Electronics Guide

The Scale of Digital Impact

Peer-reviewed estimates place the information and communication technology sector at roughly 1.8 to 3.9 percent of global greenhouse gas emissions. The range is wide because studies draw system boundaries differently, particularly over how much of the semiconductor and device manufacturing supply chain to attribute to the sector. Even the lower bound puts computing in the same range as commercial aviation, and the trend is upward as artificial intelligence workloads expand.

Data centers illustrate the scale most clearly. The International Energy Agency estimates that they consumed about 415 terawatt-hours of electricity in 2024, roughly 1.5 percent of global consumption, and projects that figure to more than double to about 945 terawatt-hours by 2030, or just under 3 percent of world electricity demand. Data center consumption has grown at roughly 12 percent per year since 2017, more than four times the growth rate of total electricity demand, with the United States, China, and Europe accounting for the large majority of the total. Artificial intelligence is the single largest driver of the projected increase.

The sector also offers leverage that heavy industry lacks. A software update can improve efficiency across millions of installed devices without replacing a single component. Shared cloud infrastructure runs at far higher utilization than the idle on-premises servers it replaced. Digital systems enable emissions reductions elsewhere through remote work, smart grid control, precision agriculture, and the replacement of physical goods with digital equivalents. These enabling effects are real, but they are frequently overstated, and they only count when a lifecycle analysis confirms a net reduction rather than a transfer of impact.

Where the Impact Arises

Environmental impact divides into embodied burden, created when hardware is manufactured and disposed of, and operational burden, created when hardware draws power. Which one dominates depends entirely on the class of equipment, and confusing the two leads to misdirected effort.

• End-user devices: Manufacturing dominates. Manufacturer disclosures for smartphones commonly attribute about four-fifths of lifecycle emissions to production, transport, and end-of-life processing, leaving a modest share for charging. The most effective software lever for a phone or a laptop is therefore not shaving milliwatts but keeping the device in service for more years.
• Servers and data centers: Operational electricity usually dominates over a server's life, though embodied carbon remains significant and grows in relative terms as grids decarbonize. Utilization is the decisive variable, because an idle server consumes a large fraction of its peak power while performing no useful work.
• Networks: Transmission equipment draws power largely independent of the traffic it carries in the short term. Average energy-per-gigabyte figures are useful for long-run capacity planning, but they overstate the savings available from simply sending fewer bytes over an existing network. Claims about the footprint of a single email or video stream deserve particular skepticism.

These distinctions have practical consequences. Compressing images reduces bandwidth, storage, and rendering work, which is worthwhile, but it does not deliver the savings that a naive per-gigabyte multiplication suggests. Extending the supported life of a fleet of laptops by two years, by contrast, avoids a genuine and large manufacturing burden.

Key Principles

Several core principles guide work in this field:

• Efficiency at every layer: Opportunities exist throughout the stack, from silicon and firmware through algorithms, data structures, and protocol design up to user interface choices that determine how much work the system performs at all.
• Carbon awareness, not only energy awareness: The emissions from a kilowatt-hour vary by hour and by region. Shifting flexible workloads such as batch training, backups, and index rebuilds to times or places with cleaner supply reduces emissions without reducing the work performed.
• Hardware efficiency: Doing more work on the hardware that already exists, and running it closer to full utilization, spreads embodied carbon across more useful output and delays replacement.
• Full lifecycle thinking: Sustainable digital systems account for manufacturing, operation, and end-of-life impacts, including the impact of the development, testing, and continuous integration pipelines that build the software.
• Sufficiency questioning: Asking whether a feature, a data retention policy, or an entire service is necessary often yields larger reductions than optimizing it. Deleting unused data and retiring dormant services costs nothing in performance.
• Transparency and measurement: Credible measurement, with stated boundaries and assumptions, enables informed decisions and distinguishes real progress from favorable accounting.

Measuring Digital Impact

Measurement in this field suffered for years from incompatible methods and unstated assumptions. Standardization has improved matters considerably.

The Software Carbon Intensity specification, developed by the Green Software Foundation and published as ISO/IEC 21031:2024, defines a method for calculating a software system's carbon emissions as a rate per functional unit rather than as a total. The functional unit may be a user, a transaction, an API call, or any other unit that reflects what the software delivers. The score combines operational emissions with embodied emissions amortized over the hardware's useful life, and it deliberately excludes offsets, so a score improves only when the system actually emits less per unit of work.

Infrastructure reporting relies on a small set of established indicators. Power usage effectiveness expresses total facility energy divided by the energy delivered to information technology equipment, so a value approaching 1.0 indicates low overhead from cooling and power distribution. Water usage effectiveness, energy reuse factor, and renewable energy factor cover water consumption, recovered waste heat, and clean supply respectively. Power usage effectiveness alone is a poor sustainability measure, because it says nothing about whether the information technology load performs useful work, and a facility can improve its ratio simply by running its servers harder on wasteful tasks.

Regulation increasingly compels disclosure. Under the Energy Efficiency Directive (EU) 2023/1791 and Commission Delegated Regulation (EU) 2024/1364, data centers in the European Union with an installed information technology power demand of at least 500 kilowatts must report energy performance and sustainability indicators annually to a European database. The first reports, covering calendar year 2023, were due by 15 September 2024, and subsequent reports are due by 15 May each year. Corporate reporting adds another layer: emissions from purchased electricity fall under scope 2 of the Greenhouse Gas Protocol, where the location-based method reflects the average grid where consumption occurs and the market-based method reflects contractual instruments such as power purchase agreements. The two methods can produce very different totals for the same data center, which is why credible reports disclose both.

Articles in This Category

The following topics examine digital sustainability from infrastructure through code, service design, and the software policies that determine how long hardware stays useful.

Software as a Determinant of Hardware Lifetime

Software decides how long hardware remains useful, and that makes software policy an environmental policy. Applications and operating systems that demand steadily more memory, storage, and processing power push functioning devices into retirement even though nothing has physically failed. When a vendor ends security updates for a model, cautious owners and every regulated organization must replace equipment that still works. The resulting waste is entirely a consequence of decisions made in software.

Because manufacturing dominates the lifecycle footprint of end-user devices, keeping them running delivers benefits that no amount of operational efficiency can match. A phone kept for four years instead of two roughly halves its annualized manufacturing emissions. Efficient, well-scoped software that performs acceptably on older hardware is therefore one of the strongest environmental levers available to a development team.

Regulators have begun to enforce this logic. Commission Regulation (EU) 2023/1670, which sets ecodesign requirements for smartphones and slate tablets and has applied since 20 June 2025, requires manufacturers to make operating system security updates available for at least five years after the last unit of a model is placed on the market, to supply professional repairers with essential spare parts for at least seven years, and to meet battery endurance criteria of at least 800 full charge cycles while retaining at least 80 percent of the original capacity. Community-maintained firmware and lightweight operating system distributions extend life further, and organizations that standardize on them can keep fleets in service well past the vendor support horizon.

Rebound Effects and Honest Claims

Efficiency does not automatically reduce total impact. Making computation cheaper tends to increase how much of it people buy, an effect long recognized in energy economics. For much of the 2010s, improvements in server hardware, virtualization, and facility design absorbed a very large increase in computing demand, so global data center electricity use grew far more slowly than the workloads it served. That decoupling has weakened: the efficiency gains available from consolidating into hyperscale facilities have largely been captured, while demand from artificial intelligence continues to rise steeply.

The practical implication is that efficiency work must be paired with attention to absolute consumption. An organization that halves the energy per inference and then runs ten times as many inferences has increased its footprint. Reporting that highlights intensity metrics while absolute emissions climb invites accusations of greenwashing, and increasingly it invites regulatory scrutiny as well. Substitution claims deserve the same discipline. Remote work reduces commuting emissions but shifts energy use to homes and can increase floor space per worker, so the net effect depends on the circumstances and must be measured rather than assumed.

Conclusion

Digital sustainability succeeds when teams treat computing as the physical activity it is. That means measuring with defined boundaries and disclosed assumptions, using a standardized method such as the Software Carbon Intensity specification, and reporting absolute emissions alongside intensity. It means running workloads on well-utilized infrastructure, shifting flexible work toward cleaner hours and regions, and deleting data and services that no longer earn their keep.

Above all, it means writing software that respects the hardware it runs on. The largest single contribution most development teams can make is to build systems that continue to perform well on devices that are several years old, and to support those devices with security updates for as long as they remain serviceable. Efficiency in the data center reduces an ongoing electricity bill; longevity at the edge avoids manufacturing that would otherwise be repeated across millions of devices.

Related Topics