Building with AI When You Care About the Planet

I’ve been feeling a growing tension.

On one hand, I’m excited by what’s possible with AI, especially the speed, the flow, the power to prototype and test ideas in minutes. Tools like Claude, Cursor, and Replit have changed how I build. It feels like a new creative language.

But here’s the rub: every time I spin up an LLM, a quiet thought lingers: what is this costing the planet?

And not in some vague, far-off way. We now know that AI has a significant planetary footprint.

• Training GPT-3 generated around 552 metric tons of CO₂e, the equivalent of driving a car around 1.24 million miles (Source: OpenAI GPT-3 Paper, 2020 + analysis by Lambda Labs)
• Inference (the everyday querying and generating) now accounts for 90–95% of the carbon footprint of large language models (Source: Berkeley/Google DeepMind, 2022)
• AI infrastructure is extremely water-intensive. Microsoft disclosed using 1.7 billion gallons of water in 2022, largely driven by cooling needs from AI projects like OpenAI. (Source: AP News, 2023)
• OpenAI’s ChatGPT consumed roughly 564 megawatt-hours of electricity in January 2023 alone.That’s the monthly usage of 17,000 average UK homes (Source)
• AI workloads require specialized chips (GPUs, TPUs), which are resource-intensive to manufacture, involving rare earth metals and large energy inputs.
• If AI usage grows unchecked, it could account for up to 3.5% of global electricity use by 2030 — rivaling aviation’s current carbon footprint (Source: International Energy Agency + University of Massachusetts Amherst)

So yes, every time I run a prompt, it’s not “free.” It’s just keeping the planetary costs invisible.

ChatGPT is a large and energy-intensive language model. The equivalent of each query was estimated at 4.32 g of CO2 (for comparison, a Google search is 0.2g per query). According to this calculator, 16 queries = the emissions generated by boiling a kettle. If each unique visit results in an average of 10 queries, that’s 15 trillion queries per month. (source)

AI Might Be One of Our Best Climate Tools

If used wisely, AI could actually help us dig out of this crisis and even restore what we’ve lost. Some of the most exciting, regenerative projects right now sit at the crossroads of climate and AI:

• 🐋 Project CETI is using AI to decode the language of sperm whales, opening entirely new doors for conservation.
• ⚡️ Tapestry uses AI to intelligently orchestrate power across the grid, helping homes and businesses reduce their carbon usage in real-time.
• 🌍 Restor maps reforestation and rewilding efforts worldwide using AI + satellite data offering hope and visibility to grassroots restoration projects.
• 🌫️ Climate TRACE, founded by Al Gore, uses machine learning to track emissions globally with startling accuracy and transparency.
• 🌊 BlueDot uses AI to detect disease outbreaks and climate risk patterns, helping governments act early.

These are signs of a different approach where AI isn’t just about extraction or efficiency, but rather about understanding, regeneration, and protection.

So… How Do We Build Responsibly?

If you’re a builder, you don’t need to opt out. But you do need to build with intention. Here are some ways I’m learning to hold both creativity and care:

If you’re a designer:

• Prototype with lower-impact tools where possible (anyway it always pays to start with pen and paper).
• Cache generative responses in early testing to avoid repeated calls.
• Push for “AI transparency” in UX: tell users when a feature is powered by AI and consider showing environmental impact too.

If you’re an engineer:

• Use model distillation or quantization to reduce energy use.
• Route workloads through low-carbon cloud regions (yes, it varies).
• Think like a systems designer: can this result be cached or shared instead of re-generated?
• Think "like an engineer and not like a magician" I read this somewhere this week, which makes a lot of sense. It involves breaking down complex problems into smaller, manageable components, using scientific principles and data analysis, and focusing on tangible results rather than relying on intuition or guesswork. 

If you’re a PM:

• Add carbon cost to your prioritization framework.
• Push back on AI overuse especially when it’s a shiny add-on rather than a meaningful contribution.

It’s not about being perfect but it is about staying aware and about bringing climate responsibility into the product room.

Try out my little prompt cost calculator here: https://velm-prompt.lovable.app/

What Are AI Companies Doing About This?

Some AI companies are starting to address their environmental impact but action is uneven and often opaque.

• Efficiency first: Labs like Meta and Google are building smaller, faster models (like LLaMA 2 and Gemini Nano), and tools like Hugging Face highlight energy-efficient options.
• Carbon claims: Microsoft has pledged to be carbon negative by 2030. Google provides regional carbon-free energy scores for its cloud. OpenAI hasn’t published GPT-4’s carbon or water footprint.
• Water use is rising: Microsoft used over 1.7 billion gallons of water in 2022, largely for AI cooling. Some providers are experimenting with recycled or closed-loop systems.
• Offsets ≠ solutions: Companies like Google and Amazon purchase renewable energy, but that doesn’t eliminate real-time energy or water impact.
• Accountability is lagging: We still lack standard disclosures for training and inference emissions, or eco-labels that let users understand the true cost of AI tools.

If AI is going to help solve the climate crisis, we need more than good intentions. We need transparency, standards and pressure from the people building and using these tools.

It is true that using technology generally has a negative impact on your carbon footprint, in fact if you check the stats here you can see that for image creation and even writing, not using AI causes more issues. I am not sure this takes into consideration the fact that you would probably not iterate as many time doing this manually.

Final Thought

AI is a powerful lever. What matters is how we apply the force.

I believe we can build with speed and integrity. We can love the planet and still prototype fast. But we have to keep asking ourselves and each other better questions ad stay accountable to our vales.