According to Maria Kirschner, CEO of Kyndryl ALPS, despite its energy consumption, "generative AI can also play an important role in energy savings by making business operations more efficient."
With a total of 17 goals – called "Sustainable Development Goals" (SDG) – United Nations Member States have set themselves the ambition of promoting sustainable development at the economic, social and environmental levels. Switzerland is also part of this approach and has introduced various regulations on ESG.
Sustainable generative artificial intelligence (AI) can help achieve sustainability goals, notably by reducing energy consumption and greenhouse gas emissions, although it generates emissions itself. Indeed, generative AI tools and solutions appear to offer significant social and economic benefits.
Our recent survey reveals that, although adoption of this technology progressed strongly last year among the companies surveyed, only 18% of them actively use AI applications. A study by ETH Zurich, carried out in collaboration with Swissmem, indicates an even lower rate of 2%. However, around 22% of companies plan to carry out their first generative AI application tests or are in the process of doing so.
This growth comes as 2024 is expected to be the warmest year on record. Operating generative AI requires immense computing power, which increases data center cooling needs and calls into question current energy management strategies.
According to an article in IEEE Spectrum, the entirety of current AI technology – not just generative AI – could each year consume as much electricity as the whole of Ireland (29.3 terawatt-hours per year). Furthermore, a single interaction with a Large Language Model (LLM), a key component of generative AI, can use as much electricity as a low-power LED bulb turned on for one hour. That may seem insignificant, except that millions of LLM interactions per day are potentially expected.
Current efficiency gains will not be sufficient to offset the increasing energy demands of generative AI. It is therefore essential to exploit this technology to optimize its own energy consumption.
Despite its energy consumption, generative AI can nonetheless contribute to energy savings by optimizing the efficiency of business operations. Companies must be aware of its impact on their carbon footprint and adapt their strategies to make its use more sustainable. To meet their sustainability goals, they should begin integrating environmental data into their business decisions and invest in technologies that provide better visibility into sustainability indicators.
We recommend that companies focus on four key elements: measure their energy consumption, optimize their energy efficiency, develop less energy-intensive generative AI, and favor clean energy sources.
The first important step is to accurately assess an organization’s current greenhouse gas emissions and energy consumption. Continuous monitoring is essential to effectively optimize energy use. However, before starting this process, companies must first determine whether generative AI meets their specific business needs.
Once a company has assessed its energy consumption, it is better prepared to optimize its generative AI systems. For example, at Kyndryl ALPS, as part of our growth strategy, we analyzed in detail the use of our infrastructures, notably our data centers and real estate portfolio.
We thus reduced our real estate portfolio and transferred our operations to more efficient, state-of-the-art data centers. In addition, many servers were consolidated through virtualization, which reduced energy consumption. An essential part of this restructuring consisted of establishing partnerships with the largest global hyperscalers in order to improve operational efficiency.
The future of generative AI will depend largely on companies’ ability to collaborate to adopt cleaner and renewable energy sources.
Current efficiency gains will not be sufficient to offset the increasing energy demands of generative AI. It is therefore essential to exploit this technology to optimize its own energy consumption. Whether through virtualization, code optimization, migration to the cloud, or the use of more efficient programming interfaces, increased visibility into energy consumption enables companies to monitor generative AI and operate it more sustainably.
Finally, the future of generative AI will depend largely on companies’ ability to collaborate to adopt cleaner and renewable energy sources. If we do not manage to save more energy than we consume with generative AI, we risk missing out on a large part of its potential.
The development of new generations of backup batteries, the use of generators running on biodiesel, and shifting compute loads to cleaner energy sources will be part of a comprehensive approach to make generative AI more sustainable.
This article has been automatically translated using AI. If you notice any errors, please don't hesitate to contact us.
"If the Federal Council is now considering abolishing the program — or at least withdrawing the federal contribution — it is mainly because of the windfall effects it generates," explains Philippe Thalmann, professor of environmental economics at EPFL.
"Today, 70% of our results come from abroad, while 70% of our investments are made in our historic service territory in Switzerland," says Cédric Christmann, Chief Executive Officer of Primeo Energie.
Interview with Philip Mäder, head of media relations at BKW.
At this stage, the investigations reveal an overvoltage problem that led to the failure, but do not yet allow its root cause to be identified.