12. Sustainable Computing and Digital Transformation

Introduction: Digital Progress and Environmental Responsibility

Digital transformation has altered how organisations produce goods, deliver services, govern institutions and interact with citizens. Cloud platforms, mobile networks, artificial intelligence, software-as-a-service and connected devices have increased computational capability while reducing the need for some physical processes. Online collaboration can replace travel, digital records can reduce paper use, and intelligent control systems can improve the efficiency of buildings, transport and energy infrastructure.

These benefits do not make digital technology environmentally neutral. Computers, servers, storage systems, network equipment and consumer devices require minerals, energy, water and industrial infrastructure. Their manufacture generates emissions and hazardous substances; their operation consumes electricity; and their disposal produces electronic waste. Software also influences environmental performance because its architecture determines how much computation, storage, data transfer and device capacity are required.

Sustainable computing therefore concerns the entire life cycle of digital systems. It asks not only whether a system performs its intended function, but also how much material, energy and carbon it requires, who bears the environmental costs, how long its components remain useful, and whether digital transformation produces a net environmental benefit. For postgraduate computer science students, sustainability is both an engineering concern and a strategic, governance and ethical responsibility.

Learning Objectives

After studying this chapter, students should be able to:

1. Explain the environmental life-cycle impacts of hardware, software, networks, data centres and AI systems.

2. Distinguish operational energy, embodied carbon, water use and electronic-waste impacts.

3. Apply green software engineering and energy-efficient architectural principles.

4. Evaluate cloud, edge, data-centre and carbon-aware computing choices.

5. Design sustainable procurement, asset-management and e-waste strategies.

6. Assess environmental claims using suitable metrics, boundaries and evidence.

7. Analyse the managerial, ethical, regulatory and professional implications of sustainable digital transformation.

The Environmental Footprint of Digital Systems

A digital service is best understood as a chain rather than as an isolated application. The chain includes extraction of raw materials, component manufacture, device assembly, transport, deployment, operation, maintenance, upgrade and end-of-life processing. A life-cycle assessment (LCA) examines these stages and identifies environmental burdens such as greenhouse-gas emissions, energy demand, water consumption, land disturbance, toxicity and resource depletion.

A useful distinction is between embodied impacts and operational impacts. Embodied carbon is associated with extracting materials and manufacturing, transporting and installing equipment. Operational carbon results from electricity consumed while equipment is running. The balance varies by product and use pattern. A device used briefly may have most of its footprint before it reaches the user, whereas a heavily utilised server may accumulate substantial operational impacts over many years. Replacing equipment prematurely can therefore increase environmental harm even when the replacement is more energy efficient.

The environmental footprint of electricity depends on its generation mix, location and time. A kilowatt-hour supplied by a low-carbon grid does not have the same emissions profile as one supplied by a coal-intensive grid. However, renewable electricity does not eliminate impacts: generation infrastructure requires materials and land, and water, reliability and transmission constraints remain relevant. Sound analysis consequently reports both energy use and the emissions factors or assumptions used to translate energy into carbon.

Data Centres and Cloud Infrastructure

Data centres contain compute servers, storage arrays, networking equipment, power-conversion systems and cooling infrastructure. Their energy demand includes both information technology (IT) equipment and supporting systems. Power Usage Effectiveness (PUE) is commonly expressed as:

PUE = Total facility energy / IT equipment energy.

A value closer to 1 indicates that a smaller proportion of energy is used by cooling, lighting, power distribution and other overheads. PUE is useful but incomplete: it does not indicate how much useful computation is delivered, whether the electricity is low-carbon, or how much water and embodied material the facility uses. Complementary indicators include carbon usage effectiveness, water usage effectiveness, server utilisation and energy per transaction or workload.

Virtualisation and containerisation can consolidate workloads and improve utilisation. Autoscaling can reduce idle capacity, while efficient cooling, server power management and workload scheduling can reduce consumption. Cloud computing may offer higher utilisation than dispersed organisational servers, but it is not automatically sustainable. Data replication, excessive storage, frequent data movement, poorly configured services and demand stimulated by low marginal prices can offset efficiency gains. Cloud selection should therefore consider workload characteristics, region, service configuration, data transfer, availability requirements and provider transparency.

Networks, Devices and Data Traffic

Communication networks consume energy in access equipment, routers, switches, mobile base stations, optical systems and data centres. Increased video traffic, Internet of Things deployments and machine-generated data can expand network demand. Efficient protocols, caching, content delivery networks, adaptive media quality and local processing can reduce unnecessary transmission. Network efficiency should be evaluated against the quantity and quality of service delivered, not simply against total traffic.

End-user devices create substantial material and social concerns. Smartphones, laptops and embedded systems require minerals and complex global supply chains. Design choices affecting repairability, battery replacement, software support and modular upgrades strongly influence useful life. A device that is technically energy efficient but difficult to repair may have a larger life-cycle footprint than a slightly less efficient device used for several additional years.

Software as an Environmental Control Surface

Software has no physical mass in the conventional sense, but it controls physical resources. An inefficient algorithm can require more processor cycles; excessive polling can prevent devices from entering low-power states; large images increase network and storage demand; and unnecessary logging expands data retention. Software sustainability is thus concerned with resource demand, durability and the environmental consequences of design decisions.

Important software-related concepts include:

• Energy efficiency: useful computation or service delivered per unit of energy.
• Resource efficiency: effective use of CPU, memory, storage, network and battery resources.
• Carbon efficiency: reduction of emissions associated with a given workload, considering location and time.
• Software longevity: ability of software to remain secure, maintainable and useful on existing hardware.
• Green software engineering: integrating environmental considerations throughout requirements, design, development, testing, deployment and retirement.

Energy is not always the only objective. Reducing runtime may increase memory use, and reducing emissions by moving a workload may conflict with latency, privacy or resilience. Sustainable engineering is consequently a multi-objective optimisation problem involving performance, cost, reliability, security, accessibility and environmental impact.

Frameworks for Sustainable Computing Decisions

Life-Cycle Thinking and Scope Definition

Before measuring an impact, an organisation must define the system boundary and functional unit. For example, “one server” is less informative than “one thousand secure transactions delivered over one year.” A boundary may include only operation, or it may include manufacture, transport, maintenance and disposal. Comparisons become misleading when one option is measured cradle-to-grave and another only during use.

A practical assessment follows these stages:

1. Define the service, functional unit and decision to be made.

2. Map hardware, software, data, energy and supply-chain dependencies.

3. Collect activity data such as watt-hours, runtime, storage volume, device numbers and replacement rates.

4. Apply documented emissions, water or material factors.

5. Test assumptions through sensitivity and scenario analysis.

6. Select interventions and monitor actual outcomes.

Carbon-Aware Computing

Carbon-aware computing changes when, where or how a workload runs in response to the carbon intensity of electricity. Flexible batch processing, model training, backups and analytics may be scheduled during periods of lower grid emissions or shifted to regions with cleaner electricity. Carbon-efficient operation differs from energy efficiency: a workload may use the same energy but generate fewer emissions at another time or location.

A carbon-aware scheduler must consider deadlines, data locality, network energy, service-level agreements, security, jurisdiction and rebound effects. It should not shift workloads merely to produce favourable accounting while increasing total energy or transferring environmental burdens elsewhere. Carbon intensity data should be timely, geographically appropriate and accompanied by uncertainty information.

Green Software Engineering Process

Sustainability should enter requirements rather than be added after coding. A green software process may include:

• setting measurable sustainability requirements, such as energy per transaction or maximum data transfer;
• selecting algorithms and data structures for computational efficiency;
• designing elastic, event-driven and right-sized architectures;
• reducing unnecessary polling, duplication, logging and data retention;
• profiling energy and resource consumption in realistic environments;
• testing under representative loads, including idle and peak conditions;
• deploying with autoscaling, efficient caching and carbon-aware scheduling;
• reviewing sustainability during maintenance, upgrades and retirement.

Measurement tools may use operating-system counters, hardware power interfaces, cloud billing data, application telemetry, workload profilers and grid-carbon APIs. Measurements require careful interpretation because virtualised environments share hardware and estimates may be based on models rather than direct observation. A benchmark should state workload, hardware, software version, sampling method and boundary.

Digital Transformation as Socio-Technical Change

Digital transformation is not simply the replacement of paper with software. It changes processes, roles, infrastructure, consumption patterns and organisational power. A digital service may reduce travel but increase device purchases; remote work may lower commuting emissions but increase household energy use and data traffic. This is the rebound effect, in which efficiency gains lower the cost of a service and stimulate additional consumption.

A transformation programme should therefore establish a baseline, identify affected stakeholders, compare alternatives and monitor both intended and unintended effects. Environmental benefits should be demonstrated through evidence rather than assumed from the presence of a digital platform.

Artificial Intelligence and Sustainable Computation

AI systems introduce distinctive environmental considerations. Training and inference require computation, storage and data movement. Large models may require specialised accelerators, cooling and repeated experimentation. Inference can dominate total impact when a model serves millions of requests. Data preparation, hyperparameter search and model retraining also contribute to resource use.

Strategies include selecting an appropriately sized model, transfer learning, pruning, quantisation, distillation, batching, caching repeated results and routing simple requests to smaller models. Efficient hardware and utilisation improve performance, but embodied impacts and hardware replacement must be included. An accurate model that performs unnecessary computation may be less sustainable than a slightly less capable model adequate for the task.

AI creates indirect risks as well. Optimisation systems can intensify extraction, surveillance or consumption, while automated decisions may distribute environmental burdens unfairly. Sustainable AI therefore includes technical efficiency, data governance, transparency, accountability and assessment of whether the application itself produces social value.

Sustainable Procurement, Governance and E-Waste

Procurement decisions often determine environmental performance more strongly than later operational adjustments. A responsible procurement specification can require energy-performance evidence, repairability, long software-support periods, recycled or responsibly sourced materials, secure data deletion, supplier emissions disclosure and take-back arrangements. Total cost of ownership should include energy, maintenance, upgrades, licensing, disposal and risk—not merely purchase price.

Organisations should maintain an asset register linking devices and infrastructure to owners, location, age, condition, software support and planned retirement. Reuse should normally precede recycling, provided that security, safety and performance requirements are satisfied. Secure erasure or cryptographic destruction of storage media is essential before reuse or disposal. Equipment should be sent to authorised recyclers capable of recovering materials and safely handling hazardous components.

Electronic waste contains valuable materials as well as substances that can harm people and ecosystems when processed informally. Exporting obsolete equipment to jurisdictions with weak controls is not a sustainable solution. E-waste policies should address worker safety, traceability, data protection, extended producer responsibility and compliance with applicable national regulations and international controls on hazardous waste movement.

Governance should assign responsibility across information technology, procurement, finance, facilities, legal, security and sustainability teams. Useful organisational measures include carbon budgets, sustainable architecture reviews, supplier audits, environmental objectives in project approval and public reporting of methodology. Green claims must be precise: “renewable electricity matched annually” is not equivalent to “zero environmental impact.”

Applied Case: Carbon-Aware Analytics for a University Cloud Platform

Background

This illustrative case concerns a large university operating a cloud-based learning and research platform. The platform served teaching applications continuously and ran nightly backups, log analysis and research data-processing jobs. Administrators had observed rising cloud costs and storage growth but had not measured environmental performance.

Situation and Analysis

An initial assessment defined the functional unit as one month of platform operation and separated always-on teaching services from flexible batch workloads. Monitoring showed that many virtual machines were lightly utilised, backups were retained indefinitely, development databases remained active outside teaching periods, and research jobs were scheduled at fixed times. The assessment also found that moving all workloads to a different region would increase data-transfer distance and create governance concerns.

The team avoided a simplistic “move everything to the greenest region” approach. It compared energy, estimated carbon, latency, data-sovereignty requirements, resilience and cost. It also examined embodied impacts because a proposed on-premises hardware refresh would replace functioning servers.

Strategy

The university adopted four measures. First, it right-sized virtual machines and introduced automatic shutdown for non-production environments. Second, it applied retention rules and deduplication to backups while preserving legally required records. Third, it scheduled flexible analytics and model-training jobs using regional carbon-intensity signals, subject to deadlines and data-location constraints. Fourth, it introduced an energy-per-job dashboard and required project teams to report workload size, runtime and resource configuration.

The institution also extended the life of suitable on-premises equipment, securely redeployed surplus devices to laboratories and contracted an audited recycler for equipment that could no longer be reused. Procurement templates were revised to request repairability, support duration and environmental data from suppliers.

Outcome

The measures reduced idle resource consumption, avoided unnecessary hardware replacement and made environmental trade-offs visible to technical and academic managers. Not every workload was shifted: latency-sensitive teaching services remained in their designated region, and privacy restrictions prevented relocation of certain datasets. The programme therefore treated sustainability as a constrained engineering decision rather than a single optimisation target.

Learning Points

The case demonstrates the importance of defining a functional unit, distinguishing flexible from critical workloads, including embodied impacts, and integrating sustainability with privacy, resilience and service quality. It also shows that measurement and governance are prerequisites for credible improvement.

Challenges, Limitations and Common Errors

Sustainable computing faces measurement uncertainty. Cloud providers may report aggregated or estimated figures, software teams may lack power data, and emissions factors vary by time and location. Comparing suppliers without common boundaries can produce false precision. Organisations should disclose assumptions and use ranges or sensitivity analysis where appropriate.

A second challenge is optimisation displacement. An application may reduce CPU use but increase network traffic; a smaller device may be replaced more frequently; or a workload may be shifted to a low-carbon region at the cost of additional transmission. Teams should assess the whole service rather than one component.

Common mistakes include treating PUE as a complete sustainability metric, equating renewable-energy certificates with elimination of all impacts, ignoring embodied carbon, retaining unlimited data, measuring only peak performance, and purchasing efficient equipment without a repair and retirement plan. Another mistake is assuming that digitalisation is automatically beneficial. Baselines and counterfactual alternatives are necessary.

Sustainability initiatives can also create inequity. Device restrictions may disadvantage users with older hardware, automated efficiency systems may reduce human control, and data-intensive services may exclude communities with limited connectivity. Accessibility, affordability, labour conditions and environmental justice should form part of project evaluation.

Emerging Directions

Current developments include liquid and immersion cooling, specialised energy-efficient accelerators, disaggregated and composable infrastructure, carbon-aware cloud orchestration, circular hardware marketplaces and software sustainability standards. Edge computing may reduce latency and backbone traffic but can multiply the number of distributed devices and complicate maintenance. Digital twins and IoT analytics may optimise buildings or transport, yet their sensors and data platforms also consume resources.

AI-assisted coding and operations may improve software efficiency through automated profiling and capacity management, while simultaneously increasing the volume of generated code and computation. The appropriate response is not to reject innovation but to establish measurable environmental requirements, independent evaluation and lifecycle accountability.

Best-Practice Checklist

A technically and organisationally mature programme should:

• define environmental boundaries and functional units before making claims;
• measure energy, carbon, water, material use and useful service output where relevant;
• include embodied impacts and equipment lifetime in architecture decisions;
• set sustainability requirements alongside security, reliability and performance;
• right-size, virtualise and schedule workloads without violating service constraints;
• design software for efficient computation, data movement, storage and longevity;
• use carbon-aware operation for genuinely flexible workloads;
• procure repairable, supportable and responsibly sourced equipment;
• prioritise reuse, secure erasure and verified recycling;
• audit suppliers and communicate assumptions transparently;
• review rebound effects, accessibility and distributional consequences;
• monitor results continuously and revise interventions when evidence changes.

Chapter Summary

Sustainable computing evaluates digital systems across their full life cycle, from material extraction and manufacture to operation, maintenance and end-of-life. Data centres, networks, devices, software and AI all create environmental impacts, but their effects depend on architecture, utilisation, energy sources, workload timing, equipment lifetime and user behaviour. Metrics such as PUE are useful only when combined with service-level, carbon, water and embodied-impact measures.

Green software engineering embeds sustainability in requirements, architecture, coding, testing, deployment and retirement. Cloud and carbon-aware computing can reduce impacts when workload flexibility, data movement, privacy and resilience are properly considered. Sustainable digital transformation also requires responsible procurement, asset management, repair, reuse, secure data deletion and verified e-waste processing. The central professional principle is evidence-based lifecycle decision-making: digital innovation should be assessed not only by functionality and cost, but also by environmental integrity, social responsibility and long-term value.

Key Terms

Embodied carbon: Greenhouse-gas emissions associated with material extraction, manufacture, transport and installation of a product or system.

Operational carbon: Emissions arising from energy consumed while equipment or services operate.

Life-cycle assessment: A systematic evaluation of environmental impacts across a product or service’s life cycle.

Functional unit: The quantified service basis used to compare environmental impacts, such as one transaction or one month of service.

Green software engineering: Software development that deliberately reduces environmental resource use throughout the software life cycle.

Energy efficiency: The amount of useful service delivered per unit of energy.

Carbon-aware computing: Scheduling or locating workloads according to the carbon intensity of electricity.

Power Usage Effectiveness: The ratio of total data-centre energy to energy used by IT equipment.

Rebound effect: Increased consumption that partially or fully offsets efficiency gains.

Virtualisation: Abstraction of physical computing resources so that multiple virtual workloads share hardware.

Autoscaling: Automatic adjustment of computing capacity in response to workload demand.

Electronic waste: Discarded electrical and electronic equipment and its components.

Circular economy: An approach that retains products and materials in use through repair, reuse, remanufacture and recycling.

Sustainable procurement: Purchasing that considers environmental, social, lifecycle and economic criteria.

Carbon intensity: Greenhouse-gas emissions associated with a unit of energy or activity.

Water usage effectiveness: A data-centre indicator relating water consumption to IT equipment energy or service operation.

Model quantisation: Reducing numerical precision in an AI model to lower computation and memory requirements.

Right-sizing: Matching infrastructure capacity to actual workload requirements.

Review Questions

Short Answer Questions

1. Distinguish embodied carbon from operational carbon.

2. Why is PUE insufficient as the sole measure of data-centre sustainability?

3. Define carbon-aware computing and identify one suitable workload.

4. What is meant by software longevity?

5. Explain the rebound effect in digital transformation.

Long Answer Questions

1. Explain the major environmental impacts of data centres, networks and end-user devices using a life-cycle perspective.

2. Discuss how green software engineering can influence requirements, architecture, coding, testing and deployment.

3. Analyse the opportunities and limitations of cloud computing as a sustainability strategy.

4. Describe a responsible organisational policy for hardware procurement, reuse and e-waste management.

5. Examine the environmental and governance challenges associated with AI systems.

Analytical / Application Questions

1. A company plans to move all workloads to a region with lower grid-carbon intensity. Develop an evaluation framework that also considers latency, data transfer, privacy, resilience and embodied impacts.

2. An online learning platform experiences rising storage and compute demand. Propose measurable interventions and identify the data required to assess their effectiveness.

3. Compare extending the life of existing servers with purchasing highly efficient replacements. Explain the assumptions, metrics and scenarios needed to make a defensible decision.

22. Multiple Choice Questions with Answers

Question 1

Which measure best represents the useful service basis for comparing two digital architectures?

A. Number of software licenses purchased

B. One thousand secure transactions delivered

C. Total number of employees in the organisation

D. Peak processor clock speed

Correct Answer: B. One thousand secure transactions delivered

Question 2

A data centre improves its PUE, but its workloads double and its electricity comes from a carbon-intensive grid. What conclusion is most justified?

A. Its total environmental impact must have fallen

B. PUE alone cannot establish its overall sustainability

C. Its embodied carbon has become zero

D. Its network traffic is necessarily lower

Correct Answer: B. PUE alone cannot establish its overall sustainability

Question 3

Which workload is generally most suitable for carbon-aware scheduling?

A. An emergency control system with strict real-time requirements

B. A latency-sensitive online examination session

C. A flexible batch analytics job with a deadline

D. A medical device requiring continuous local operation

Correct Answer: C. A flexible batch analytics job with a deadline

Question 4

Why can replacing functioning hardware with newer efficient equipment increase environmental impact?

A. New equipment never consumes electricity

B. The embodied impacts of manufacturing and disposal may outweigh operational savings

C. Efficient equipment always requires more network bandwidth

D. Older equipment cannot be securely reused

Correct Answer: B. The embodied impacts of manufacturing and disposal may outweigh operational savings

Question 5

Which software design choice most directly reduces unnecessary resource consumption?

A. Increasing polling frequency for all services

B. Retaining every log indefinitely

C. Using caching and event-driven processing where appropriate

D. Duplicating all datasets without a retention policy

Correct Answer: C. Using caching and event-driven processing where appropriate

Question 6

What is a major limitation of treating renewable-energy procurement as proof of zero environmental impact?

A. Renewable generation requires no materials

B. It ignores embodied, water, land and timing-related impacts

C. It prevents any form of energy measurement

D. It makes software optimisation impossible

Correct Answer: B. It ignores embodied, water, land and timing-related impacts

Question 7

An organisation is preparing a laptop procurement specification. Which requirement most supports sustainability across the life cycle?

A. Shortest possible warranty period

B. High purchase price without supplier evidence

C. Repairability, long software support and take-back arrangements

D. Replacement of all devices every year

Correct Answer: C. Repairability, long software support and take-back arrangements

Question 8

Which AI technique can reduce inference resource demand by using lower numerical precision?

A. Quantisation

B. Indefinite data retention

C. Uncontrolled hyperparameter search

D. Increasing model size without evaluation

Correct Answer: A. Quantisation

Question 9

Why should a sustainability assessment disclose its system boundary?

A. To avoid measuring any environmental impact

B. Because comparisons can be misleading when different life-cycle stages are included

C. To guarantee that all estimates are exact

D. Because operational energy is never relevant

Correct Answer: B. Because comparisons can be misleading when different life-cycle stages are included

Question 10

A university shuts down idle development environments, deduplicates backups and schedules flexible jobs using grid-carbon data. What principle does this combined strategy demonstrate?

A. Single-metric optimisation based only on purchase cost

B. Lifecycle and workload-aware sustainable computing

C. Replacement of all governance with automation

D. Elimination of embodied carbon through cloud use

Correct Answer: B. Lifecycle and workload-aware sustainable computing