Your Kettle Is More Dangerous Than ChatGPT

Let me say something that might surprise you, coming from someone who lives and breathes AI for a living.

The AI energy panic is mostly misdirected.

Not fake. Not irrelevant. Just misdirected.

And if you are making decisions about your carbon footprint based on guilt about your ChatGPT usage, you are optimizing the wrong variable. Completely.

I have spent years delivering AI and automation outcomes for enterprises across four continents. I have sat in rooms where the ROI question was the only question that mattered. And increasingly, the climate question is arriving in those same rooms. That is good. We should be asking it. But we should be asking it with accurate data, not headlines engineered to make you feel bad about pressing “Send.”

So here is the version built on actual numbers.

---

The Headline Is Wrong. Here Is What the Data Says.

According to 2025 figures cited by BBC climate editor Justin Rowlatt in The Climate Question podcast, AI consumed approximately 0.5% of the world’s electricity. Half of one percent. And electricity itself is only a fraction of total global energy use, which means AI’s actual share of total energy consumption sits somewhere between 0.1% and 0.2%.

Read that again.

Zero point one to zero point two percent.

Now compare that to the media coverage. Based on headlines alone, you would think every prompt you send is melting a glacier. That is not analysis. That is noise. And it crowds out the conversations that actually need to happen.

Bloomberg Green’s Akshat Rathi was equally direct in the same discussion. The internet as a whole, factoring in everything from device manufacturing to data transmission to running the infrastructure, accounts for roughly 3 to 5% of global CO2 emissions. AI sits inside that number. It is not the dominant story.

The dominant story, if you actually want to reduce your carbon footprint, remains the same as it has been for years. How do you travel? How you heat and cool your home. What you eat. Those three levers still move the needle more than any number of AI queries you could run in a lifetime.

---

Nobody Tells You This Comparison. They Should.

Boiling a kettle produces more carbon than running an AI search.

This is not a metaphor. It is a direct comparison made by Rowlatt during the podcast discussion. Anything involving heating, physically transforming matter through thermal energy, requires orders of magnitude more power than processing digital information. A kettle. A washing machine. A dishwasher. These dominate your personal energy consumption in a way that your AI usage simply does not.

Now, is an AI query more energy-intensive than a traditional Google search? Yes. The rough number cited in the podcast is about ten times more. That is a real difference, and it should drive efficiency investment on the part of AI companies. But ten times a tiny number is still a small number. If a Google search is negligible in your carbon budget, then an AI query at 10x is still negligible.

The multiplier that does change the picture is agentic AI. When you ask an AI to go out, take actions, commission additional searches, book something on your behalf, or complete multi-step tasks autonomously, the energy footprint can scale dramatically. Rowlatt put a figure on it: an agentic task like booking a flight could cost the equivalent of around a thousand standard Google searches. That is a different order of magnitude, and it is a legitimate area to watch as AI agents move from demos into production.

But that is a system design and infrastructure problem. It is not a “stop using AI” problem.

---

Stop Counting Your Prompts. Start Asking Where the Data Centre Is.

The meaningful concern around AI energy is not your individual query count. It is where data centres are being built, how they are being powered, and what resources they are consuming.

In the United States, data centre build-out represents a disproportionately large share of new electricity demand, partly because US electricity consumption growth from electric vehicles and industrial electrification is slower than in other markets, meaning data centres loom larger in the mix. The major AI companies are headquartered there, and they are scaling infrastructure at a pace that local grids were not designed for.

More importantly: water. Rathi flagged this explicitly, and it rarely makes the headline. Data centres require enormous volumes of water for cooling. They are frequently cited in regions already experiencing water stress. This is not a carbon story. It is a resource depletion story, and it is arguably more urgent in the near term for the communities affected.

If you want to apply real pressure to AI companies on environmental grounds, the water footprint and site selection questions are far more productive than asking whether your team should send fewer prompts.

---

The Efficiency Gains Are Real. So Is the Incentive Behind Them.

There is a dynamic in AI infrastructure that the doom narrative systematically ignores.

AI companies are aggressively optimizing energy consumption for purely commercial reasons. Most users are not paying per query. That means every watt of compute is a cost the company absorbs directly. The financial incentive to drive efficiency is enormous, independent of any climate commitment.

We have already seen this play out. Model efficiency has improved dramatically in a short period. Techniques like caching familiar query patterns, routing simple requests to smaller models, and reducing unnecessary compute are all in active deployment. Rathi made the observation clearly: these efficiency gains will continue because the economics demand it.

This does not mean the total energy footprint of AI will shrink. Usage will almost certainly grow faster than per-query efficiency improves. The honest answer is that total AI energy consumption will rise. But the carbon intensity of that energy is simultaneously falling as grids integrate more renewables. These two curves are in tension, and neither can be read in isolation.

The IEA data referenced in the podcast illustrates the speed of this shift. Renewables deployment triggered partly by the 2022 energy shock has already meaningfully cushioned European energy prices within four years. The transition is happening, at scale, faster than most people’s mental models have updated.

---

AI Is Also a Climate Tool. The Serious Analysts Know This.

This is the part that tends to disappear under the alarm.

AI is not only consuming energy. It is being deployed to use energy more intelligently.

The examples from the podcast are operational, not theoretical. AI-powered sorting systems at recycling facilities are improving material separation accuracy. Weather forecasting models precise enough to allow grid operators to anticipate renewable generation windows and reduce reliance on fossil fuel backup in real time. Optimization loops across logistics, agriculture, and manufacturing that compress waste out of systems at a scale no human team could replicate.

Rowlatt acknowledged what most honest observers acknowledge: no comprehensive peer-reviewed analysis has yet quantified the net climate impact of AI at the system level. The technology is too new and the applications too diverse for that accounting to be clean. His read, which I share, is that right now the energy draw likely exceeds the climate benefit, simply because the majority of current AI usage is consumer-facing productivity, not systemic decarbonization.

That balance shifts as the technology matures and as the applications that actually move the needle on emissions get built out, funded, and deployed at scale.

I am tracking several of those. They are not science fiction. They are in production.

---

Five Questions Every Enterprise Leader Should Be Asking Their AI Vendors Right Now

Stop letting the energy narrative function as a reason for delay. The data does not support using it to avoid investing in AI. It supports applying the right scrutiny to the right decisions.

These are the questions that separate serious buyers from organizations that are either ignoring the issue entirely or hiding behind it:

1. Where are your data centres located, and what is the grid mix at each site? Renewable sourcing varies enormously by geography. A facility in Norway and one in the American Midwest are not the same environmental proposition.

2. What is your water consumption per query, and how is it trending? If a vendor cannot answer this, that is itself the answer.

3. How has your per-query energy cost changed in the last twelve months? Efficiency improvement should be measurable, and a serious vendor should want to tell you about it.

4. What is your approach to agentic workloads? As multi-step AI tasks become standard, energy and cost profiles change materially. Understand this before you run it at scale.

5. Do you have independently verified emissions data, or only self-reported figures? The gap between those two is often significant.

At IAC.AI, outcome-based delivery means accountability for results, not just activity. That same discipline applies to the infrastructure decisions we recommend. Environmental impact is a real variable in the equation. It is just not the dominant one right now, and treating it as such without the data to back it up is not rigorous. It is a theatre.

---

The Right Level of Concern Is Harder to Hold Than the Headlines Suggest

The podcast surfaced a concept from climate psychology called the Goldilocks Zone: the optimal level of concern where there is enough anxiety to motivate action, not so much that it produces paralysis, and not so little that you simply disengage.

Most AI and climate coverage sits outside that zone. It either catastrophizes individual usage in a way that generates guilt without producing any useful action, or it dismisses the topic entirely as tech-company PR.

The honest position sits between those. And it requires actual numbers rather than a narrative.

Here is what the numbers say.

AI’s current energy footprint is real, measurable, and smaller than the coverage suggests. The companies building this infrastructure have both the financial incentive and the technical capability to improve it. The grids powering them are getting cleaner. The applications that could make AI a genuine net climate asset are being built, slowly and imperfectly, but built.

Your kettle, your car, and your diet are still doing more damage than your entire prompt history.

Focus your scrutiny where the leverage actually lives.

---

This article draws on the BBC World Service podcast “The Climate Question,” featuring host Graihagh Jackson and panellists Akshat Rathi (Senior Climate Reporter, Bloomberg Green), Caroline Steel (host, Crowdscience, BBC World Service), and Justin Rowlatt (BBC Climate Editor). All data points referenced originate from that episode and have not been modified or re-attributed beyond what was stated on air.