AI and the Environment: What We Know, What Is Being Reported, and What We Still Don’t Know
AI and the Environment: What We Know, What Is Being Reported, and What We Still Don’t Know
AI’s environmental impacts are real. The full picture is also more complicated than many headlines make it sound.
A quick note before we begin: I am not an environmental scientist. The deeper and more complex environmental questions surrounding AI should include environmental scientists, energy experts, researchers, and the communities directly affected.
My focus is responsible and safe AI. That includes helping people understand AI’s environmental impacts so they can ask better questions and make more informed decisions.
AI can feel almost weightless. We type something into a box, receive an answer, and move on with our day.
But that answer did not appear out of thin air.
Behind every AI tool are data centers, computer chips, electrical grids, cooling systems, networks, and water. All of that physical infrastructure has an environmental impact.
That impact is real. It is also more complicated than some of the headlines make it sound.
To have a useful conversation about this, we need to separate three things: what we know, what is being reported, and what we still do not know.
What we know
AI requires a lot of physical infrastructure.
Developing and operating AI systems requires powerful computing equipment. Data centers need electricity to run that equipment and additional energy to keep it cool. Depending on where a data center is located and how its cooling system works, it may also use water directly at the facility and indirectly through electricity generation.
Then there is the hardware itself.
Producing advanced chips and servers requires raw materials, manufacturing, and transportation. Eventually, that equipment also contributes to electronic waste.
The International Energy Agency reported that data centers consumed approximately 415 terawatt-hours of electricity worldwide in 2024. That was about 1.5% of global electricity consumption.
The agency projects that data-center electricity demand could more than double by 2030, with AI as the most important driver of that growth.
The local impact can be especially important. A new data center may represent a relatively small part of worldwide electricity use while still placing significant pressure on a particular electrical grid or local water supply.
That is one reason national percentages do not tell the entire story.
What is being reported
We have all seen the comparisons.
An AI request is described as using a certain number of bottles of water. Training a model is compared with household electricity consumption or the emissions from a flight. An AI search is compared with a traditional internet search.
These comparisons can make an invisible issue easier to understand. They can also create a level of certainty that the available evidence does not always support.
There is no single environmental cost that applies to every AI request. The footprint can change based on the model, the request, the amount of computation, hardware efficiency, server usage, cooling, location, timing, and the source of electricity.
Training a large AI model can require substantial resources. However, training is only one part of the picture. Once millions of people begin using a model every day, its ongoing operation can also become a major part of its total footprint.
There is another side to the reporting as well. AI is being explored as a way to improve electrical grids, forecast renewable-energy production, reduce industrial waste, optimize buildings, monitor ecosystems, and support climate research.
Both things can be true at the same time: AI can consume environmental resources while also helping people address environmental problems.
What we still do not know
We know AI has an environmental footprint. What we do not have is a complete and consistent way to measure it.
Technology companies do not always disclose energy and water consumption at the individual model level. Even when estimates are available, researchers may be measuring different things.
One estimate may focus only on the electricity used by computing equipment. Another may include cooling. Others may try to account for water, hardware manufacturing, transportation, or emissions from the electrical grid. Those different boundaries can produce very different numbers.
We also do not know whether improvements in efficiency will be enough to offset the rapid growth in AI use.
AI systems and computer chips are becoming more efficient. That is important progress. But when technology becomes faster, cheaper, and easier to access, people usually use more of it. This is sometimes called a rebound effect.
In other words, individual AI tasks may become more efficient while the industry’s total consumption continues to grow. That is why projections about AI’s future environmental impact should be understood as scenarios—not guarantees.
This is being taken seriously
Researchers, governments, standards organizations, energy planners, and technology companies are actively studying these issues.
The National Institute of Standards and Technology recommends that organizations assess and document the environmental effects of AI training and operations. Its guidance includes monitoring energy, water, efficiency, and greenhouse-gas indicators, establishing baselines, and evaluating sustainability tradeoffs.
The OECD has called for measurement across the full AI lifecycle. That includes the production, transportation, operation, and eventual disposal of AI equipment.
The International Energy Agency is examining both AI’s growing energy demands and the ways AI may improve energy systems.
This does not mean the problem has been solved. It means the issue is being recognized as a serious part of AI planning and governance.
Responsible AI must include the environment
Responsible AI is often discussed in terms of privacy, fairness, accuracy, transparency, safety, and human oversight. Environmental responsibility belongs in that conversation too.
Organizations do not need to wait for perfect information before they begin asking better questions:
- Are we using AI to solve a meaningful problem?
- Does this task require AI at all?
- Are we using more computing power than the task requires?
- Could a smaller or more efficient system produce an acceptable result?
- Does the provider disclose information about energy, water, or emissions?
- Are we measuring usage and eliminating wasteful processes?
- Are environmental considerations included in procurement and governance?
The answer is not to label every AI interaction as either harmful or harmless. The goal is to understand the tradeoffs and make deliberate decisions.
The bottom line
AI’s environmental impacts are real.
Some of those impacts can be measured today. Others are being estimated with incomplete information. Headlines can help communicate the scale of the issue, but highly specific per-request comparisons should be treated carefully.
AI may also help address environmental challenges. Those possible benefits do not erase the resources required to build and operate it.
Responsible AI means being willing to hold both realities at once.
We should acknowledge what we know, stay honest about what we do not know, and continue asking for better measurement and greater transparency.
Sources
International Energy Agency, “Energy and AI”
Lawrence Berkeley National Laboratory, “2024 United States Data Center Energy Usage Report”
NIST AI Risk Management Framework Playbook, Measure 2.12
OECD, “Measuring the Environmental Impacts of Artificial Intelligence Compute and Applications”
Responsible AI includes environmental responsibility.
Better questions, better measurement, and more deliberate use all matter.
AI LITERACY, MADE PRACTICAL
You do not need to become a developer to use AI well.
AI literacy is a workplace skill: knowing how to choose the right tool, give useful instructions, evaluate the result, protect sensitive information, and recognize when human judgment must take over.
WHY AI LITERACY MATTERS
AI already appears in writing, research, recruiting, customer service, analysis, scheduling, and everyday office software. Understanding the basics helps you:
• Ask better questions and give clearer instructions
• Recognize confident-sounding mistakes
• Protect sensitive information
• Decide which tasks should—and should not—use AI
• Collaborate more effectively with technical teams
• Adapt as your role and industry change
AI literacy is not about knowing everything. It is about knowing enough to work thoughtfully.
FOUR PRACTICAL SKILLS TO BUILD
1. Understand the basics
Learn what generative AI can do, where its answers come from, and why it can produce inaccurate or invented information. Beginner-friendly courses, demonstrations, and workshops can build this foundation without requiring coding.
2. Practice prompting
A useful prompt includes the goal, relevant context, constraints, desired format, and criteria for success. Treat the first response as a draft. Ask follow-up questions, request alternatives, and refine your instructions.
3. Evaluate outputs
Do not confuse fluent language with accuracy. Check facts, calculations, quotations, links, and important recommendations. Review for missing context, bias, privacy concerns, and unintended consequences.
4. Apply human judgment
AI can accelerate work, but accountability remains human. People still need to understand the audience, make decisions, handle exceptions, and recognize when a task is too sensitive or consequential to delegate.
LEARN BY USING AI ON REAL WORK
The fastest way to build confidence is to practice on low-risk tasks you already understand.
Good starting points include drafting an outline, rewriting a message for clarity, summarizing approved notes, brainstorming interview questions, turning a process into a checklist, comparing options, or drafting training materials.
Start small. Compare the AI-assisted result with your usual process. Note what improved, what became harder, and what still required your expertise.
THE FIVE-STEP PRACTICE LOOP
TRY — Choose one repetitive, low-risk task and define the result you want.
CHECK — Review the output for accuracy, usefulness, bias, and missing information.
ADJUST — Improve your instructions or change the workflow based on what you learned.
DOCUMENT — Save effective prompts, verification steps, and examples your team can reuse.
SHARE — Discuss what worked and what failed so learning spreads beyond one person.
KNOW WHEN TO SLOW DOWN
Do not enter confidential employee, customer, health, legal, financial, or proprietary information into an unapproved tool. Do not rely on AI alone for high-stakes decisions. Do not assume a polished answer is correct. And do not automate a broken process before understanding why it is broken.
A responsible beginner is more valuable than a careless power user.
CREATE A LEARNING HABIT
Follow a small number of trustworthy sources. Test one use case at a time. Keep a prompt and workflow library. Share lessons and failures. Revisit policies as tools and risks change. Measure quality and outcomes—not just usage.
A NOTE ON RELIABILITY
Generative AI can fabricate facts, sources, quotations, and explanations. Verification is essential—especially in legal, medical, financial, employment, and other high-impact work.
Use AI to support your thinking, not replace it.
YOUR NEXT STEP
Choose one repetitive, low-risk task this week. Test an approved AI tool, record what happened, and identify one improvement for your next attempt.
That is how AI literacy grows: one practical, responsible experiment at a time.
BUILD PRACTICAL AI CONFIDENCE
Rocky Phoenix AI helps non-technical professionals and teams adopt AI with clear training, responsible workflows, and human-centered guidance.