A false choice
Sustainability in technology is often framed as a trade-off: you can grow fast, or you can be green, and every step toward one costs you something on the other. For most of what organizations do with technology, that framing is simply wrong.
Wasted computing power is wasted money. Oversized servers, idle environments, bloated web pages and data nobody uses all cost something to run, and they all consume energy for nothing. Cut the waste and you lower both the bill and the footprint. That is why the first round of sustainability work in technology usually pays for itself.
There are harder trade-offs further down the road, and we will come to those. But most organizations are nowhere near them yet.
Where the footprint really comes from
Before choosing what to fix, it helps to know where the impact sits. In a typical organization's digital footprint, three sources dominate:
- Devices. Laptops, phones, monitors and servers carry a large share of their lifetime emissions before they are ever switched on, from manufacturing and shipping. Replacing them frequently is one of the biggest sources of impact, and one of the least discussed.
- Data centers and cloud. The electricity used to run and cool computing infrastructure, which depends heavily on where and when it runs.
- Networks and data transfer. Every page load, video stream and file sync moves data that has to be transmitted and stored.
AI adds a new and fast-growing line to this picture. Training and running large models is energy-intensive, and the pace of adoption means data center electricity demand has become a topic for energy planners as well as IT departments.
Five levers that cut cost and carbon together
1. Switch off and right-size
Development and test environments that run around the clock, servers sized for a peak that never comes, storage holding years of forgotten backups: most cloud estates carry a surprising amount of idle capacity. Scheduling non-production environments to shut down outside working hours and right-sizing what remains is usually the fastest, easiest saving available.
2. Choose where and when
The carbon intensity of electricity varies widely between regions and across the day. Where your applications and data can live in more than one region, choosing a region powered by cleaner energy can reduce emissions significantly at little or no extra cost. Work that is not time-sensitive, such as batch processing, reporting and model training, can often be scheduled for times when cleaner power is available.
3. Build lighter products
A web page that loads megabytes of scripts, images and video to display a few paragraphs costs energy on every visit, on your servers, across the network and on the visitor's device. Lighter pages are also faster pages, which means better user experience, better search visibility and better conversion. Efficient code, sensible caching and only collecting the data you actually use all point the same way.
4. Keep hardware longer
Extending the life of laptops and servers by a year or two, through better maintenance, refurbishment and choosing durable equipment, avoids a large share of manufacturing emissions. When equipment does reach the end of its life, responsible reuse and recycling matter.
5. Use AI deliberately
Not every task needs the largest model available. Smaller models are often good enough for classification, extraction and routine drafting, at a fraction of the energy and cost. Caching repeated answers and avoiding unnecessary calls helps too. Choosing the right-sized model for each job is good engineering, good economics and good for the climate.
Measure in units that matter
Total emissions are important for reporting but not very useful for engineering decisions. A more practical measure is intensity: emissions per transaction, per customer or per user session. It shows whether you are becoming more efficient as you grow, which is the whole point.
Major cloud providers now offer carbon reporting for the services you use, and the Green Software Foundation's Software Carbon Intensity specification gives teams a consistent way to measure the footprint of an application. Neither is perfect. Both are far better than not measuring at all.
The reporting landscape
In Europe, sustainability reporting has been in flux. The Corporate Sustainability Reporting Directive set out broad new requirements, and during 2025 the EU moved to simplify and delay many of them. For some organizations, formal obligations are now further away than expected.
It would be a mistake to treat that as a reason to stop. Large customers increasingly ask their suppliers about emissions as part of their own reporting, investors continue to look at climate risk, and energy costs reward efficiency regardless of regulation. The organizations that build measurement and efficiency into their technology now will have an easier time, whatever the final rules look like.
Where the real trade-offs are
Once the waste is gone, genuine choices remain. Redundancy across several regions improves resilience but increases resource use. Real-time features cost more energy than batch ones. Ambitious AI products consume significant computing power. These are legitimate business decisions, and sustainability should be one of the inputs alongside cost, performance and risk, made explicitly rather than by accident.
Where to start
- Get visibility. Use your cloud provider's cost and carbon reports to find the biggest sources of spend and emissions.
- Take the quick wins. Switch off idle environments, right-size servers and clean up storage.
- Set an intensity metric for your main digital products and track it alongside performance.
- Make efficiency a design requirement for new products, including page weight and model choice.
- Extend hardware lifecycles and choose suppliers who report their own impact.
Growth and green are not opposites. In technology, efficiency is where they meet, and it is one of the few investments that improves the balance sheet and the footprint at the same time. Our Sustainability Advisory and DevOps and Infrastructure teams help organizations find that overlap and build on it.
