A virtual machine runs all weekend. An AI assistant summarizes a three-page document. Backups pile up “just in case.” From a distance, the waste seems obvious. On the ground, each use may have its reasons: supporting on-call duties, absorbing a workload spike, preventing data loss. Digital sustainability begins when criticism gives way to a question: what need are we trying to safeguard? By September 2026, this capacity for dialogue could matter as much as management tools.
Digital technology is not immaterial, but neither is work
The rise of cloud computing and generative AI has brought a long-abstract reality into sharper focus: behind an interface lie servers, networks, electricity, cooling and equipment that must be manufactured. Research by Ademe and Arcep on the environmental footprint of digital technology in France has notably highlighted the significant contribution of end-user devices and their manufacture. Cloud computing is therefore only part of the issue, even when it dominates discussions.
The International Energy Agency has also documented the growing pressure that data centers place on electricity demand. But translating this collective finding into individual instructions remains difficult. An AI query does not have a universal footprint: the model, hardware, response length and operating conditions all affect the result. Striking comparisons can raise awareness; they become counterproductive when they create an illusion of precision.
Within companies, another resource is in short supply: time. Asking a team to reduce its consumption without changing its objectives or resources amounts to passing a contradiction on to it. Environmental messaging then risks being heard as yet another productivity demand, accompanied by moral judgment.
Listen before cutting back
Imagine a product team being asked to shut down its test environments every evening. The measure seems straightforward. Yet its developers work with a provider in another time zone; some tests run overnight; restarting requires manual intervention. Faced with these constraints, a blanket instruction will mainly produce covert exceptions.
The right starting point is a short discussion with those who use and maintain the service. Not to secure their buy-in for a predetermined solution, but to understand when the service is needed, its dependencies, the incidents they fear and the invisible tasks involved. One question helps move beyond entrenched positions: “What would need to change so that this reduction does not make your work harder?”
A compromise can then emerge: shutting down only inactive environments, adapting the schedule, automating restarts and allowing a simple exemption during releases. The environmental ambition has not disappeared. It has become a workable rule. This kind of listening requires reflecting back what people say, but also genuine room for negotiation.
Turn disagreement into useful information
Not all resistance stems from a lack of awareness. A security manager may require long-term log retention. A sales team may fear losing responsiveness. A developer may have learned, after an incident, to overprovision as a precaution. These positions reveal the risks each person bears.
The manager’s role is to make those risks open to discussion without dismissing the people raising them. Rather than saying “you store too much,” they can ask: “Which data must remain immediately accessible, which can be archived, and who approves its deletion?” The discussion becomes technical and organizational, rather than pitting supposedly virtuous people against everyone else.
A useful agreement fits on one page
- The need being protected: availability, security, deadlines or expected quality.
- The change being tested: precise scope, person responsible and trial duration.
- The indicators: consumption or resources used, but also effects on work.
- The rollback rule: conditions for stopping and the exception procedure.
This framework avoids vague promises. It also makes it possible to acknowledge that a measure can fail without its advocates losing face. This sense of interpersonal safety is essential: no one will report a deterioration if raising the alarm risks being interpreted as hostility toward environmental protection.
Cloud computing: measure more than the bill
FinOps practices, which bring together finance, technology and business teams to manage cloud spending, offer a starting point. Identifying resources with no assigned owner, adjusting capacity or removing duplicates can reduce both costs and some resource consumption. But a lower bill is not sufficient proof of a smaller footprint: commercial discounts and pricing obscure the relationship.
A sensible dashboard therefore combines several dimensions: compute hours, stored data volumes, data transfers, environmental estimates where available, and service quality. Data limitations must remain visible. Comparing two estimates produced using different methods can yield a misleading ranking.
Total volume must also be tracked. A less resource-intensive process that runs far more often can cancel out the expected savings. This risk of a rebound effect deserves collective discussion: what do we do with the capacity freed up? Digital sustainability is not just about optimizing each operation, but also questioning why operations are multiplying.
AI: negotiate the right level of service
With generative AI, the debate easily becomes polarized between compulsory enthusiasm and outright prohibition. A more productive approach starts with the tasks. Classifying messages, searching documentation and producing a complex analysis do not necessarily require the same tool. A conventional search, deterministic automation or a smaller model may sometimes suffice, provided quality is verified.
Rather than imposing a uniform query quota, the team can test a tiered rule: a simple tool by default, and a more powerful model when the difficulty warrants it. The workflow must still remain smooth. If the “resource-efficient” alternative requires five extra steps, adoption will be fragile and the human cost very real.
Evaluation must account for corrections, errors and verification time. A fast but unusable answer is not progress. Conversely, a more resource-intensive use may be justified if it delivers demonstrable value. The aim is not to cast suspicion on every interaction, but to avoid reflexive uses with no established benefit.
Give teams the power to act
Ultimately, credibility hinges on leadership decisions. You cannot ask people to slow resource growth while rewarding only delivery speed and the addition of features. Time must be set aside for maintenance, purchasing must be simplified, and discontinued services must be recognized as achievements, not as a lack of innovation.
What next? For September 2026 and beyond, a plausible path would embed digital sustainability in everyday trade-offs rather than awareness campaigns alone. Starting with one service, establishing a baseline, testing a compromise and then sharing the results is an accessible method. The sign of maturity will not be the number of employees made to feel guilty, but the collective ability to decide what truly deserves resources.
