While “AI” was the primary buzzword of 2025 Enterprise Connect, “sustainability” came in second. To those who rely on AI and other power-heavy applications to run their businesses, governments and lives, the issues of sustainability have gained increasing attention and importance as we plan for the future.
Per Akoh Atadoga, Uchenna Joseph Umoga, Oluwaseun Augustine Lottu and Enoch Oluwademilade Sodiya, writing in the World Journal of Advanced Engineering Technology and Sciences, sustainability can best be defined as “the utilization of computing resources in an environmentally responsible manner, with a focus on minimizing energy consumption, reducing carbon emissions and mitigating ecological impact.”
Government and enterprise policies, practices and procurements that are environmentally friendly have taken on new importance in the face of daily climate challenges. In order to be effective sustainability leaders, technology decision-makers must recognize that such considerations are more about future than present operations (simply as a result of the time necessary to procure, deliver, install and deploy such products and systems). Nonetheless, as temperatures rise and as power consumption grows exponentially, it is the time to address three goals to sustainability: minimizing energy consumption, carbon emissions and electronic waste.
Related:No Jitter Roll: GoTo Announces New Grasshopper Capabilities
Artificial intelligence typically requires significant energy consumption to support the massive amounts of computer horsepower to complete the assigned tasks. But the actual computing power necessary to complete the task is only one part of the equation. There’s also the time and energy used to train the AI, as well as the power necessary to keep devices cool, plus the costs of actually getting to the equipment that’s doing the processing. Lastly, there are a number of issues about the equipment itself: servers and other technology-based devices are frequently made obsolete by newer, faster, and smaller models, and while new devices may offer improvement in efficiency and processing, they are still environmental burdens, both in terms of production and in terms of disposal challenges for the old technology, which often has toxic components requiring specialized disposal.
According to Ryan McPherson, Chief Sustainability Officer at the University at Buffalo, “AI has an insatiable thirst for power.” Over the next few years, he expects an increase in demand for electricity between 15 and 20% greater than current levels. “One of our goals is to recapture heat generated by servers and other equipment so we can utilize it for heating and cooling in other applications and thus reduce overall energy demand.”
Related:No Jitter Roll: Informatica and Carnegie Mellon Partner for AI Research
I asked McPherson about practical approaches and finite actions that can measure the success (and failure) of reducing carbon emissions—not simply shifting them from one responsible party to another. That is, outsourcing data processing and AI work to a third party does not, in and of itself, provide a sustainable model, simply because the work that draws electricity is not being eliminated, just being shifted from one location to another.
In a recent whitepaper on sustainability, Genesys suggests, “Choosing a foundational model instead of training a new one avoids much of the energy needed for training and spreads the energy that is used across the model’s life. Processing can be streamlined through techniques like quantization (compressing models to reduce memory usage of parameters) and dimensional reduction (transforming data from a high-dimensional space to a low-dimensional space) to further improve model efficiency, although these methods can introduce tradeoffs with model accuracy.”
On the software and hardware fronts, there are steps that can be taken to lessen the burden on the power draw. Efficiency is really the key here. Code creation, and the hardware on which it runs, should be designed and utilized to rely on algorithms that are as streamlined as possible, thus requiring minimal resource utilization. The creation of useful -- this is a key word here -- AI tools, which will require not only code compilation, training, testing and ongoing adjustment and modification as warranted are also key considerations. Genesys also further indicates that system designers should consider the training and retraining frequency of any AI model and these designers can choose energy-saving alternatives to model training like retrieval augmented generation (RAG), a way to connect an AI neural network to a data store (such as a new technical paper or a database of images) without retraining.
Related:Google’s Agentic AI Vision for the Workplace Rewrites Both Code and Culture
Practical Considerations
As enterprise users look to design, create, test and deploy increasingly sophisticated AI based tools, those who are doing the acquisition should have a toolkit to ask the right questions. As a start, what are the magic words or phrases to use with a vendor who is talking about sustainable processes? This may sound overly obvious, but the first step is to actually ask the questions: what are the efficiencies offered by these tools? What is the track record? While it’s not like a major corporation will be able to necessarily recognize a drop in its electric bill, it’s a good question to ask how the enterprise will benefit in terms of reaching sustainability goals. The second question, which applies to all acquisitions, is whether or not you can believe the vendor response. Essentially, how will you know if the system is working as intended after the enterprise has plunked down considerable cash to make the acquisition.
The next question is how will success be measured? There’s no right answer here, but if you’re promised better gas mileage on a new car, it’s easy to measure whether the mileage is actually better than on the old jalopy. But here, where the power demands are both huge and movable, success may be very tough to measure. But that doesn’t mean that you shouldn’t ask the question – or find a way to measure the results somehow.
Two more recommendations: Where certified Energy-star products can be deployed, they should be. Lastly, before signing on the dotted line, enterprise consumers should consider options and costs associated to recycle, reuse or re-deploy products that are about to be replaced by products that are deemed “more sustainable.”
Sustainability may be the magic word of the moment, but it’s tough to know, certainly in the short term, what the best moves are to reduce the carbon footprint and increase efficiency without breaking the bank.
