Author: Anahida Bhardwaj
AI is transforming our world at an astonishing pace—but beneath the buzz of innovation, are we ignoring a looming environmental crisis? By 2027, AI servers could devour 85-134 TWh of electricity annually, nearly 0.5% of global consumption. To put this in perspective, Bitcoin mining, widely criticized for its energy intensity, consumes an estimated 155-172 TWh per year. But this is not just an energy problem; it is a question of global equity. As AI’s resource demands rise, are we risking a repeat of the same North-South divide that has long defined the climate crisis?
Overlooked Environment Toll
Behind AI’s impressive capabilities lies an energy-hungry reality. Training large models like OpenAI’s GPT-3 requires immense computational power, with a single generative AI image consuming as much energy as fully charged a smartphone. Discussions around AI’s environmental impact surged after reports revealed that generating a single Studio Ghibli-style image using AI could consume enough energy to power a 60-watt lightbulb for approximately 3 minutes. While this might seem minor, the cumulative effect is staggering. GPT-3’s training alone emitted around 552 metric tonnes of carbon dioxide, equivalent to the yearly emissions of 123 gas-powered cars, or nearly 100 round-trip flights between London and San Francisco. Beyond the software, the hardware enabling AI (servers, GPUs, and specialized processors) comes with its own environmental footprint. Mining the raw materials for these components contributes to soil degradation and pollution, while improper disposal of outdated machines contaminates land and water. AI’s growing thirst for water is another overlooked consequence.Data centers require massive amounts of water to prevent overheating. A single ChatGPT conversation can use about 50 centiliters of water. This becomes particularly concerning in regions like India, where AI is becoming central to digital infrastructure. The United Nations projects that by 2025, India’s per capita water availability will drop below 1,000 cubic meters annually, which is the threshold for water scarcity. This raises a critical question: who secures access to these dwindling resources as AI expands, and who gets left behind?
Who benefits and who pays?
AI appears to promise global benefits, but its socio-economic rewards remain unevenly distributed globally. In 2022, Google’s data centers in Finland ran on 97% carbon-free energy, while their Asian counterparts relied on just 4-18%. This stark contrast reveals deep disparities in access to clean energy. The Global North leverages renewable energy to power advanced AI systems. At the same time, the Global South remains hitched to fossil fuels that intensify pollution and environmental degradation.
This power imbalance extends beyond energy consumption to the very core of AI innovation and development. As of this moment, we appear to be at a crossroads: AI could potentially either foster inclusive, sustainable growth or entrench existing inequalities through techno-colonialism. The development and regulation of AI technologies remain concentrated in the Global North- primarily the United States and Western Europe. According to the Stanford AI Index Report 2024, over 80% of AI policy initiatives originate from North America and Europe. This dominance makes the Global South dependent on foreign technology and governance systems, increasing structural inequalities.
The environmental consequences of AI’s energy demands further compound these inequities. Resource extraction for hardware and e-waste disposal disproportionately affects the Global South, while the Global North reaps the economic rewards. Extracting rare earth minerals like lithium and cobalt, which are crucial for AI hardware, leads to deforestation, pollution of water sources and destruction of habitats. As AI-driven industries scale, their carbon footprint grows alongside them. The environmental burden, however, is rarely shared equally. Emissions linked to training AI models are often outsourced to regions with weaker environmental regulations, shifting the cost onto communities already facing climate stress. The imbalance caused calls for policies that enforce shared environmental responsibilities. Furthermore, countries in the Global South are often reduced to passive consumers, adopting technologies they did not help shape. Without a shift toward equitable participation, the AI revolution risks widening the gap between those who create technology and those who bear its costs.
A Responsibility Divide
An overlooked but important framework in this conversation is ‘common but differentiated responsibility’ (CBDR) principle in international environmental law. Formalized at the 1992 UN Conference in Rio de Janeiro, CBDR acknowledges that while all States share responsibility for environmental protection, their obligations vary based on socio-economic conditions and historical contributions to pollution. This concept holds powerful relevance for AI governance. Applying CBDR to AI suggests that regions leading AI development and innovation should shoulder greater responsibility through technology transfer, equitable regulation, and sustainable practices.
However, building the infrastructure for successful, large-scale AI adoption poses a major challenge for the Global South. Sustainable innovation could help bridge this gap. Researchers from Stanford, Meta AI, and McGill University have developed an impact tracker to measure machine-learning-origin electricity and carbon emissions. For example, training Hugging Face’s BLOOM model released 25 metric tonnes of carbon dioxide despite a nuclear-powered French supercomputer, and continues to emit 19 kg of carbon dioxide daily, equivalent to 54 miles driven in a new car. Some researchers, like those at IBM, suggest moving AI operations to cloud-based data centers to reduce this carbon footprint. India, for example, is witnessing a surge in AI adoption alongside climate commitments. As the demand for AI grows, so does the need for sustainable energy solutions to support its expansion. The National Green Hydrogen Mission aims to accelerate the deployment of green hydrogen as a clean energy source, potentially powering AI with renewables and significantly cutting its carbon footprint.
However, identifying the problem is just the first step. AI models, for example, can be effective without vast data. Domain-specific models optimize resources, enhance efficiency, and reduce environmental impact, aligning AI development with sustainability goals. Measures such as prompt engineering optimize hardware usage and reduce the carbon footprint. Meanwhile, the technology industry is investing in ‘AI for Sustainability’ initiatives, which integrate AI into smarter energy grids, optimize carbon capture and use Life Cycle Assessment tools to track emissions across industries.
Regulation also plays a key role. The European Union’s AI Act requires companies to report energy consumption under Article 40.2. This model could help the Global South balance technological growth with environmental responsibility. However, for such frameworks to be effective, they must be adaptable to local realities and supported by public-private partnerships. These partnerships could combine government oversight with private-sector innovation, fostering responsible AI development without stifling progress. Global governance bodies are also starting to recognize these challenges. The UN AI Advisory Body has proposed an International Scientific Panel on AI to assess its environmental impacts. However, infrastructure limitations and funding gaps remain significant barriers for the Global South — highlighting that technical solutions alone are not enough. Tackling these inequalities demands global cooperation and a commitment to shared responsibility.
Crossroads for AI’s Future
While AI seems to be providing groundbreaking assistance in sustainable solutions such as designing energy-efficient buildings, tracking deforestation, monitoring air quality and improving renewable energy use, there is a growing environmental cost to these benefits. The challenge gets even more complex with the issue of techno-colonialism, where countries in the Global North are often at the forefront of technological advancements. Concurrently, those in the Global South struggle to keep pace.
But the path forward does not have to mirror these inequalities. Addressing this divide means rethinking global AI policies to prioritize accessibility, transparency, and equitable technology transfers. It means empowering local researchers and innovators in the Global South with funding, infrastructure, and decision-making power to build AI systems that tackle their region-specific environmental and social challenges. It also requires holding tech giants accountable for the environmental footprints of their data centers and supply chains; especially when those burdens fall on regions already facing severe climate stress. AI’s future is still being written. The real challenge is whether we can harness AI’s potential without letting its environmental costs fall on those least equipped to bear them.
The views expressed are solely those of the author and are not affiliated with the organization.
