, better than both the public cloud industry average of 1.25 and most on-premises enterprise data centres. These figures are achieved through liquid cooling, hot aisle/cold aisle containment, outside air economisation, and AI-driven cooling management that adjusts cooling loads dynamically based on real-time server conditions.
Key Strategies in Green Computing
Renewable Energy Procurement
The most direct path to reducing carbon emissions from computing is switching the electricity source. The tech sector accounted for around 40% of all corporate power purchase agreements for renewables signed in 2025 — making technology companies one of the primary drivers of renewable energy expansion globally. By the end of 2024, Amazon had matched 100% of the electricity consumed by 24 AWS data centre regions with renewable, carbon-free energy sources, achieving its 2025 goal five years ahead of its original 2030 target.
Server Virtualisation
Traditional servers ran at shockingly low utilisation — a McKinsey study found typical utilisation at around 6%, meaning 94% of a server's capacity was idle but still consuming power. Virtualisation — running multiple workloads on the same physical hardware — dramatically improved this. Modern hyperscale cloud environments run far higher utilisation rates than enterprise-owned server rooms, which is one of the genuine environmental arguments for cloud migration over maintaining on-premises infrastructure.
Hardware Efficiency
Modern AI chips use nearly 99% less power to perform the same computations as a model from 2008. That improvement is one of the few bright spots in the data centre energy story — efficiency per computation has improved dramatically, even as the total number of computations demanded has grown faster than efficiency gains. The net result is still rising absolute energy consumption, but the counterfactual — running today's AI workloads on 2008 hardware — would be catastrophically worse.
Electronic Waste (E-Waste) Management
Green computing isn't only about operational energy — it covers the full hardware lifecycle. Manufacturing a computer requires significant energy and rare earth materials; disposing of one improperly releases toxic materials including lead, mercury, and cadmium. E-waste is now one of the fastest-growing waste streams globally. Extended producer responsibility schemes, device refurbishment programmes, and circular economy design — where components are designed to be recovered and reused — are all part of the green computing picture.
Software Efficiency
A program that performs the same task in fewer operations uses less energy. Green software engineering — measuring and minimising the carbon footprint of code — is an emerging discipline with its own tools, the Green Software Foundation (founded in 2021 by Microsoft, GitHub, ThoughtWorks, and others), and increasingly, organisational reporting requirements. This layer is consistently deprioritised in software development, where developer time is expensive and compute time appears cheap — but at scale, inefficient code translates directly into avoidable energy consumption.
Where AI Makes the Problem Harder
The single biggest complication in green computing right now is AI. Electricity consumption from data centres is set to double by 2030, and power use from AI-focused data centres is poised to triple.
Google disclosed that its global data centre operations consumed approximately 6.1 billion gallons of water in 2024, a 20% increase year-over-year driven primarily by AI infrastructure expansion. Water, not just electricity, is increasingly a constraint — data centres in water-stressed regions are drawing from local supplies that communities depend on.
The pipeline of conditional offtake agreements between data centre operators and small modular reactor (SMR) nuclear projects has grown from 25 gigawatts at the end of 2024 to 45 gigawatts — a sign that renewable energy alone may not scale fast enough to meet data centre demand, and that the industry is turning to next-generation nuclear as a complement.
Why Green Computing Matters Beyond the Tech Sector
There were 5,426 data centres in the US alone as of March 2025, and the number is skyrocketing. About 56% of the electricity used to power US data centres comes from fossil fuels. Data centres' projected US electricity demand by 2030 is set to increase to up to 130 GW — close to 12% of total US annual demand.
The computing industry's energy problem is not a niche concern for data centre engineers. It's a mainstream energy policy issue, a water management issue, a land use issue, and a climate issue. Green computing is the set of technical, design, and policy responses to that problem — responses that have already delivered measurable improvements in efficiency, and will need to deliver far more as AI, Edge Computing, and continued digital expansion drive demand higher through the rest of this decade.
Frequently Asked Questions (FAQs) - Green Computing: Tech's Growing Energy Problem
Q1. Why does green computing matter if AI keeps getting more efficient every year?
Efficiency gains don't offset scale. AI chips use nearly 99% less power per computation than 2008 hardware, but total demand grows faster than efficiency improves. Data centre electricity demand is set to double by 2030, with AI facilities tripling usage in that window.
Q2. What is PUE and why do data centre companies keep talking about it?
PUE measures total facility energy versus energy actually reaching computing equipment; 1.0 is the theoretical ideal. Industry average improved from 2.5 in 2007 to 1.55 in 2022. Top performers like Amazon now report figures near 1.14 through liquid and AI-managed cooling.
Q3. Is water usage really a bigger concern than electricity for data centres?
Not bigger, but increasingly just as constrained. Google alone reported roughly 6.1 billion gallons used across its global data centres in 2024, up 20% year-over-year, driven largely by AI infrastructure. In water-stressed regions, this competes directly with local community supply.
Q4. Does moving workloads to the cloud actually help the environment?
Generally yes, though it depends on the provider. On-premises enterprise servers run notoriously low utilisation — one McKinsey study found around just 6%, meaning most capacity sits idle while drawing power. Hyperscale providers achieve far higher utilisation through virtualisation, a genuine sustainability argument for migration.
Q5. Where does nuclear power fit into green computing?
It's becoming a bigger part of the answer than expected. Renewable energy alone hasn't scaled fast enough to match data centre demand, so operators are turning to next-generation nuclear. Offtake agreements with small modular reactor projects nearly doubled in a year, from 25 to 45 gigawatts.