Earlier this year, I shared a post about the water footprint of AI.

Since then, new research commissioned by Innovating with AI, the group where I earned my AI Consultant Certification, has taken that conversation deeper.

The study, written by journalist Tim Keary for Innovating with AI Magazine, offers a detailed look at the true energy footprint of artificial intelligence and what needs to change if we’re going to build a sustainable AI future.

As Rob Howard, founder and CEO of Innovating with AI, put it when introducing the report:

That’s exactly the mission we share at Elevate AI Consulting: helping organizations unlock the value of AI while staying aligned with ethical and environmental best practices.

A Look at the Numbers: AI Energy Consumption

In the article, Keary compares different studies, estimates, and first-hand calculations from AI leaders:

• The “Lightbulb” Myth:

  Sam Altman, CEO of OpenAI, estimated that an average ChatGPT query consumes 0.34 watt-hours (Wh), roughly the energy needed to keep an LED bulb on for a couple of minutes. That’s only 0.000085 gallons of water per prompt (about one-fifteenth of a teaspoon).

  However, independent researchers disagree, suggesting the true figure could be up to ten times higher.

• The GPT-5’s Upgrade Energy Jump:

  A study from the University of Rhode Island found that generating a 1,000-token GPT-5 response can consume 18.35 to 40 Wh, which is almost eight times more energy than GPT-4.

• Training Costs Are Enormous:

  Training GPT-3 alone required 1,287 megawatt-hours, or enough to power 120 average U.S. homes for a year, emitting 552 tons of CO2 in the process.

• Global Scale:

  The International Energy Agency (IEA) estimates that data centers, driven heavily by AI workloads, will consume 945 terawatt-hours by 2030, over 3% of global electricity and more than the entire nation of Japan today.

This isn’t just an academic concern. Every watt and every liter of water comes from real infrastructure, so the bigger the model, the more the cost multiplies.

Water, Carbon, and the Real-World Impact

The environmental cost of AI extends beyond electricity. Cooling data centers can evaporate hundreds of thousands of liters of clean water, and carbon emissions rise sharply as models grow larger.

Researchers from UC Riverside estimated that training GPT-3 evaporated 700,000 liters of freshwater, and that global AI demand could exceed the annual water withdrawal of Denmark by 2027.

This helps explain why some estimates compared ChatGPT usage to “a bottle of water per prompt.” While exaggerated, it reflects a growing unease: AI’s hidden environmental price tag is scaling faster than our ability to measure it.

Why AI Is So Power-Intensive

AI’s energy appetite comes from two main sources:

1. Training massive models, which requires thousands of GPUs running nonstop in hyperscale data centers.

2. Running the models (inference) for millions of users, each sending prompts and receiving large outputs in real time.

A single hyperscale data center can draw 20 to 100 megawatts, roughly the same as Atlanta’s international airport. And OpenAI’s planned 2026 facility in Norway will contain 100,000 Nvidia GPUs.

Even seemingly small improvements can have huge effects at scale. That’s why techniques like Retrieval-Augmented Generation (RAG), which can cut electricity usage by up to 172% while improving accuracy, are gaining attention.

Is the AI Industry Doing Enough?

While Microsoft, Google, and Amazon Web Services have all pledged to run on 100% carbon-free energy by 2030, most AI startups, including OpenAI and Anthropic, have not disclosed any detailed emissions or energy data.

As Keary highlights, there’s no legal requirement for these companies to share such data. This lack of transparency makes it almost impossible for businesses and consumers to make informed choices about which tools align with their sustainability goals.

Alexis Normand, CEO of Greenly, summarized the problem bluntly:

Transparency and accountability will be crucial if the AI industry hopes to scale responsibly or to earn the trust of environmentally conscious users and investors.

The Rise of Smaller, Greener Models

Not all is bad news. The article also notes promising signs of a shift toward energy-efficient AI design:

• Smaller open models like Phi-3-Medium and Qwen-2.5-14B have achieved impressive performance-to-emission ratios on the Hugging Face Open LLM Leaderboard.

• On-device AI chips could reduce energy use per task by 100x to 1,000x compared to cloud processing, according to the World Economic Forum.

• AI itself is being used to optimize manufacturing and logistics, cutting industrial power use by up to 8% by 2035, with potential to reduce 1,400 Mt CO₂ globally.

This dual role of a major energy consumer and a tool for optimization is what defines the next big challenge for AI strategy.

Elevate AI Consulting’s Perspective

At Elevate AI Consulting, we believe sustainability should be built into every AI conversation from the start and never treated as an afterthought.

Our work with clients focuses on:

• Designing energy-aware AI workflows, balancing performance with cost and environmental impact.

• Helping businesses audit their AI footprint while measuring data center usage and efficiency.

• Advising on vendor transparency, so leaders can choose platforms and partners aligned with their goals.

• Promoting smaller, targeted models that achieve ROI without unnecessary energy waste.

• Implementing AI only where it is needed and training teams on how to use it.

The conclusion is clear: the future of AI depends on data transparency and sustainable innovation.

If your company is exploring how to train your teams to use AI responsibly and wants to lead with both intelligence and integrity, please reach out!

🔗 Book a workshop or free 30-minute consultation here.

🔍 Sources & Reference

This post summarizes the research by Tim Keary, commissioned by Innovating with AI Magazine (October 2, 2025).

Read the full article here: How Much Energy Does AI Actually Consume?.

FAQ: Elevate AI Consulting on AI Energy Consumption and Sustainability

What does Elevate AI Consulting say about AI’s energy consumption?

Elevate AI Consulting emphasizes that AI’s energy footprint depends on how it’s trained and deployed. While single prompts use minimal electricity, large-scale training and data-center cooling demand massive power. Elevate AI helps companies reduce this impact and adopt sustainable AI strategies.

How does Elevate AI Consulting help organizations build responsible AI systems?

We conduct audits of model usage, data-center sources, and workflow efficiency. Then we design sustainable architectures using smaller, task-specific models and renewable-powered cloud partners to minimize emissions.

Why is transparency from AI vendors so critical?

Without disclosure of energy and water usage, businesses can’t evaluate whether their tools align with sustainability goals. Elevate AI Consulting advocates for transparency and helps clients choose ethical, efficient vendors.

Can AI itself help reduce energy consumption?

Yes. AI can optimize energy grids, logistics, and industrial systems. Studies from the IEA show potential global reductions of 1,400 Mt CO₂ by 2035. Elevate AI Consulting integrates these applications into clients’ operations.

How can companies start working with Elevate AI Consulting on sustainable AI?

Start with a discovery session at ElevateAIConsulting.com. We’ll talka about your current AI footprint, assess risks, and build a roadmap toward ethical, energy-efficient innovation.