Jevons’ paradox — a 160-year-old economic theory first described by English economist William Stanley Jevons — suggests that when technological advances improve the efficiency of a resource, we paradoxically end up using more of that resource, not less. In the 1860s, Jevons observed this with coal: as steam engines became more efficient, coal usage increased because the improved efficiency lowered costs and expanded demand.

Today, this paradox sits squarely at the intersection of artificial intelligence and climate change. As AI systems become more efficient and accessible, their total energy use continues to rise — mirroring the pattern Jevons identified.

Microsoft CEO Satya Nadella all but celebrated this dynamic when DeepSeek was announced in January 2025: “As AI gets more efficient and accessible, we will see its use skyrocket, turning it into a commodity we just can’t get enough of,” he wrote on social media.

AI’s growing energy footprint

The energy demands of artificial intelligence systems have reached worrisome levels—and continue to rise steeply. In 2023, AI-related servers consumed approximately 40 terawatt-hours (TWh) of electricity, up from just 2 TWh in 2017 — a twentyfold increase in six years. This trend shows no sign of abating. Forecasts suggest AI operations could exceed 90 TWh annually by 2026 and reach 146 TWh by 2027.

By 2030, global data centers — driven largely by AI workloads — are projected to consume between 945 and 1,050 TWh per year, more than double today’s usage. At that scale, data centers would rank among the world’s largest electricity consumers, comparable to entire industrialized nations. For perspective, the 146 TWh projected for AI alone by 2027 would nearly match Malaysia’s total electricity consumption in 2023 (148 TWh), and more than double Chile’s 2014 usage (69 TWh).

The energy impact is also evident at the individual level. A single ChatGPT query consumes roughly as much electricity as a light bulb running for twenty minutes. ChatGPT also produces 260,930 kilograms of carbon dioxide every month. As adoption accelerates across sectors, the challenge is compounded by the geographic concentration of data centers. These facilities are often clustered in specific regions, placing acute pressure on local energy infrastructures.

Rising demand is also driving up electricity prices. For data centers, electricity constitutes the largest ongoing operational cost — accounting for 46% of enterprise facility spending and 60% for service providers. According to Bain & Company, meeting the projected demand could require more than $2 trillion in new energy generation resources globally, potentially triggering what analysts are calling an “extraordinary” rise in electricity bills.

Can AI help optimize energy use across sectors — and in doing so, offset its own intensifying footprint?

The opportunities

Buildings account for a substantial share of global energy use — approximately 18% of total consumption, according to the International Energy Agency. Much of this demand stems from outdated and inefficient HVAC systems, which often fail to respond dynamically to changing weather and occupancy patterns, resulting in significant energy waste.

AI offers a powerful tool to address this inefficiency. A 2024 study estimates that AI integration could reduce energy consumption and carbon emissions from buildings by at least 8%. This promise lies in AI’s ability to deliver continuous, predictive optimization in real-time.

Consider 45 Broadway, a 32-story office tower in Manhattan built in 1983. In February 2022, it became the first building in New York City to implement BrainBox AI’s smart energy management system. Using a dense network of sensors to collect data on temperature, humidity, sun angle, wind speed, and occupancy patterns, the system makes automated HVAC adjustments every five minutes. With a predictive accuracy of 99.6%, BrainBox’s AI can anticipate the building’s future state and act accordingly.

After 11 months, the results were compelling: a 15.8% reduction in HVAC-related energy consumption, over $42,000 in savings, and a reduction of 37 metric tons of CO? equivalent.

Yet this success raises a critical question: how much electricity did the AI itself consume to produce these savings? And what was its own carbon footprint?

Source: A Review of Green Artificial Intelligence: Towards a More Sustainable Future; Bolón-Canedo, Fernández, Cancela, Alonso-Betanzos

Green AI: Can AI clean up after itself?

One promising response to AI’s growing energy appetite is the development of Green AI — a concept encompassing two complementary strategies. “Green-in AI” focuses on designing energy-efficient models and computing systems to reduce AI’s environmental impact directly. Meanwhile, “green-by AI” refers to deploying AI to advance sustainability across other sectors, from clean energy grids to smart agriculture.

Efforts to green the AI ecosystem are multiplying. Researchers have catalogued more than 55 initiatives worldwide across six major domains: cloud optimization, model efficiency, carbon footprinting, sustainability-driven AI development, open-source platforms, and community-building for environmental AI. The common goal: reduce energy and carbon costs without compromising performance or accessibility.

Technical innovations are central to these efforts. Efficient algorithms — enabled by model pruning, quantization, and knowledge distillation — can dramatically cut computational overhead. The choice of model architecture also matters: sparse models and efficient architectures can maintain or improve outcomes while using less energy. On the hardware front, next-generation GPUs and Tensor Processing Units (TPUs) with higher FLOPS-per-watt ratios offer significant gains in efficiency.

Yet a critical question remains: how much energy does AI require to deliver these environmental benefits — and can those benefits ultimately outweigh the system-wide energy costs?

In February 2025, momentum around Green AI took a major step forward with the launch of the Coalition for Environmentally Sustainable Artificial Intelligence in France. Led by the UN Environment Programme (UNEP) and the International Telecommunication Union (ITU), the coalition brings together over 100 stakeholders — 37 technology firms, eleven national governments, and five international organizations — to coordinate standards, frameworks, and methodologies that align AI development with global environmental goals.

These are important steps, but they raise a deeper challenge: can Green AI scale quickly and deeply enough to keep pace with the exponential growth of AI adoption — and ultimately offset or even outweigh its own emissions?

Green-by AI: Applying intelligence to nature’s defense

A wave of innovation is demonstrating how artificial intelligence can serve as a powerful ally in environmental protection. Across universities, startups, and research labs, new initiatives are emerging that harness AI to monitor ecosystems, accelerate climate research, and enable smarter conservation.

In the fight against illegal fishing, UK-based OceanMind uses machine learning, satellite imagery, and advanced data analytics to detect suspicious vessel behavior in real time. At Leeds University, researchers have developed an AI system capable of analyzing satellite images to measure iceberg melt rates up to 10,000 times faster than human capabilities — processing each image in just 0.01 seconds. This acceleration in analysis aids predictions of sea-level rise and improves understanding of how meltwater disrupts ocean currents.

Meanwhile, London startup Olombria is exploring the use of hoverflies as alternative pollinators, deploying AI to track and optimize their behavior. In the field of biodiversity, scientists are building AI-enhanced acoustic monitoring systems that identify species and track populations by analyzing animal vocalizations, enabling large-scale, non-invasive ecosystem monitoring.

Major tech players are also stepping in. In early 2025, Google launched its AI for Nature Accelerator to support a new generation of sustainability-driven ventures, underscoring the growing alignment between climate tech and AI innovation.

These examples illustrate the immense promise of “green-by AI.” Yet a critical question remains: how much energy does AI require to deliver these environmental benefits — and can those benefits ultimately outweigh the system-wide energy costs?

Increased energy demand, expanding clean supply

As the AI revolution accelerates — and global electrification expands across sectors—the world’s appetite for energy is surging. Speaking before the Senate Committee on Commerce, Science and Transportation on May 8, 2025, Sam Altman called it the “dual revolutions”. The good news: renewable energy capacity is scaling at an unprecedented rate.

Meeting this demand sustainably will require not just more renewables, but smarter deployment.

According to the International Renewable Energy Agency (IRENA), global renewable power capacity grew by a record-breaking 585 gigawatts in 2024, marking an annual growth rate of 15.1%. Solar has emerged as the engine of this transition, doubling its output in just three years. Together, renewables and nuclear passed a major milestone in 2024 — surpassing 40% of global electricity generation for the first time since the 1940s.

This trend is expected to continue. Renewables-based electricity is projected to overtake coal-fired power in 2025. Wind and solar alone will surpass nuclear by 2026. By 2030, renewable sources are forecast to deliver 46% of global electricity. In the U.S., battery storage capacity nearly doubled in 2024 and is set to grow another 47% in 2025 — bolstering grid stability as solar and wind scale further.

Renewable energy demand growth by sector, 2023-2030; Source IEA

Still, challenges remain. Rapidly rising energy demand — driven in part by AI and digital infrastructure — is outpacing even this clean energy growth. Global electricity demand grew by 4.3% in 2024 alone, the fastest rate ever recorded and well above both overall energy demand and GDP growth.

Meeting this demand sustainably will require not just more renewables, but smarter deployment. That includes expanding transmission infrastructure, accelerating energy storage adoption, removing policy barriers, and ensuring equitable access to clean energy finance — particularly in the Global South.

With thoughtful design, responsible regulation, and intentional deployment, AI could become a powerful ally in our climate transition — optimizing buildings, supply chains, ecosystems, and energy systems. But if we fail to plan and govern wisely, it risks becoming a force that deepens the crisis it might otherwise help solve.

The stakes couldn’t be higher: the choices we make now will determine whether AI becomes a cornerstone of climate progress — or an accelerant of ecological strain.