Can We Use AI and Still Love the Planet?

A grounded, hopeful look at the numbers, updated for 2026

If you’ve been feeling eco-guilt every time you open an AI tool… I understand. You care deeply about your impact on the planet, and that’s a beautiful thing.

(Little-known fact: I have a master’s degree in green/sustainable business. So believe me, I get the hesitation! 😅)

A while back I started digging into the data on AI’s energy and water use. Not the fear-mongering headlines, but the real numbers.

And the real picture turns out to be more hopeful than the mass anxiety would suggest…

This article is lovingly written to help you, dear soulpreneur, feel more at peace about using these tools, for the purpose of growing yourself and serving others better.

I’ve combined what were originally two separate articles (one on energy, one on water) and updated everything with the most current research I could find. May you find the answers to be both reassuring and empowering.

---

The short version (for skimmers)

• One AI prompt (one message you send and the response AI gives back) uses about as much electricity as nine seconds of TV, and about five drops of water.

• Your active use of AI is likely a rounding error next to your video streaming, your meals, and your travel.

• The whole industry’s footprint is real and growing, yet it’s getting greener fast, and most importantly, it was never your load to carry.

• The kindest thing you can do for the planet, yourself, and the people you serve, is to use AI with purpose, and aim your societal concerns at policy rather than foster personal guilt.

If that’s all you needed, breathe easy, and go create! If you’d like the receipts, read on. 🙏

Before we start: where these numbers come from

I want this to be trustworthy, not only comforting. So, a few ground rules.

Some of the most detailed studies come from the AI companies themselves. Google in particular published a comprehensive technical report in 2025. Company data deserves a raised eyebrow… so wherever possible, I’ve checked it against independent sources, and thankfully, they line up closely. I’ll also be clear about what the numbers do and don’t include.

I’m not claiming AI has zero footprint, or that every concern is silly. There are real, societal conversations to be had about the build-out of data centers, and water stress in particular communities. What I want to show you is how your individual use compares to the rest of your life. That’s where the guilt lives, and without that context, it’s how the headlines get it wrong.

Let’s take a look.

What one AI prompt (and the response) costs

In 2025, Google measured the energy, water, and carbon of a typical text prompt to its Gemini AI. A single prompt used:

• 0.24 watt-hours of electricity, about the same as watching TV for nine seconds.

• 0.26 milliliters of water, roughly five drops.

• 0.03 grams of CO₂, a tiny fraction of almost anything else you’ll do today.

Five drops of water. Not even a cup. Five drops.

You’ve probably seen the scary version: “every AI query drinks a whole bottle of water!” That figure (and yes, the half-liter estimate I used in my own earlier article) turned out to be off by a wide margin. For one thing, it was measuring a whole multi-prompt session, not a single prompt. And to give the skeptics their due: you should count the full lifecycle, including the water used back at the power plant that makes the electricity. Even when you do, one typical prompt comes to only a few milliliters, nowhere near a bottle. Google’s narrowest figure, just the data-center cooling, is 0.26 mL; yet if you add the upstream power and chip-making, a prompt still lands in single-digit milliliters of water.

And this isn’t only Google patting itself on the back. Independent researchers at Epoch AI put a typical ChatGPT query at about 0.3 watt-hours, and OpenAI disclosed a similar figure (0.34 watt-hours, about a fifteenth of a teaspoon of water). Three independent teams, measuring slightly different things, all land in the same small range. That’s about as solid as a number like this gets.

(One quick footnote: these are medians for everyday text prompts. A long “reasoning” request, an AI-generated image, or a few seconds of AI video can cost far more. A short AI video can use as much energy as running your microwave for an hour. So “a few drops” is the right number for the kind of text-based help most of us use AI for.)

To be fair, you need to compare AI to everything else…

We rarely consider the energy of our daily habits, yet we fixate on AI because it’s in the news.

If one prompt is nine seconds of TV, then an hour in front of the TV uses about as much electricity as 400 AI questions. So… which is more likely to improve your life and your business: an evening of autoplay, or 400 thoughtful exchanges about your work, your clients, your growth? :)

On water, the comparison is just as lopsided:

• A typical American household runs through more than 1,000 liters of water a day (showers, laundry, dishes, the lawn). An hour of back-and-forth with AI might cost a few teaspoons.

• A single beef hamburger takes roughly 1,800 liters of water to produce, counting everything from the feed to the bun. To be fair, count everything on the AI side too (the power plants and chip-making, not just the data center): a year of daily AI chatting still uses only a small fraction of that one burger. (And yes, a plant-based burger is far kinder to the water table, very roughly 85% less than beef by some lifecycle estimates. Even so, it uses several times more water than a year of typical AI prompts.)

So while AI does have a footprint, for your daily use it’s tiny next to things you might already do, without a second thought.

If you’re comfortable streaming a show, eating lunch, or driving to a meeting, you can relax about asking AI to help you draft an email, plan a launch, or think something through.

The real question isn’t “Is using electricity sinful?” It’s “Is this a worthwhile use of it?” Spending a few drops of water and a few seconds of TV’s worth of power to become more capable, more thoughtful, and more helpful to others? That’s one of the best uses I can imagine.

A single flight puts it all in perspective

One more comparison keeps it all in proportion…

A transcontinental flight emits roughly a tonne, a thousand kilograms, of CO₂ per passenger. At 0.03 grams per prompt, that single flight equals the carbon of millions of AI prompts.

Flip it around. A heavy user running 100 prompts every single day for a year racks up barely a kilogram of CO₂ from it, somewhere around a thousandth of that one flight. So if you’re at peace with the occasional vacation, the carbon of your AI habit can come right off your worry list.

Efficiency is improving at a breathtaking pace

In Google’s measurements, the electricity used for a typical prompt dropped 33-fold in a single year (May 2024 to May 2025). The carbon per prompt dropped 44-fold, even as the answers got better.

Read that again. Not 33%. 33 times less energy, in a single year. (Put another way: a prompt now uses about 3% of the energy it did twelve months before.)

That’s the latest year-over-year figure Google has published, through May 2025. They haven’t released the following year’s number yet, but nothing suggests the rapid improvement has stopped.

Two things drive this. First, the models themselves are getting dramatically more efficient. Second, the industry keeps shifting to specialized chips. For the kind of math AI runs, a purpose-built AI chip is often 10 to 30 times more energy-efficient than the general-purpose processor it replaces, like trading an all-around athlete for an Olympic sprinter who does one thing astonishingly well.

Train once, use forever

“But isn’t training an AI model hugely energy-intensive?” Yes, but it’s a one-time, fixed cost.

It’s like filming a movie. The production is expensive and happens once; afterward, millions of people enjoy it for years. Once a model is trained, that same finished “library of intelligence” serves everyone who uses it.

As of early 2026, ChatGPT alone has around 900 million weekly users, roughly one in nine people on Earth. Spread the one-time training cost across that many people and that many interactions, and the share for any one of us becomes vanishingly small.

(Yes, models do get retrained and refreshed; it’s not literally one time forever. But for a model used by almost a billion people, that big upfront cost still divides down to an extremely small amount per person.)

The companies have real skin in this game

AI isn’t the first technology to raise energy questions. Cloud computing sparked nearly identical worries in the early 2000s, and efficiency innovations soon made those worries moot.

The same forces are at work now, only stronger. AI companies spend billions on electricity and water. Every wasted watt is money out of their own pocket. They have enormous financial incentive, sharpened by fierce competition, to make every query cheaper and cleaner. The engineers building these systems are working on this far harder than any of us could, and not purely out of altruism: their survival depends on it.

This cuts both ways, though. Even as each query gets more efficient, the big tech companies’ total emissions have risen in recent years, because they’re building so fast. That’s a real concern. But notice where it lives. It’s a question about how a handful of giant companies build and power their data centers. It is not a question about whether you, personally, ask AI to help with rewriting your website. Your individual prompts aren’t what’s moving that number, and your individual guilt isn’t what will fix it.

So the efficiency problem, making each query leaner and cleaner, is being handled, hard, by the people whose job it literally is. The bigger question of the whole build-out’s total footprint is real too. We’ll come back to whose job that one is, and where your care can do some good.

AI is helping stabilize the grid, and pull clean energy forward

This surprised me:

AI’s electricity demand is flexible. A data center can pause heavy training work when the grid is stressed, then resume it at night when wind power is abundant and cheap. That makes AI one of the useful tools we have for balancing the grid.

In one recent demonstration, an AI data center cut its power draw by about a third within 30 seconds of a grid signal, with no loss of service. As Varun Sivaram, who runs a company doing exactly this, put it:

“We need to stop thinking of AI as inherently inflexible and start seeing it for what it is: the Holy Grail of demand-side management.”

And because AI developers want clean, cheap, abundant power, their demand is accelerating the build-out of renewables:

• Renewables already supply about 27% of data centers’ electricity, among the largest single sources, and the fastest-growing.

• That renewable supply is projected to grow about 22% a year through 2030, meeting roughly half of all the new demand AI creates.

So the race to power AI is helping fund the very clean-energy infrastructure the world needs.

On water, the three companies whose clouds run almost all AI (Google, Microsoft, and Amazon) have each committed to becoming “water-positive” by 2030, replenishing more freshwater than they consume. And the progress is measurable: Amazon reports it’s about 75% of the way there and now several times more water-efficient than the industry average, and Microsoft has designed a new generation of data centers that use essentially zero water for cooling, with the first sites coming online in 2026. (These are commitments and early pilots, not finished achievements. They also lean partly on replenishment projects and clean-energy contracts that independent watchdogs say are hard to fully verify, so this deserves ongoing scrutiny. The direction of travel is solid; the pace is fairly debated.)

Now zoom out: the whole sector is a rounding error

Let’s look at the big picture.

All of the world’s data centers, every website, every video, every cloud app, and AI combined, used about 1.5% of global electricity in 2024 (per the International Energy Agency, the newest data I could find). The CO₂ from all that electricity is less than 1% of global emissions. AI is only a slice of that slice.

For comparison, here is what moves the needle (Our World in Data):

• Road transport: about 12% of global emissions

• Agriculture: about 12%

• Heating, cooling, and powering our homes: about 11%

Each of those is more than ten times the entire data-center sector. So while it’s worth being thoughtful, the footprint of your AI use is a rounding error next to how everyone travels, eats, and heats their homes.

One fair caveat: that’s true today. The data-center sector is growing fast. The International Energy Agency expects it to roughly double by 2030, to around 3% of global electricity, as AI scales up. Keeping it small will take real work. But once again, that growth is driven by a global industrial build-out, not by your personal prompts. Your individual use is a rounding error of a rounding error.

A world that spends electricity freely on autoplay video, doom-scroll feeds, and ads nobody asked for, can surely spare a little for tools that help people learn faster, communicate more clearly, and create things that matter.

AI can actually be a net saver of energy and water

This is the part that turns guilt into optimism. The same intelligence that costs a few drops of water is being pointed at saving enormous amounts of energy, water, and carbon:

• Contrails. Those thin white lines behind planes cause more than a third of aviation’s total warming impact. Google’s AI predicts where they’ll form, so pilots can nudge their route slightly to avoid them. In a 2025 trial with American Airlines, the flights that followed the AI’s recommended route cut contrail formation by 62%, with no meaningful increase in fuel use. (It’s early: only a fraction of trial flights executed the plan fully, so the fleet-wide average was smaller. But the mechanism clearly works.)

• Farming. A 2025 peer-reviewed meta-analysis of about 135 studies found that AI-driven irrigation can cut agricultural water use by 30 to 50% while improving yields, and farming is the single largest user of fresh water on Earth.

• Industry. One oil-and-gas company (ADNOC) used AI across its operations to avoid about 1 million tonnes of CO₂ over 2022 to 2023, the equivalent of taking around 200,000 cars off the road. (That’s a self-report from an oil company, so hold it loosely. But it’s in line with other heavy-industry results.)

• Data centers themselves. Google’s AI has cut the energy needed to cool its data centers by up to 40%, a system it’s relied on since 2016 for many other uses besides AI.

When AI spends a little electricity to optimize a flight path, a farm’s watering, or a factory’s energy use, the net impact isn’t small. It’s wildly positive, like using a teaspoon of water to save a swimming pool.

The concerns that are worth taking seriously

Let me lay out the concerns a thoughtful skeptic would raise. They’re fair, and when you look closely, none of them lands on your own AI usage.

“Won’t all this efficiency just make us use way more?” Quite possibly. It’s a real pattern (economists call it the rebound effect): when something gets cheaper, we use more of it. Total AI demand may well keep climbing even as each query gets leaner. That’s a well-argued concern, and it’s an argument for clean power and smart policy keeping pace with the build-out, not an argument for you to feel guilty about your prompts.

“What about the towns where data centers strain the local water?” This one is real and specific. A single large data center can use as much water as a small town, and some have been built in already-stressed, drought-prone areas, drawing justified local pushback and, increasingly, new regulation. But notice that this is a problem of where and how a few giant facilities are sited and cooled, a corporate and policy question. It has almost nothing to do with the few drops behind your individual chat. (In fact, most of AI’s water footprint isn’t even in the data center. It’s upstream, in generating the electricity and making the chips.)

“Aren’t those green pledges just good PR?” Partly fair. “Water-positive” and “100% renewable” claims often rest on offsets and energy contracts that are hard to verify, and as I mentioned, the companies’ total footprints have been rising. Healthy skepticism here is warranted.

All three have something in common: they’re systemic questions, about how a handful of enormous companies build, site, and power their infrastructure. They deserve real scrutiny, real journalism, and real regulation. None of that is a reason to talk yourself out of a tool that helps you grow your skills and serve people better. Weigh the big concerns where they belong, at the level of policy and corporate accountability, and take a few deep breaths about your own usage.

---

So whose fight is it, really?

Let me say this as plainly as I can: the environmental cost of AI is not yours to carry.

Not because the big questions don’t matter. They do. It’s because your part in them (unless you work on policy at the major AI companies) is imperceptibly small. You didn’t build the data centers. You don’t decide where they’re sited or how they’re cooled. As we saw above, your own use is a rounding error of a rounding error. The things that move the needle, how a handful of trillion-dollar companies build and power their infrastructure, and how governments choose to regulate them, sit entirely outside your zone of control.

So if you’ve been carrying guilt about this, you can put it down. Truly. It was never yours to hold.

And yet. I know you. You care, and you want that care to go somewhere good, so let’s see what we can do to help:

On the big-picture side, the most useful thing you can do is get behind the people working to make the whole system cleaner and more efficient. Using less AI yourself doesn’t even make the list. There’s a growing, hopeful movement, sometimes called abundance or ecomodernism, built on a simple idea that I find sane and pragmatic: we solve our environmental problems with better technology and more clean energy, not by retreating from progress. A few trustworthy places to start, drawn on purpose from across the political spectrum (this kind of optimism isn’t owned by the left or the right) –

• Hannah Ritchie, an environmental scientist at Our World in Data whose book Not the End of the World (and newsletter Sustainability by Numbers) makes the calm, data-grounded case for hope.

• Abundance, by Ezra Klein and Derek Thompson, the center-left case that we build our way to a cleaner, more plentiful world rather than shrink our way there.

• Superabundance, by Marian Tupy and Gale Pooley (with the data at HumanProgress.org), the free-market counterpart: human ingenuity makes resources more abundant over time, not less.

• The Roots of Progress Institute, a community devoted to “techno-humanism,” the belief that progress matters because it serves human flourishing. More generally, check out the Progress Movement community.

Share their work, support them, vote for saner policy. That’s where a citizen’s energy counts, far more than any prompt you skip.

And this is what I most want you to hear: regarding AI and its effects on the world — the single most beneficial thing you can do with your time, your energy, and your good intentions is to keep using AI to become better at what you’re here to do. To sharpen your skills. To serve your people more deeply. To get your message to the ones still waiting for it. Used that way, AI is a higher use of your one precious life than the worrying ever was.

An invitation to lean in

What you read shapes what you believe, and what you believe shapes what you do. If you fill your days with media that paints AI as destructive, you’ll hold back from tools that could help you and the people you serve, and that hesitation has a cost. Not in carbon, but in unrealized good: the book you don’t finish, the client you don’t reach, the product that never gets created. Your time and energy are precious too. If avoiding AI means three extra hours on something it could help you do in thirty minutes, what did that really save, and who didn’t get your work because of it?

So weigh the concerns. Just weigh them accurately, in proportion to the facts.

And hold on to the bigger pattern. Every time humanity has invented a tool that amplifies our intelligence, the printing press, the computer, the internet, energy use spiked at first… but then efficiency caught up, and ultimately, human well-being expanded. We’re watching that same story unfold with AI, only faster than anyone expected. The electricity it uses doesn’t vanish like gas burned in a traffic jam. More and more, it’s invested in solving the very problems that take energy to fix.

So, dear soulpreneur: let AI help you. The part of all this that’s truly yours, your handful of prompts, is so small it barely registers, and each one keeps getting leaner every year. The bigger system is growing, yes, and it’s also getting cleaner, smarter, and more thoughtful about doing right by the communities it touches.

Let these new tools help you move past your blocks, draft your offerings, and reach the people who need your presence and your wisdom.

Aim your care at the systems that can use it, and let your daily work be exactly that: your work, made lighter and more beneficial.

It’s not about perfection. It’s about progress, one intentional, openhearted step at a time. 🙏