On this page
At Nordic Business Forum 2026 I sat under a conference screen the size of a house. On stage was Paul Polman, who led Unilever from 2009 to 2018. Behind him was the sustainability story from those years. It showed three goals and a grid of coloured Sustainable Development Goal icons. At the end came a shareholder value figure. The room liked it. So did I, once.
What the screen did not show was the electricity behind it. Every AI demonstration at that conference ran on power that somebody paid for and nobody had counted. That is the subject of this piece. AI is a sustainability question, and the industry has been here before.
What cloud taught, too late
Companies moved to the cloud first and asked about its footprint years later. The early promise was simple. Shared data centres are more efficient than a server room in the basement. That was true, and it ended the conversation. Nobody defined a number for what a workload used, where it ran, or how much of the day it sat unused.
A large part of my work in the first years of cloud adoption was that retrofit. I measured workloads that had already been running for years. I turned a region choice and an idle hour into figures a finance team could read. The lesson was not that the numbers were hard. It was that nobody had designed them in, so every one had to be reconstructed afterwards.
AI is repeating it, faster
The same thing is happening with AI, and the scale is larger. A model uses electricity every time it answers, and the amount is no longer small. In 2025 the International Energy Agency published its first dedicated report on data centre demand. Boards have started reading it.
Most companies that funded AI work did so with no line in the record for what it uses. That is the cloud pattern again: adopt first, measure later. A board will notice, because the question is arriving from outside. A customer will ask for the company's energy numbers, and the AI work will be part of the answer.
This time, the numbers come first
The difference is available, and it is a decision rather than a technology. Define the number before the work starts. Which region the system runs in. Which model it uses. How many hours a day it sits idle. What it used in a month, in kilowatt hours, written next to the hours it saved or the revenue it moved.
Those numbers belong in the same record as everything else the board oversees. It is the same ledger, and the same person answers for it. The same rule applies: nothing goes in without a source and a date. A board cannot oversee an intention. It can oversee a number with a name on it.
Planned use is the lighter answer
Putting the number first changes what gets built. Not every problem needs AI. Of the problems that do, most do not need the largest model available. A smaller model sized to the task is cheaper, faster, and uses far less power. That is not a reason to use less AI. It is a reason to use it on purpose.
We take that seriously in our own work. We start from the business problem. Then we ask whether AI solves it at all, and if so, how small a model can do the job. We host in the European Union and can name the region. That is an answer a board can check. A commitment is not.
Where the reporting duty now sits
The regulation reaches a mid-market company indirectly. The Corporate Sustainability Reporting Directive turned sustainability into audited numbers. The Omnibus directive of February 2026 raised the thresholds. From the 2027 financial year, only large companies report directly. But those large companies are your customers and your lender. They need numbers from their whole value chain. So the duty arrives as a supplier questionnaire, with a deadline set by a customer. AI energy will be one of the lines on it.
Take it to the next meeting
Do not wait for the questionnaire. Pick the AI work the company has already funded. Find out where it runs, which model it uses, and what it used in its last full month. Find out who holds those numbers today and how long they take to produce. Put that time in the minutes.
Then decide where the numbers will live from now on, with a date and a source on every line. Write down who answers for them. The next AI project starts with those numbers already recorded.
The board is not really asking about sustainability. It is asking whether the company can account for what it does. This time, the answer can be planned rather than reconstructed.