Can I Use ChatGPT if I Care About the Climate?

Note: While this blog post focuses on OpenAI & ChatGPT, we’re fortunate to be able to work with the most cutting edge AI companies like Microsoft and Google to help manage their emissions as they scale. We want to help more rapidly scaling AI companies manage their footprint & build a removals portfolio from the start. Whether you’re in hypergrowth or an early stage AI company, reach out to accelerate your climate program! 

Sam Altman recently published a post The Gentle Singularity that shared a critical datapoint for our climate. In the words of Sam, each ChatGPT prompt consumes “about 0.34 watt-hours of electricity and a fifteenth-of-a-teaspoon of water (0.000085 gal)”.

Image Caption: from The Gentle Singularity

That small disclosure gives us the numbers we need to do the back-of-the-envelope carbon math and calculate total emissions for chatGPT. Is AI cooking our planet? Should we avoid using ChatGPT because of the emissions? 

Our analysis—consistent with prior analysis—shows that the carbon footprint per user is minimal when viewed at the scale of an individual or household. The environmental impact is significant when you zoom out to the billions of prompts and countless GPU-hours that power today's AI models. 

AI is quickly becoming the engine behind research breakthroughs, productivity gains, and—at times—the fight against climate change itself. But this transformative tech comes with a serious global emissions price tag, and the AI industry has the responsibility to mitigate and manage their emissions from the beginning. 

It turns out that, in addition to an aggressive renewables strategy, OpenAI and similar AI labs could remove the inference emissions at a very low cost per user. In the case of ChatGPT, total inference emissions could be removed for $0.25 cents per user per year!

Let’s put some hard numbers to help us understand AI emissions and what to do about them. 

Zooming in on your Prompts

Current internet traffic estimates put ChatGPT at ~122 million daily users and roughly one billion messages a day. That pencils out to ~8 messages per user per day. Multiply the two figures and you get a platform draw of roughly 340 MWh of electricity and 85,000 gallons of water every 24 hours.

What does that mean for carbon emissions? Using the International Energy Agency’s 2024 global grid average of 445 g CO₂ per kWh, each ChatGPT reply lands at ~0.15 g CO₂. 

For a typical user with average usage patterns that scales to ~1.2 g per day, or 0.45 kg CO₂ per year—roughly the emissions from driving two miles or drinking a single latte. That’s not much! 

Our takeaway is that your own usage of ChatGPT — if following average patterns — is unlikely to be a huge driver of your overall carbon footprint.

A note on methodology: this analysis focuses on AI inference emissions. There is a whole other set of embodied emissions and data center buildout emissions that we don’t account for in this analysis. The data center build outs are rapidly evolving (Meta’s announcement this week, OpenAI Stargate) with less available data so this post focuses specifically on inference. 

Zooming Out on Global Emissions

Aggregate usage is another story, and where the climate challenge starts to arise. Based on the math Altman shared, we estimate that ChatGPT emits 55,000 tCO₂ a year in inference emissions, primarily driven through the use of 124 GWh of electricity. Electricity consumption accounts for 99% of this inference footprint. 

While AI companies are aggressively pursuing renewable energy strategies — including 24/7 clean energy matching and investments in carbon-free power — that doesn’t fully offset the emissions generated by inference nor the emissions generated by the building of the data centers themselves.

Running large language models like ChatGPT still requires vast amounts of electricity, much of which comes from carbon-intensive sources, especially during peak usage or in regions where clean energy is scarce.

So what should AI companies do?

1. Site on clean power from day one. Wind in Texas, hydro in Québec, nuclear in Sweden—take your pick, but grid fuel mix matters.

2. Thoughtful carbon accounting. Working with our carbon accounting services like to measure your footprint and develop strategies to decarbonize from the start. 

3. Be Transparent. We applaud OpenAI for sharing these metrics and believe standard reporting and transparency will help consumers and policymakers make more informed decisions. There’s a lot of FUD out there! 

4. Keep squeezing efficiency. Mixture-of-experts routing and next-gen GPUs have already cut per-token energy requirements by an order of magnitude since GPT-3; keep going.

5. Remove what’s left. Even the best-sited data center will have residual tonnes. That’s where permanent carbon removal comes in.

What would it cost a service like ChatGPT to remove all their annual emissions with Charm? Using bio-oil sequestration, we estimate the per-user cost would be…drumroll…just $0.25 cents per user per year. 

That’s it! Well within reach for the high-margin, rapidly growing businesses that OpenAI, Anthropic and leading labs are building. 

Do we use AI at Charm?

So, what does this carbon accounting mean for Charm’s use of AI? Do we use AI at Charm?

The answer is a resounding yes. From Agent350 for high quality verifications to accelerating siting and permitting decisions with Geoforge to lab co-pilots accelerating new research breakthroughs, we believe AI will help us scale our carbon removal operations faster than we dreamed possible. 

But as with any great technological breakthrough, how AI gets deployed can have tremendous impact on our planet and those effects need to be closely measured and managed. 

So yes, we believe that AI can be accelerative for our work and we will continue to be thoughtful in the ways in which we adopt and deploy AI technologies given the emissions profile. But the big lever won’t come from individual abstinence; it’s industry leadership on clean energy and a commitment to real, permanent removals — that’s willing to spend $0.25c per user for a cleaner climate. 

That’s a future we’re ready to build together!