This blog was co-created with AI tools to ensure accuracy, then refined by our human team for clarity.
Prefer to listen?
Join Alex and William as they explore the real carbon impact of artificial intelligence in a warm, insight-packed conversation. From surprising emissions data to values-led decision frameworks, this audio experience brings the article to life through clarity, compassion and practical relevance.
Is AI Greener Than Humans?: Audio Overview Transcription
Alex: Welcome back to The Umuti Podcast. Today we’re asking a big question that doesn’t get nearly enough airtime—what’s the true environmental cost of artificial intelligence?
William: Great to be here, Alex. I think a lot of people see AI as this invisible, weightless helper. But every interaction—every prompt, every image generation—is powered by very real, very physical infrastructure.
Alex: Right. We’re talking about servers, data centres, cooling systems. Even minerals that have to be mined. There’s a footprint. It’s just one we don’t see.
William: Exactly. And as more conscious leaders start to integrate AI into their day-to-day, it’s essential to understand the systems beneath the software. Because otherwise, we risk outsourcing our labour to machines without seeing the hidden ecological costs.
Alex: And yet, it’s not all doom and gloom. There’s emerging research that shows AI, when compared to human output in some tasks, can actually be much more energy efficient. It depends on how it’s used.
William: That’s the nuance we need. This conversation isn’t about saying AI is good or bad. It’s about helping people ask better questions—and use these tools with more clarity and care.
Alex: So let’s talk about those numbers. Is AI really greener than humans?
William: Well, when it comes to tasks like writing and illustrating, the data is startling. One study showed that BLOOM, an open-source language model, emits about 1,400 times less CO₂e per page than a US-based writer. ChatGPT? Roughly 1,100 times less.
Alex: That’s wild. And even more so when you factor in illustrators. Tools like DALL·E 2 and Midjourney use 300 to nearly 3,000 times less energy per image compared to professional human illustrators.
William: It’s a big shift. Especially for businesses producing content at scale. AI can take on the high-volume, repetitive tasks, and do it with far less environmental load. But—
Alex: And there’s always a “but” here.
William: Always. These figures don’t tell the full story. Most emission stats leave out the wider ecosystem—like the energy used by the dev teams, the carbon cost of manufacturing GPUs, and even the social impact of replacing human creativity.
Alex: It’s a good reminder. This isn’t just a carbon equation—it’s also about values, livelihoods, and the ripple effects of our choices.
William: That’s where discernment comes in. We shouldn’t be thinking in binaries. It’s not “replace humans with AI” or “ban tech altogether.” It’s about integration—where it makes sense, and where it brings genuine efficiency.
Alex: And some parts of AI are far from efficient, right? Especially training those big models.
William: Yes. Training GPT-3, for instance, emits over 550 metric tons of CO₂e. That’s comparable to the lifetime emissions of five petrol-powered cars. It’s a one-time load, but it’s huge.
Alex: Whereas inference—when AI responds to our prompts—is less intensive per task. But it happens millions of times a day, so the impact adds up.
William: That’s exactly the trade-off. And here’s another layer: about 30% of training energy is wasted due to poor scheduling or low GPU usage. It’s not just about how much energy is used, but how much is wasted.
Alex: I love that you brought that up. Because sustainability isn’t just about output—it’s also about restraint. About asking the system to do less, more wisely.
William: There’s a whole field called Green AI that’s emerging. It focuses on building models that use fewer resources, reusing pre-trained systems, and scheduling workloads when cleaner energy is available.
Alex: This feels like the real heart of the conversation. Not just measuring emissions, but rethinking how we design and deploy tech in the first place.
William: Absolutely. We also need to zoom out. The infrastructure behind AI is massive. Data centres alone could consume nearly 1,000 terawatt-hours by 2030.
Alex: And where those centres are built matters. If they’re powered by coal-heavy grids, then all the AI efficiency at a micro level gets wiped out by macro emissions.
William: There’s also water for cooling, rare earth minerals for hardware, and the constant churn of electronics. These aren’t just carbon costs—they’re ecosystem costs.
Alex: So if AI is going to help lower emissions, it has to be used with intention. Especially in areas like content creation, where it can really outperform human benchmarks in energy efficiency.
William: Yes. When used for first drafts, summaries, batch tasks—it can massively cut the need for manual repetition. That doesn’t replace creativity. It supports it.
Alex: And the Green AI movement is showing us that better design can start from the ground up. Smarter architecture, cleaner scheduling, more ethical tooling.
William: Exactly. This isn’t about waiting for perfect conditions. It’s about doing what we can now—embedding sustainability into the foundations, just like you’d start regenerative farming with the soil.
Alex: That’s such a powerful metaphor. And for leaders listening—there’s a framework, right? A way to evaluate whether your AI stack aligns with your values?
William: There is. You start by measuring what matters—ask vendors for emissions data. Then use AI where it makes the most impact. Don’t automate just for the sake of it.
Alex: Optimise your workflows, turn off background processes, reduce duplication. Educate your team too—so they know when not to use AI.
William: And keep reviewing. Tools evolve. Systems need updates. But your values—that’s your compass.
Alex: I want to close with this: AI is not immaterial. It may feel invisible, but it has a body. A cost. And also, an opportunity.
William: Used well, it can support regenerative growth. But it requires conscious leaders—people willing to choose wisely, not just conveniently.
Alex: So, if you’re building something that matters, ask yourself—are your tools aligned with your vision? Because every system you build is also a reflection of what you believe.
William: Beautifully said. Let’s build wisely. Let’s build for more than just speed.
Alex: Thanks so much for joining me, William.
William: Always a pleasure, Alex.
AI’s Carbon Footprint
Artificial intelligence is everywhere now. It generates emails, designs logos, analyses patterns, and writes blog posts like this one. It feels seamless. But the truth is every digital interaction draws from a vast physical network of servers, processors, and energy sources. The result is an environmental cost that is often invisible, yet far from negligible. (We did the math on AI’s energy footprint. Here’s the story you haven’t heard.)
As conscious service leaders begin weaving AI into their work, it is vital to understand the systems beneath the surface. Each query, each generated image, each model training cycle runs on electricity, water, and minerals. (AI has an environmental problem. Here’s what the world can do about it)
If we do not question the origin and impact of these resources, we risk building efficiencies that come at a hidden ecological price.
Yet the story is not all bleak. Compared to traditional methods of writing, illustrating, and processing, AI can in some cases be radically more energy-efficient. Studies show that AI-generated content often emits a fraction of the carbon associated with human outputs.
That opens up space for hope. But only if the efficiency gains are paired with integrity, and used with intention.
This article explores the full environmental lifecycle of AI. You will:
- Discover how energy consumption breaks down across training and inference. (Can AI Be More Eco-Friendly Than Humans? A New Study Says Yes)
- Learn how AI’s impact compares to that of humans in common creative and analytical tasks.
- Explore emerging ideas around Green AI, carbon-aware infrastructure, and responsible adoption.
The goal is not to argue for or against AI. The goal is to equip you, as a thoughtful leader, with the insight needed to ask better questions. AI’s footprint can be light, but only if we step forward with clarity and care.
Table of Contents
Is AI Really Greener Than Humans? A Comparative Overview
Want to see the real carbon difference?
Download Umuthi’s AI vs Human: Carbon Impact Comparison Sheet here:
Writing and Illustrating with AI vs. Humans
A landmark study published in Scientific Reports found that AI systems like BLOOM and ChatGPT produce dramatically lower emissions than their human counterparts:
- BLOOM emits around 1400 times less carbon dioxide equivalent per page than a US-based human writer
- ChatGPT emits 1100 times less CO₂e per page compared to a US-based writer
- ChatGPT emits 130 times less CO₂e per page compared to an Indian-based writer
- Even the energy consumed while a human types or pauses between thoughts can exceed what the AI uses to complete a full page
(The carbon emissions of writing and illustrating are lower for AI than for humans)
AI Illustration Efficiency
When it comes to illustrations, AI tools like DALL-E2 and Midjourney also show remarkable energy efficiency:
- Their emissions per image range from 310 to 2900 times lower than those of professional human illustrators
- While generating images requires more power than generating text, AI still significantly outperforms human benchmarks in this area
These numbers reveal a meaningful shift.
For repetitive, high-volume creative tasks, AI holds the potential to reduce emissions at scale without compromising output quality.
What These Numbers Don’t Tell You
Despite the promising comparisons, these figures do not tell the whole story. Most published emission stats exclude the wider ecosystem costs of AI, such as energy consumed by support teams, emissions from software development, and the embodied carbon in hardware manufacture.
There are also social dimensions that pure carbon calculations miss. Job displacement, legal uncertainties, and cultural implications of replacing human creativity with machine output do not appear in carbon ledgers.
This is not a binary debate. AI should not be framed as a substitute for human labour in all cases.
Instead, it should be integrated where it brings genuine energy savings and efficiency. And that integration must come with discernment.
Umuthi insight: True sustainability lives in the tension between innovation and intention. AI’s energy advantage is compelling, but it is not license for unchecked automation. Like planting a new species in a shared ecosystem, introducing AI requires care, compatibility, and a long view of the consequences.
Where AI’s Environmental Footprint Gets Complicated
Training vs. Inference
Training vs. Inference AI's carbon impact begins long before we ever type a prompt. The process of training large language models like GPT-3 demands immense computing resources. One training run of GPT-3 emits an estimated 552 metric tons of CO₂e, which is roughly equal to the lifetime emissions of five petrol-powered cars.
Inference at Scale
The inference stage is the period when the trained model generates responses to user input. This stage uses less energy than training. However, inference happens millions of times a day across countless systems. So while it is lighter per task, the global scale of usage makes its cumulative impact significant.
Full Lifecycle Assessment
The distinction matters. A one-time training load can be amortised across years of use. But inference represents a continuous energy drain. This is why understanding the full lifecycle is so important when assessing environmental cost.
A Closer Look at AI’s Carbon Cost
Understanding the energy behind AI is not always straightforward. Training large models can use as much power as five petrol cars over their lifetime. But what does that really mean in context?
This short video by BBC News offers a grounded, real-world view of how AI’s carbon footprint compares to other sectors. It breaks down the numbers with clarity and speaks to the bigger questions we’re asking in this section.
It’s a useful companion if you’re reflecting on how scale, intention, and infrastructure shape the environmental story of AI.
Wasted Energy in AI Systems
Even within the training process, there is a surprising amount of inefficiency. A 2024 study from the University of Michigan revealed that up to 30 percent of training energy is wasted, often due to poor GPU utilisation and outdated scheduling methods. (Up to 30% of the power used to train AI is wasted. Here’s how to fix it)
Tools like Zeus, an open-source optimiser developed by PyTorch, are helping to reclaim that waste. By fine-tuning how AI models allocate energy use during training, Zeus and similar innovations can significantly cut emissions without sacrificing performance.
Green AI is an emerging discipline that aims to make machine learning more efficient and less environmentally damaging by reducing emissions throughout the entire development cycle.
Umuthi insight: Not all efficiency gains come from output. Some begin with restraint. Before we ask AI to do more, we must ask it to do better. Addressing energy waste is not just about saving resources. It is about building systems that honour both innovation and impact.
The Bigger Picture
945
Terawatt-hours
Global data centre electricity use by 2030
50%
Growth
AI’s share of US electricity demand growth by decade end
2028
Tipping Point
When AI may consume over half of all data centre power
AI’s environmental story does not stop with carbon emissions per task. It is embedded in a larger infrastructure narrative that spans continents, climate policies, and power grids. Data centres are the backbone of every AI tool. Their energy demands are growing quickly.
According to the International Energy Agency, global data centre electricity use could reach 945 terawatt-hours by 2030. (Energy and AI: Executive summary.)
In the United States alone, AI could account for nearly half of all projected electricity demand growth by the end of this decade. By 2028, more than half of all data centre power may be linked to AI-driven processes. These figures remind us that efficiency per query does not always translate into overall environmental benefit. If growth outpaces our ability to green the grid, the net impact will remain heavy.
Geographic placement matters as well. Data centres located near coal-heavy grids generate far more emissions than those powered by renewables. Without deliberate alignment between AI infrastructure and clean energy sources, we risk offsetting the sustainability gains AI makes at the micro level.
Beyond energy, AI systems also consume:
Water Consumption
Used extensively to cool servers, especially in heat-intensive regions
Rare-earth Minerals
Essential to manufacturing chips and GPUs, which often come from unsustainable or exploitative supply chains
E-waste Generation
Short-lifecycle electronics, leading to rapid e-waste as hardware is frequently replaced to keep up with demand
These impacts are less visible but just as important. They reflect an ecosystem cost that can be easy to overlook in carbon-centric discussions.
Umuthi insight: Just as a single tree cannot restore a forest, individual query efficiency cannot fix extractive infrastructure. Conscious innovation means looking not only at outputs, but also at inputs, processes, and aftereffects. To build truly sustainable systems, we must connect our tools to ecosystems that regenerate, not deplete.
When AI Can Help Lower Environmental Costs
AI’s reputation as an energy-hungry technology is well earned. But in specific contexts, it can actually reduce emissions and resource consumption when used intentionally.
Lifecycle Efficiency of Creative Tasks
One of the clearest examples is in content creation. Studies have shown that for tasks like writing blog posts or generating illustrations, AI can outperform human benchmarks by hundreds or even thousands of times in terms of energy savings. This is particularly valuable for businesses that produce content at scale. When AI is used for first drafts, summaries, or batch content production, it eliminates the need for repeated rounds of labour-intensive iteration.
Using AI in these roles does not mean replacing human creativity. It means streamlining low-value or repetitive elements so that human time can be spent on refinement and strategy. The energy savings from cutting down on hours of human processing adds up.
Green AI and the Push for Smarter Design
Another area of promise lies in the design of AI systems themselves. The Green AI movement encourages the industry to prioritise efficiency from the start. That includes:
- Choosing model architectures that require less computation
- Reusing pre-trained models rather than training from scratch
- Scheduling AI workloads when cleaner energy is available
- Using tools like Zeus to profile and optimise GPU energy use
Green AI is not about slowing innovation. It is about designing systems with the whole lifecycle in mind. When you reduce waste at the foundation, the benefits compound over time.
Umuthi Insight: A conscious business does not wait for perfect conditions. It begins optimising with what it has. In the same way that regenerative agriculture starts with the soil, sustainable technology starts at the model layer. When we embed care into the foundation, growth becomes more ethical by design.
A Conscious Leader’s Framework for Evaluating AI Tools
To use AI ethically and sustainably, conscious leaders need more than carbon comparisons. They need decision-making tools that reflect their values, business goals, and ecosystem responsibilities. Our Ethical AI pillar page offers strategic guidance on integrating sustainability into everyday AI decisions.
Here is a five-part framework for evaluating AI tools through a conscious and regenerative lens:
1. Measure what matters
- Ask vendors for emissions data or model training disclosures
- Choose tools that are transparent about their energy use
2. Use AI where it makes the most impact
- Delegate high-volume, energy-intensive tasks like content drafting or document sorting
- Keep humans in the loop for sensitive, values-driven, or relationship-centred work
3. Optimise your workflows
- Use tools like Zeus to monitor GPU energy performance
- Turn off background inference where it is not needed
- Reduce duplication across AI and human systems
4. Educate your team
- Build literacy around AI lifecycle impacts and sustainability considerations
- Offer training on when not to use AI to conserve energy and uphold ethical clarity
5. Revisit your stack regularly
- Technologies evolve fast. Review tools every 6 to 12 months
- Replace bloated or inefficient systems with leaner, purpose-fit models
Check in with how your business is doing with Umuthi’s AI Sustainability Scorecard
Umuthi Insight: Responsibility begins with awareness, but it matures into action. The tools we adopt today shape our systems tomorrow. A conscious leader understands that operational decisions are climate decisions too. By choosing wisely, we do more than reduce emissions. We lead by example.
FAQ's
1. Does using ChatGPT contribute to climate change?
Yes. Every AI interaction requires energy. However, studies suggest that for tasks like writing or illustration, the emissions per output are significantly lower than traditional human methods.
2. Is AI really more energy efficient than people?
For high-volume tasks like drafting or image generation, AI systems like ChatGPT and Midjourney emit hundreds to thousands of times less CO₂e than human equivalents. But these figures depend on use case and frequency.
3. What makes an AI tool sustainable?
A sustainable AI tool is designed with lifecycle emissions in mind. It may use energy-efficient models, optimise training workflows, rely on renewable-powered data centres, or provide emissions transparency to users.
4. How can I reduce the impact of the AI tools I use?
Use AI selectively. Avoid redundancy. Choose tools that report their energy use. Where possible, schedule heavy workloads during low-demand times or when cleaner power is available.
5. Should I avoid using AI altogether to help the planet?
Not necessarily. The focus should be on responsible use. Use AI where it replaces resource-heavy workflows, not where it duplicates them.
Our Final Thoughts
AI is not immaterial. It may feel weightless, but its impacts are grounded in the real world. From server energy and cooling systems to rare minerals and e-waste, every AI interaction carries an environmental cost. Yet within that complexity lies an opportunity. Used wisely, AI can become a tool for energy savings, efficiency, and regenerative growth.
To recap:
- AI systems can drastically outperform humans in terms of emissions per page or image
- The real challenge lies in scale, infrastructure, and rebound effects
- Tools like Zeus and the Green AI movement show there is momentum for optimisation
- Conscious leaders can shape this momentum through strategic decisions and team education
The question is no longer whether AI consumes energy. It is whether we are willing to shape its development in line with planetary boundaries.
A Final Step
Review your own digital ecosystem. Are your tools optimised? Are you choosing AI intentionally, or by default?
Need a guide?
Umuthi’s RootScan offers a practical, self‑help roadmap grounded in the same principles that shaped our own journey from restless agency life to rooted impact.
Align your growth with what matters most.
Umuthi Final Insight:
Sustainability is not a feature. It is a practice. When technology grows with intention, it becomes part of something greater. Not just a system of outputs, but a living ecosystem of values, decisions, and impact.
References:
- Tomlinson, B., Black, R. W., Patterson, D. J., & Torrance, A. W. (2024). The carbon emissions of writing and illustrating are lower for AI than for humans. Scientific Reports, 14, 54271. https://doi.org/10.1038/s41598-024-54271-x
- Mondillo, M. (2025, January 9). Can AI Be More Eco-Friendly Than Humans? A New Study Says Yes. LinkedIn. https://www.linkedin.com/pulse/can-ai-more-eco-friendly-than-humans-new-study-says-yes-mondillo-md-e9eyf
- Person, S. (2025, April 17). The Environmental Cost of AI-Generated Images (and how they compare to human-created art). LinkedIn. https://www.linkedin.com/pulse/environmental-cost-ai-generated-images-simple-person-me-simone-5hxef
- University of Michigan. (2024, November 6). Up to 30% of the power used to train AI is wasted. Here’s how to fix it. Michigan News. https://news.engin.umich.edu/2024/11/up-to-30-of-the-power-used-to-train-ai-is-wasted-heres-how-to-fix-it/
- International Energy Agency. (2025). Energy and AI: Executive summary. https://www.iea.org/reports/energy-and-ai/executive-summary
- United Nations Environment Programme. (2024, August 10). AI has an environmental problem. Here’s what the world can do about it. UNEP. https://www.unep.org/news-and-stories/story/ai-has-environmental-problem-heres-what-world-can-do-about
- Tripp, C. (2024). Measuring the Energy Consumption and Efficiency of Deep Neural Networks: An Empirical Analysis and Design Recommendations. arXiv. https://arxiv.org/abs/2403.08151
- García, J., et al. (2025). Towards an Energy Consumption Index for Deep Learning Models: A Comparative Analysis of Architectures, GPUs, and Measurement Tools. Sensors, 25(3), 846. https://doi.org/10.3390/s25030846
- Tomlinson, B., et al. (2024). The Carbon Emissions of Writing and Illustrating Are Lower for AI Than for Humans. SSRN. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4778246
- MIT Technology Review. (2025, May 20). We did the math on AI’s energy footprint. Here’s the story you haven’t heard. https://www.technologyreview.com/2025/05/20/1116327/ai-energy-usage-climate-footprint-big-tech/
- Strategic Energy. (2025, April 14). Data center energy consumption will double by 2030. https://strategicenergy.eu/data-center/
- Hesser, L. (2024, April 30). Flawed ‘LCA’ study on AI gives false impression about the state-of-the-art. IS4IE. https://is4ie.org/blog/1903
- Brussels Times. (2025, April 10). Data centres will double global electricity demand by 2030 – IEA. https://www.brusselstimes.com/1528980/data-centres-will-double-global-electricity-demand-by-2030-iea
- University of Kansas. (2024, February 4). Study: AI writing, illustration emits hundreds of times less carbon than humans. https://news.ku.edu/news/article/study-ai-writing-illustration-emits-hundreds-of-times-less-carbon-than-humans
- PlanBe.Eco. (2024, May 9). AI’s carbon footprint – how does the popularity of artificial intelligence affect the climate? https://planbe.eco/en/blog/ais-carbon-footprint-how-does-the-popularity-of-artificial-intelligence-affect-the-climate/
- Scientific Reports (Nature). Comparative analysis of AI and human emissions. https://www.nature.com/articles/s41598-024-54271-x
