The Hidden Cost of AI Energy Consumption

AI will deplete our natural resources if leaders don’t act now, warns Wharton visiting scholar Cornelia Walther.

The following article was written by Dr. Cornelia C. Walther, a visiting scholar at Wharton and director of global alliance POZE. A humanitarian practitioner who spent over 20 years at the United Nations, Walther’s current research focuses on leveraging AI for social good.

Artificial intelligence is reshaping public and private sectors, from academia to the arts, and no arena is safe. In industrialized countries around the world, millions are either enthused about AI’s potential to revolutionize their work and life or depressed by its impact on their livelihood and the underdefined threat of Artificial General Intelligence/Artificial Super Intelligence.

Amid the hype about the prospects, we tend to overlook several significant downsides of our new artificial asset that have already begun to manifest. These include the way in which it jeopardizes our natural resources.

On the one hand, the training and deployment of frontier models come with a heavy environmental footprint. On the other hand, constant AI assistance comes with the risk of agency erosion. While one endangers the planet that we depend on, the other weakens the very human beings we are. Both risks can be averted or at least mitigated once we are aware of them. However, magic does not happen by itself. We need to address the dangers of natural resource depletion before they hijack the hybrid future that we are moving toward. In the following article, we will focus on the planetary problem.

The Cost of AI Energy Consumption

Every time we interact with AI — whether it’s a business utilizing cloud computing for decision-making or an individual asking ChatGPT for help — there’s an invisible but substantial energy cost. As AI models grow in complexity, the energy required to train and deploy them skyrockets, raising concerns about the environmental footprint of AI-driven technologies.

As businesses embrace AI, they not only inherit its energy burden but also contribute to a broader cycle of ecological strain. Both are situated in a geopolitical spider net where money and power tend to take precedence over stability and shared quality of life.

Training large language models like GPT-4 demands immense computational resources. Recent research shows that training GPT-3 consumed approximately 1,287 megawatt-hours (MWh) of electricity, emitting 502 metric tons of CO₂, which is roughly equivalent to the emissions of 112 gasoline-powered cars over a year. This is only the training phase. The power required for “inference” — when models process real-time queries — can account for up to 60% of AI’s total energy consumption.

This energy-intensive nature of AI is increasingly activated in our everyday interactions. For instance, a ChatGPT request uses about 100 times more energy than a typical Google search.

AI’s Water Footprint and the Burden of Environmental Inequity

Beyond the energy demands for training and usage, the infrastructure required for bringing AI systems to scale strains the planet’s natural resources, especially water. Data centers, which power AI models, use vast amounts of water for cooling. According to recent estimates, the average data center uses 1.7 liters of water per kilowatt-hour of energy consumed. As AI becomes more integrated into everyday business operations, the increasing demand for water places additional stress on already scarce resources. While the tech industry tends to emphasize the benefits of AI, including its potential to devise creative solutions to address and alleviate climate change, it is time to also account for the hidden costs of our algorithmic affinity, which directly affect our ecosystems and have consequences that have begun to manifest.

For now, it is easy to play the ostrich, turning a blind eye to the downsides, because so far not everyone has been affected. Communities most affected by climate change, toxic water waste, and resource depletion are often not the ones benefiting from AI’s technological advances. For example, while Google’s data center in Finland runs on 97% carbon-free energy, its centers in Asia rely heavily on fossil fuels, contributing to local pollution. Similarly, water-intensive data centers in drought-prone regions burden local communities with increased water scarcity​.

The Hybrid Nexus: Quality of Life, Social Cohesion, Sustainability, and Geopolitics

AI’s environmental footprint is not an isolated problem. It sits at the intersection of several critical global challenges: quality of life, social cohesion, environmental sustainability, and geopolitical stability. This hybrid nexus shapes the future of both personal and planetary power dynamics. Here is a nonexclusive list of examples for each component:

Quality of Life: On an individual level, the strain AI places on natural resources has indirect effects on human well-being. Water shortages, rising energy costs, and the degradation of local environments will increasingly lead to socioeconomic disparities and a reduced quality of life. (Moreover, reliance on AI may diminish cognitive capacity and emotional well-being, as agency decay reduces our ability to solve problems independently. AI addiction is a slow process — and if left unaddressed, digital attachment disorders may become a common disease soon.)

Social Cohesion: AI systems have (in)direct consequences on communities. On the material side, the sheer impact of climate change and resource depletion impacts people and their livelihoods. On a more subtle level, AI companionship comes with the risk of isolation and decreased appetite for interpersonal relationships.

Environmental Sustainability: AI expands the climate footprint of high- and medium-income countries. As industries rely more on AI, their environmental toll will increase unless we adopt sustainable energy sources. AI’s hunger for electricity puts additional pressure on energy grids, further emphasizing the need for alternative energy solutions. Nations that fail to secure reliable, renewable baseload power will fall behind in the global AI race, losing both economic and geopolitical ground.

Geopolitical Stability: The race to dominate AI and renewable energy is shaping new geopolitical realities. Countries that lead in AI and energy innovation will dictate the global economic agenda. Conversely, those that remain reliant on fossil fuels or outdated energy infrastructure risk falling behind in both technology and influence, leading to heightened geopolitical tensions.

In short, the environmental costs are compounded by social inequity, further deepening the divide between those benefiting from AI and those disproportionately bearing its environmental and social costs​.

AI can be a force of good. But looking at humanity’s track record with past waves of innovation, cautious pessimism is permitted. If AI remains determined by commercial interests, as did past technologies, the hybrid nexus may spin out of control.

Ironically, the stronger our artificial assets become, the more our natural assets are jeopardized, from the depletion of cognitive agency to the climate change conundrum. Cynically, one might argue that those currently deprived of AI access will be the ones laughing at the end because without the facility of 24/7 artificial assistance, their brains, bodies, and communities are forced to withstand adversity, growing stronger and more resilient over time — whereas those who surf the curve of AI-powered convenience become ever more dependent. Ultimately, however, we only have one planet.

Practical Steps to Combat AI Energy Consumption and Environmental Costs

To mitigate AI’s environmental and human costs, both businesses and individuals need to take actionable steps to ensure AI serves as a force for good. The A-Frame — Awareness, Appreciation, Acceptance, and Accountability — offers a structured approach to address these challenges, including one’s own mindset:

1. Awareness

Recognizing the energy, water, and resource consumption accompanying every AI interaction is the first step toward more responsible usage. Businesses can become savvy in systematically assessing the energy intensity of their AI operations, while individuals can choose to track how often they engage with AI tools.

Practical Action: Conduct regular energy audits to track consumption. Tools such as Microsoft’s Sustainability Calculator or Google’s Environmental Insights Explorer can help assess the environmental impact of cloud-based AI services. Individuals can minimize their carbon footprint by consciously reducing unnecessary AI usage.

2. Appreciation

AI offers significant potential to increase the efficiency and effectiveness of businesses and workers. This must be weighed against the hidden opportunity costs. As stated by economists since the 1930s, there is no free lunch. Institutions and individuals should appreciate the immediate convenience of AI, while being aware of its longer-term consequences on natural resources.

Practical Action: Shift data center operations toward renewable energy. Tech giants like Microsoft, Google, and Amazon have pledged to power their data centers with 100% renewable energy. Whether and when these commitments will be translated to practice is open. In the meantime, individuals can choose services that commit to prioritizing sustainability in AI development. In that sense, ESG (environment, social, governance) — which has been tainted by critics as “greenhushing” — might get a revival as ESG4AI.

3. Acceptance

Accepting responsibility for the environmental costs of AI is a mindset shift that matters. Individually and collectively, we must balance AI’s convenience with its long-term impacts on the planet.

Practical Action: Implement energy-efficient AI models and reduce unnecessary retraining, which consumes vast amounts of energy. Techniques such as transfer learning and model distillation can significantly cut institutional energy use. With acceptance of the environmental implications, we can deliberately choose to lessen delegation, shifting from blind consumption to informed citizenship.

4. Accountability

Responsibility for AI’s sustainability extends beyond the corporate sector. Setting tangible goals for reducing the carbon footprint of AI usage is critical. Ultimately, humans remain accountable for the outcomes of AI, in the short- and long-term. Regardless of our growing possibilities to delegate ever more tasks to AI assistants, and soon AI agents, we must face the (in)direct consequences of that delegation.

Practical Action: Businesses should publicly commit to sustainability benchmarks for AI operations. Reporting on AI energy consumption and setting measurable targets aligned with international frameworks like the Science-Based Targets Initiative (SBTi) are interesting steps in the right direction. Individuals can seek energy-efficient alternatives and opt for less resource-intensive solutions, such as Google/Safari vs. ChatGTP, where possible.

A Path to Sustainability with AI

The rapid expansion of AI offers immense opportunities for growth and innovation, but we must carefully balance this progress with thoughtful, responsible practices.

By acknowledging AI’s hidden costs, taking action to reduce energy consumption, and setting clear benchmarks for sustainability, we can ensure AI drives productivity without depleting our most valuable resources. It’s time to act to preserve our planet.

The future of AI depends on responsible stewardship. The A-Frame guides us to balance AI’s potential with its impacts. By reducing AI’s environmental footprint and protecting cognitive autonomy, we can ensure AI serves us sustainably. Today’s choices shape tomorrow’s world.