We optimised the hardware. We forgot the software. It's time to build better by design.
Guest blog by Luc Burnip, Architecture Lead at DXC.
AI's environmental impact has become impossible to ignore. Data centres consumed approximately 415 terawatt-hours globally in 2024, with projections suggesting this could more than double to 945 TWh by 2030—equivalent to Japan's entire electricity consumption. Yet amid the headlines about power-hungry GPUs and water-cooled server farms, we're collectively missing a crucial piece of the puzzle: IT equipment accounts for 40-50% of a data centre's energy consumption—and software design choices ultimately dictate how much of that energy is productive versus wasted.
Hardware efficiency has come a long way. The rise of ARM-based processors in the data centre—pioneered by AWS Graviton chips since 2018—has delivered dramatic improvements in performance per watt, with Graviton instances using up to 60% less energy than comparable x86 alternatives. Hyperscale operators have driven PUE ratings down to 1.09–1.2, and the GPUs powering AI workloads are 99% more efficient than their 2008 counterparts.
Regulators have taken notice of hardware. Decades of policy have produced extensive efficiency standards—from ENERGY STAR-certified servers to EU Ecodesign requirements and mandatory PUE reporting. Yet no equivalent standards exist for software. We've regulated how much energy a server can draw, but nothing stops inefficient code from running that server flat-out doing work that could be accomplished in half the cycles. This regulatory blind spot represents one of the most significant untapped opportunities for meaningful carbon reduction in the age of AI.
The Software Sustainability Blind Spot
When organisations assess their IT carbon footprint, they typically focus on infrastructure: server specifications, cooling systems, and renewable energy procurement. But as the Green Software Foundation notes, if the software generated is not as efficient as hardware advances, overall energy consumption remains high, negating the entire goal of building green data centres.
Software inefficiency manifests in countless ways: bloated legacy applications consuming resources unnecessarily, poorly optimised algorithms running millions of times daily, redundant data processing, and user interfaces that demand excessive compute on every page load. Consider a government digital service accessed thousands of times per day. Every unnecessary API call, every unoptimised database query compounds into significant energy consumption at national scale.
Modernisation as a Sustainability Strategy
Legacy software presents both a challenge and an opportunity. Many organisations operate applications-built decades ago, designed without consideration for energy efficiency. These systems often run continuously at low utilisation, wasting both compute resources and electricity.
Outdated infrastructure management compounds the problem. Traditional approaches to meeting SLA targets often rely on over-provisioning—keeping machines idling "just in case." In a world of auto-scaling and predictive load management, this precautionary waste is increasingly unnecessary.
Modernising to cloud-native, microservices-based architectures delivers substantial sustainability benefits. Containerised applications can scale dynamically with demand, eliminating the energy waste of always-on infrastructure provisioned for peak loads. When combined with hosting in renewable energy-powered hyperscaler regions, AWS and Microsoft research suggests carbon footprint reductions of up to 98% compared to traditional on-premises deployments.
Architecture decisions made today will determine energy consumption for years to come. Large enterprise systems typically remain in place for a decade or more; government and financial services platforms routinely operate for 25 years and beyond. A slightly inefficient algorithm doesn't just waste resources today—it compounds across millions of transactions over decades. In cloud environments, where costs are consumption-based, this compounding becomes financially visible. Sustainability, like security, needs to be better by design.
Building a Culture of Software Sustainability
Technology alone won't solve this challenge. Sustainable software requires cultural transformation within product and engineering teams. Just as FinOps practices have brought cost consciousness into development workflows, GreenOps must embed carbon awareness into every stage of the software lifecycle.
This starts with measurement and accountability. Assign carbon budgets to product teams alongside their cost and performance targets. When teams understand the environmental impact of their design decisions, they make different choices.
Frameworks and standards provide essential scaffolding for this cultural shift. AWS's Well-Architected Framework now includes a dedicated Sustainability Pillar. DXC's Modern Architecture Framework embeds sustainability requirements into its reference architectures. These aren't optional nice-to-haves—they're becoming baseline expectations for responsible software engineering.
The Green Software Foundation provides resources for writing better code. Their Software Carbon Intensity (SCI) specification—now ratified as ISO/IEC 21031:2024—offers a globally recognised methodology for calculating the carbon emission rate of software systems. This isn't simply another reporting metric; it's a rate-based score (emissions per unit of work) that teams can own and actively reduce.
You Can't Improve What You Can't Measure
Meaningful progress requires visibility. Observability platforms like Dynatrace now incorporate comprehensive carbon footprint capabilities through their Cost & Carbon Optimization tooling. The platform tracks emissions across hybrid and distributed environments, translating key platform and application metrics into energy consumption and CO2 equivalent values.
For organisations running containerised workloads, the Kubernetes-specific capabilities are particularly valuable. The platform calculates energy consumption and carbon emissions at cluster and namespace level, identifying idle pods, underutilised nodes, and scaling opportunities. Teams can filter by optimisation state to quickly identify where resource waste is occurring.
The real power emerges when this data integrates with automation. Dynatrace's AI Operations capabilities enable intelligent, automated responses—orchestrating scaling decisions, right-sizing resources, and optimising workload placement while tracking the carbon impact of these changes.
Green AI: Efficiency by Design
The principles that make software sustainable—efficient algorithms, right-sized resources, processing only what's necessary—become even more critical when applied to AI systems, where energy intensity and data volumes are typically much higher. Training large language models can generate hundreds of tons of CO2, and inference at scale adds continuous energy demand. Yet AI also offers sustainability opportunities—Google reports reducing energy consumption per Gemini prompt by a factor of 33 between 2024 and 2025 through better algorithms and architecture.
The lesson is clear: sustainable AI isn't solely about renewable energy or hardware efficiency. It's about designing models and deployment architectures with energy efficiency as a first-order concern. Smaller, more efficient models deployed thoughtfully often deliver better outcomes than massive models run without consideration for their environmental cost.
Where to Start
The path forward doesn't require a multi-year transformation programme. Start by raising awareness within your technology teams—many engineers simply haven't been asked to consider energy efficiency as a design constraint.
Then get visibility. All three major cloud providers now offer carbon footprint tools—AWS's Sustainability console, Google Cloud's Carbon Footprint, and Azure's Emissions Impact Dashboard. These tools aren't perfect, but they're free, already in your console, and can reveal hotspots you didn't know existed. From there, consider integrating carbon metrics into your existing FinOps practice. If you're already tracking cloud spend by team or service, extending that discipline to carbon is a natural next step.
The software we forgot is too significant to ignore. Every inefficient algorithm, every bloated application, every poorly optimised query represents energy spent on work that could be done with a fraction of the resources. The good news: sustainability improvements start at the drawing board. Event-driven architectures, efficient data models, right-sized compute—these aren't afterthoughts but design decisions that compound in value over years of operation. Sustainable software isn't just good environmental stewardship—it's increasingly good business practice.
U.S. Congress Research Service, "Data Centers and Their Energy Consumption: Frequently Asked Questions" (2025). Computing power and server systems account for roughly 40% of electricity consumption in a data center. https://www.congress.gov/crs-product/R48646
U.S. Government Accountability Office, "Information Technology: IRS Needs to Complete Modernization Plans and Fully Address Cloud Computing Requirements" (GAO-23-104719, January 2023). IRS legacy applications range from 25 to 64 years in age. https://www.gao.gov/products/gao-23-104719
The techUK Cloud Programme ensures the UK stays at the forefront of cloud adoption and is a single point of contact for UK government and key stakeholders on issues impacting the development of the UK cloud market and industry. Visit the programme page here.
Call for contributions: Digital Infrastructure Report
Members are invited to help us champion how digital infrastructure (compute, connectivity and data centres) underpins a broad base of economic activity across the economy.
Our members develop strong networks, build meaningful partnerships and grow their businesses as we all work together to create a thriving environment where industry, government and stakeholders come together to realise the positive outcomes tech can deliver.
