Emissions displacement: the sustainability gain that never left the atmosphere
Getty Images
Getty Images· 5 min read
A server running at 15 per cent of its capacity still draws close to half its maximum power.
That single fact, documented in energy-efficiency research on the data centre industry, should unsettle anyone who has signed off a green IT report in the last three years. It means the machine idling in your estate is not cheap. It is nearly as expensive, in watts, as the machine working hard. Idleness is not free. It is merely invisible.
I have spent three decades building technology estates for banks. I have also signed the reports. And I have learned to distrust the moment when the numbers improve faster than the behaviour did.
Right now, across global enterprise IT, something is improving on paper but not in the atmosphere.
The International Energy Agency has projected that data centre energy use could climb from roughly one per cent of worldwide electricity demand to a multiple of that within about a decade. Global data centre power capacity has grown sharply in just a few years. Meanwhile, corporate IT emissions disclosures have improved across many sectors.
Both things are true. That is the problem.
Here is what I see in advisory engagements. An enterprise migrates workloads to a hyperscaler. Its own energy consumption falls. Its Scope 2 figure improves. The sustainability slide turns green. Nothing about the electricity drawn, the water evaporated, or the hardware manufactured has changed. The emissions did not fall. They moved outside the boundary the organisation reports on.
I call this emissions displacement: the improvement in your sustainability position that comes from relocating impact rather than reducing it.
It is not fraud. Almost nobody does it deliberately. It is an artefact of measuring what sits inside your accounting boundary while the atmosphere recognises no boundary at all. And in the AI era it is accelerating, because the largest new computing loads in history are being provisioned by enterprises that will never see the substation.
What share of your computing emissions has moved from your direct reporting into a supplier's disclosure over the past three years, and did the absolute number fall when it moved? If your consumption dropped 40 per cent in the year you migrated, you did not decarbonise. You changed address.
What proportion of your provisioned capacity does real work? The gap between utilisation and power draw is the single largest source of waste in most estates, and it is entirely within your control. No supplier negotiation is required to switch off what nobody uses.
Training a model is a headline. Serving it is a habit. The energy cost of a model is increasingly dominated not by the weeks spent building it but by the years spent answering with it — often at far higher precision, and far higher frequency, than the decision actually warrants. Most enterprises have no measure of energy per decision. They should.
None of this is unmapped territory. The responsible computing framework developed at IBM and published by Parmar, Peters and Thomas sets out six pillars for leaders who want their IT to be green, ethical and trustworthy: data centres, infrastructure, code, data, systems and impact. It is the most practical blueprint I have encountered, and I recommend it to boards without reservation.
My argument is narrower, and it sits underneath theirs. A framework only governs what your measurement boundary contains. Draw the boundary conveniently, and even an excellent framework will certify a comfortable answer.
The first act of responsible computing is not adoption. It is honest boundary-setting.
In financial services, the fastest-growing load is fraud and risk inference running continuously against every transaction — enormous aggregate energy, almost never measured per decision. In healthcare, diagnostic models are retrained on cycles set by clinical caution rather than by need, and the compute cost of that caution goes unexamined. In automotive, the emissions of a software-defined vehicle increasingly sit in the cloud that serves it, not the plant that built it. In telecom, network intelligence is the substrate everyone else's efficiency claims quietly rest on. And in agriculture, advisory models serving smallholders must justify their energy against the water and yield they actually save — a standard the sector is well placed to meet, and rarely asked to.
The efficiency conversation often skips an equity dimension.
Compute is being sited where power is cheap, and regulation is light. Water is being drawn in places already under stress. The communities hosting the infrastructure of the AI economy are frequently not the communities benefiting from it. Any responsible computing position that counts carbon while ignoring who bears the local cost is incomplete — and the same holds inside the organisation, where the diversity of the team designing a system determines whose needs the system was built to notice.
Responsible computing is not only about what technology consumes. It is about who it serves, and who pays for it in ways that never appear on an invoice.
Restate your boundary before you restate your target. Ask for one number: total computing emissions including everything you outsourced, expressed on the same basis as three years ago. If nobody can produce it, that is the finding.
Measure energy per decision, not just per workload. Make it a standing metric for every AI system in production. It will change model selection faster than any policy document.
Put displacement in the audit committee pack. Label sustainability improvements achieved by relocation as such, separately from reductions achieved by design. Boards can only govern distinctions they are shown.
The organisations that will lead the sustainable transition are not the ones with the best-looking disclosures. They are the ones willing to draw the boundary where it hurts, and then do the harder work inside it.
A green report and a green estate are not the same document.
Which one does your organisation actually have?
illuminem Voices is a democratic space presenting the opinions of leading Sustainability Thought Leaders, their views do not necessarily represent those of illuminem.
The world needs sustainability knowledge. At illuminem, no interest group or shareholder can influence our work. Thank you for supporting our mission to make high-quality and independent sustainability information free for all. Every contribution helps. Thank you for donating today.
John Elkington

Corporate Governance · Sustainable Business
illuminem briefings

AI · Public Governance
John Elkington

AI · Corporate Governance
MIT Technology Review

AI · Green Tech
The Washington Post

AI · Corporate Governance
The Wall Street Journal

AI · Public Governance