AI and the Exponential Growth in Energy Demand

Although it seemed futuristic and distant five years ago, Artificial Intelligence (AI) is now an inherent part of our daily internet consumption: a simple Google search now solicits the help of Gemini, the search engine’s AI assistant. AI has naturally gained importance in day-to-day activities and is now considered a crucial part of a company’s competitive edge, regardless of its sector. Moreover, AI-enabled tools have been growing prominently in the financial space, given the large quantities of data involved. Thus, AI is now inherent to scope 3 emissions’ calculations of many financial institutions. Indeed, increasing disclosure requirements have compelled financial institutions to track their scope 3 emissions, including financed and facilitated emissions, therefore increasing the need for meticulous management of emissions across all activities, as this broader range also signifies a rapid increase of additional emissions to account for.

There is a growing concern around the energy needed to support this exponential growth of AI. This has even been emphasized by OpenAI’s CEO, who noted that every single query to an AI system incurs a tangible additional cost, even seemingly trivial additions like “Please” and “Thank you.” This cost not only includes electricity demand but also a significant increase in water use, mostly for cooling systems in  data centers. It is becoming increasingly clear that the rapid integration of AI into daily operations is not just a technological leap forward but also a significant environmental burden, one that demands urgent attention. The critical question for investors and policymakers alike is clear: how much additional energy does AI’s exponential growth require, and how can we prepare to meet this new demand?

The magnitude of the issue

From an environmental perspective, the rapid rise in AI usage has created two major impacts: (1) direct^[1] emissions from computing, energy, and water consumption, as well as pollution and e-waste^[2]; and (2) indirect emissions from the purchased energy to power broader AI and machine learning applications. The surge in demand for data storage has driven the rapid expansion of data centers, dramatically increasing overall energy use. The scale of this growth is striking: from just 500,000 data centers in 2012 to over eight million globally today. Moreover, data centers consume disproportionately high amounts of electricity, and by 2028, AI alone could require as much electricity as 22% of all U.S. households.

To put this into the perspective of an individual, consider an analyst using ChatGPT for their work: according to a study by the MIT Technology Review, a conversation with 15 questions plus an image request involving 10 queries would consume an estimated 2.9 kilowatt-hours of energy. This is the equivalent of running a microwave for three and a half hours or traveling 100 miles on an e-bike. This example illustrates the magnitude of energy demand generated by a single user, let alone by an entire organization with hundreds of employees relying on AI daily.

How are companies and governments addressing this issue?

There are two main avenues through which governments and large technology companies are addressing this growing issue. First, the surge in energy demand requires a restructuring of the energy mix. For example, according to the IEA, the current energy mix of North America^[1] is composed of 37% of natural gas, 36% of oil and related oil products, and 9% of nuclear. By diversifying available sources, communities can better meet rising consumption needs. While renewable energy is a critical part of the solution, the scale of additional supply required is too great for any single source to cover. A diversified mix that combines renewables with other complementary low-carbon solutions such as nuclear power plants will be essential. For instance, the United States of America have recently focused on the development of Small Modular Reactors (SMR) as a nuclear-based alternative in order to diversify toward cleaner energy sources. Moreover, the technology giant Meta has invested in the development of nuclear-based power plants to support the energy demand of its data centers, thus reflecting a shift in energy sourcing.

On the other hand, major technology companies are building new data centers and working to improve their efficiency, reducing the electricity demand per facility. Technological advances in materials, infrastructure design, and cooling systems are central to lowering the energy footprint of each data center. Another way of tackling the vast energy demand is to address the issue of dark data : the data which is collected and stored, never to be used again. Refining the algorithms that power AI platforms can help better identify the data that is worth storing and reduce the amount of wasted storage.

Finally, governments and policymakers also play a decisive role. Stricter disclosure requirements and minimum efficiency standards can accelerate progress, shaping the way data centers are developed and managed. Some initiatives have already been established: in 2023, the European Parliament adopted the EU Energy Efficiency Directive, requiring data centers to report on energy use, water consumption, and renewable integration. This represents an important first step toward regulating and reducing AI’s environmental impact in Europe.

On a global scale, organizations are also stepping in. The World Economic Forum launched the Artificial Intelligence Governance Alliance, an initiative encouraging collaboration on AI’s most pressing challenges. Signatories include Apple, Accenture, Cisco, and other industry leaders, signaling a shared recognition that addressing AI’s environmental footprint requires joint action.

What practical tips should sustainability leaders in the financial sector keep in mind ?

It is essential for sustainability leaders in the financial sector to closely monitor the rapid evolution of AI’s energy footprint. This is important as they consider their own operations, provide financing for data center expansion, and manage the downstream impacts of this boom. Frameworks such as the Science-Based Targets initiative (SBTi) and the Global Reporting Initiative (GRI) already require companies to follow clear reduction pathways and may soon introduce specific requirements addressing AI-related energy use. Similarly, standard-setting organizations like the International Sustainability Standards Board (ISSB) could tighten expectations, potentially restricting some technology firms within investment portfolios.

Moreover, as energy prices fluctuate in the short to medium term, financial markets may experience volatility directly linked to AI’s rising energy consumption. Large technology companies should be scrutinized for changes in their net-zero commitments, which may become increasingly difficult to achieve given higher emissions tied to expanded energy demand. This represents another critical dimension for sustainability leaders to watch.

Finally, with AI contributing significantly to both Scope 2 and Scope 3 emissions for financial institutions, raising awareness internally is vital. Understanding the rising energy consumption of AI is not just about compliance; it is about anticipating market risks, safeguarding portfolio resilience, and capturing opportunities in the transition to a low-carbon economy. Ensuring employees in your organization have the necessary knowledge around the risks of AI energy consumption can be accomplished through sustainability training that focuses on the energy consumption at work, as well as the efficient ways to use AI.

Why Does Employee Sustainability Training Matter?

Overall, companies can reduce energy waste risks by increasing internal expertise through sustainability training for employees, with particular emphasis for the sustainability teams, investment and product professionals, and risk assessment teams. They may also turn to sustainability training providers, compliance courses or customizable corporate ESG training.

^[1] Emissions that come from sources that are directly owned or controlled by a company, also referred to as Scope 1 emissions. On the contrary, indirect (or Scope 2) emissions refer to those coming from sources that are purchased by the company (i.e. electricity, heating, etc).

^[2] Also known as electronic waste, refers to solid waste coming from computing and electronic devices that are challenging to recycle and dispose of.

^[3] Canada, United States of America, and Mexico

---

Sources

Crownhart, Casey; O’Donnell, James. (May 20, 2025). We did the math on AI’s energy footprint. Here’s the story you haven’t heard. MIT Technology Review, https://www.technologyreview.com/2025/05/20/1116327/ai-energy-usage-climate-footprint-big-tech/

Gluski, Andres; Bruch, Christian. (Mar 19, 2025). Powering the future: How to meet global energy demand in the age of electricity. World Economic Forum,  https://www.weforum.org/stories/2025/03/key-steps-for-meet-the-growing-global-demand-for-electricity/

The Guardian. (June 3rd, 2025). Meta signs deal with nuclear plant to power AI and datacenters for 20 years. The Guardian, https://www.theguardian.com/technology/2025/jun/03/meta-nuclear-power-ai 

IBM. (October 2, 2023). What is dark data? IBM, https://www.ibm.com/think/topics/dark-data

IEA. (2025). North America. IEA, https://www.iea.org/regions/north-america/energy-mix 

IFRS Foundation. (April 2025). Horizon scanning and corporate reporting in the future. IFRS, https://www.ifrs.org/content/dam/ifrs/meetings/2025/april/ac/ap5-horizon-scanning-corporate-reporting.pdf

Kemene, Eleni.; Valkhof, Bart.; Greene-Dewasmes, Ginelle. (July 22, 2024). AI and energy: Will AI help reduce emissions or increase power demand? Here's what to know. World Economic Forum, https://www.weforum.org/stories/2024/07/generative-ai-energy-emissions/

Risk Business Newsletter. (November 12^th, 2025). The green dilemma: AI and tech in finance – balancing innovation with environmental responsibility. RiskBusiness, https://riskbusiness.com/blog/the-green-dilemma-ai-and-tech-in-finance-balancing-innovation-with-environmental-responsibility/#:~:text=Scope%202%3A%20Indirect%20emissions%20from,companies%20to%20achieve%20carbon%20neutrality.

Quist, Zazala. (March 26th, 2020). What To Know About Direct & Indirect Emissions: Scope 1, 2, and 3 Emissions. EcoChain, https://ecochain.com/blog/scope-1-2-and-3-emissions-overview-to-direct-and-indirect-emissions/#:~:text=Scope%201:%20Direct%20Emissions,and%20employee%20travel%20and%20commute.

Shemfe, Mobi. & Kooroshy, Jaakko. (June 28th, 2024). Conundrum Cubed: Scope 3 for Financials. CFA Institute, https://blogs.cfainstitute.org/investor/2024/06/28/scope-3-emissions-spotlight-on-financial-institutions/ 

Sullivan, Helen. (July 14, 2025). Meta to spend hundreds of billions to build AI data centres. BBC, https://www.bbc.com/news/articles/c1e02vx55wpo

United Nations Regional Information Center for Western Europe. (July 4^th, 2025). Artificial intelligence: How much energy does AI use? United Nations,  https://unric.org/en/artificial-intelligence-how-much-energy-does-ai-use/

University of Copenhagen. (April 8, 2024). Revolutionizing AI With the Power of Energy Efficiency. SciTech Daily, https://scitechdaily.com/revolutionizing-ai-with-the-power-of-energy-efficiency/ 

U.S. Department of Justice. (2025). Advanced Small Modular Reactors (SMRs). U.S. Department of Justice, https://www.energy.gov/ne/advanced-small-modular-reactors-smrs 

Widuto, Agnieszka. (July 4^th, 2025). AI and the energy sector. European Parliament Think Tank, https://www.europarl.europa.eu/thinktank/en/document/EPRS_BRI(2025)775859

Wilkins, Joe. (April 19, 2025). Sam Altman Admits That Saying "Please" and "Thank You" to ChatGPT Is Wasting Millions of Dollars in Computing Power. Futurism, https://futurism.com/altman-please-thanks-chatgpt

World Economic Forum. (2025). AI Governance Alliance. World Economic Forum, https://initiatives.weforum.org/ai-governance-alliance/home

World Health Organization. (October 1, 2024). Electronic waste (e-waste). World Health Organization, https://www.who.int/news-room/fact-sheets/detail/electronic-waste-(e-waste)