For a while, AI has had the rare privilege of being sold as everything at once: a productivity tool, a growth story, a labor-saving device and, by implication, a cleaner, smarter way of running a business. It was the corporate miracle diet. It has helped cut costs, increase production, automate tedious parts, and, one way or another, stay on track to achieve net zero.
However, miracles tend to seem more expensive when the electricity bill arrives. as Power requirements for AI infrastructure It’s becoming harder to ignore, and there’s a strange truth moving from the margins to boardrooms: the economics of AI and sustainability politics are no longer well aligned.
This tension is now evident in clear numbers. Google’s latest environmental reports show greenhouse gas (GHG) emissions An increase of 48 percent compared to the base year of 2019driven in part by energy use in data centers and supply chain growth associated with the expansion of artificial intelligence. Microsoft reports total greenhouse gas emissions in Scope 1, 2, and 3 An increase of 29.1 percent from the 2020 baselineMuch of the increase is due to the creation of the infrastructure required for new AI services.
Meanwhile, Amazon reported a High emissions Partly related to construction and expansion across its holdings, including the construction of data centres. For many years, the digital economy liked to present itself as airy and weightless. It turns out that the cloud is made of concrete, copper, water and a great deal of energy.
More of this energy will be needed as we do more with artificial intelligence. The International Energy Agency (IEA) recently reported that electricity use from data centers is critical. On track to more than double by 2030With artificial intelligence being the main reason for this increase. In the United States, energy demand is expected to reach record levels as loads on data centers rise. Of course, some of this electricity will come from renewable energy sources, but a significant share of it will not. As demand increases, the risk is not simply that AI consumes more energy. It even helps keep dirty energy around longer, because the business imperative for reliable supply arrives before the green grid.
This is where the company’s script starts to waver. Almost every big company now wants to talk about two things at once: its enthusiasm for AI and its commitment to sustainability. In theory, the two can coexist. In practice, they have reluctantly begun sharing space like a bonfire at the climate summit.
However, most boards are not discussing AI primarily as an environmental problem, but rather as a governance issue. From their perspective, this concern is perfectly reasonable. The immediate fear in boardrooms is not that the chatbot has used too much water in Cheshunt. That is, an employee pasted confidential material into a large public language form (LLM), violated a company data policy, exposed customer data, or created an industry-wide legal issue. The immediate concern is control: who uses what, over what systems, with what information, and under whose authority.
That’s why many IT managers and technology managers are now issuing a different version of the same command: Use the approved enterprise tool, not the latest consumer toy. Keep data within the fence – use our “Private MBA”. Stop employees working autonomously with “shadow AI” in the same way that previous generations were told to stop storing files on personal USB devices or redirecting work to private email accounts. The ruling is where the heat is Because governance is where responsibility lies.
But there is a trick being played here, most of it unintentionally. By treating AI primarily as a security, compliance, and policy issue, companies can feel like they are managing it responsibly while leaving its environmental costs largely unexamined. If the form is properly authorized, permissions are in order and claims occur in a sanctioned institutional environment, the organization can tick the governance box. What it cannot easily do is explain the resource density of the infrastructure underneath.
This is important because many companies still treat AI with the intellectual seriousness of a gold rush. They know they need an AI strategy in the same way that previous generations knew they needed a “digital transformation strategy,” whether or not anyone can say exactly what problem was solved in the process. The result is predictable: too many projects, too little discipline, and the instinct to get to the biggest possible technical answer before smaller, cheaper, less energy-intensive projects are exhausted.
The answer? Most organizations should start with a use case, not a model. What decision, workflow, or reporting load really needs improvement? What data is required? Can it be governed correctly? Is a full generative AI layer necessary or will analytics, retrieval, or a simpler rules-based approach do the job better? This seems almost plausible offensively, which is precisely why it’s useful.
This perspective is shaped in part by our roots in environmental, social and institutional analysis, including work to detect misstatement in the performance figures published by listed companies. This background provides a revealing perspective on the boom in artificial intelligence today. The next issue may not be whether companies use AI at all. It may be the case whether they overestimate the upside while underestimating the overall environmental costs. Companies are very good at advertising efficiency gains. She is less enthusiastic about discussing rapidly rising electricity and water bills.
None of this suggests that AI is a hoax or that companies should retreat to candlelight and spreadsheets. If used well, AI can reduce waste, improve reporting, streamline operations, and help companies make better decisions. But it’s not magic. It’s the infrastructure. Infrastructure requires gas.
The companies that are able to successfully navigate this balancing act will not be the ones that are most vocal about transformation. They will be the ones with the discipline to ask a somewhat unfashionable question before scaling up another business: Is our next use of AI really worth what it costs? This is not against innovation. This is what human supervision should look like.
