Your Daily Dose of AI is Costing the Planet More Than You Think

I'll be honest - I have completely fallen in the trap of asking ChatGPT and Claude for useless things. I recently asked it (her? him?) the following:

• To create a photo of my dog Louis and me on a canal in Venice.
• To create a photo of my dog Louis as a human
• To roast a (okay, several) family photos
• How my children would get along as adults given their birth charts

…And so many more that I’m too embarrassed to even put here. Every time I ask AI to generate an image, I'm using roughly 1,000 times more energy than a simple web search. I know what you’re thinking…a web search DEFINITELY couldn’t create a special picture of you and Louis on a Venetian Canal! But put that in perspective: generating one image through AI is roughly equivalent to driving four miles in a gas car. Suddenly, my weekend spent asking my AI pals to create a picture of what Louis’ wedding decor would look like if he married our neighbor’s Maine Coon cat (hypothetically) feels a lot less innocent.

The stats are staggering. The computational power required to train generative AI models like OpenAI's GPT-4 can demand a staggering amount of electricity, which leads to increased carbon dioxide emissions and pressures on the electric grid. We're talking about training GPT-4 taking over $100 million and consuming 50 gigawatt-hours of energy - enough to power San Francisco for three days.

The Hidden Truth About Data Centers

Remember when we thought data centers were just those hypothetical warehouses in the cloud? Well, "AI data centers are much more power intensive. So, if you have a normal data center, an AI data center would be up to 10 times more power intensive", according to Dr. Noman Bashir from MIT's Climate and Sustainability Consortium.

Amazon has more than 100 data centers worldwide, each of which has about 50,000 servers, and that was before the AI boom really took off. AI data centers are massively contributing to the continued rise in power demand, which itself contributes to the continued rise in global emissions. And most importantly, it's growing faster than grid capacities.

Here's the kicker: it's growing faster than renewable energy growth, which means that even as data center energy use grows, clean energy from wind, solar, nuclear perhaps is lagging behind. Translation? We're filling the energy gap with fossil fuels.

It's Not Just About Electricity

The environmental impact goes way beyond your electric bill. A massive amount of water is needed to cool the hardware used for training, deploying, and fine-tuning generative AI models, which can strain municipal water supplies and disrupt local ecosystems. These chips that run these models get very hot, so you need water to cool them down.

Think of it like this: a study by researchers at the University of Massachusetts Amherst estimated that training a large AI model could produce over 626,000 pounds of carbon dioxide equivalent. This is more than five times a car's emissions over its entire lifetime.

And then there's the hardware itself. Market research firm TechInsights estimates that the three major producers (NVIDIA, AMD, and Intel) shipped 3.85 million GPUs to data centers in 2023, up from about 2.67 million in 2022. That number increased by an even greater percentage in 2024. Each of those GPUs has its own manufacturing carbon footprint, plus the environmental impacts of mining the raw materials.

The Reality Check: Who's Paying for AI's Energy Appetite?

Here's what really gets me: the business model of the utility industry for more than 100 years has been to invest in their systems and then spread those costs for everyone. Meaning when tech companies build massive data centers that strain the grid, utilities are building billions of dollars of infrastructure to support them and spreading those costs to all of us.

Electricity bills are climbing nationwide, rising faster than inflation in many places, and AI is a significant driver of this increase. As Harvard Law School's Ari Peskoe puts it: there are energy markets where utilities buy their power. And because demand from these A.I. energy centers is booming, it's driving up prices, and we're all paying those higher prices.

So basically, we're all subsidizing Big Tech's AI ambitions through our monthly electric bills. Cool.

The Policy Problem

Meanwhile, the Trump administration is pushing to ease environmental regulations to accelerate AI development. WIRED magazine has found that big tech companies are asking the White House to ease those protections and that the Trump administration is all in. Naturally. They're specifically targeting things like Clean Water Act permits that protect wetlands and waterways from data center construction.

Amazon, for instance, is building a data center in Indiana. They would fill in almost 10 acres of wetland and impact thousands of streams in the region. The industry's argument? In order for the U.S. to compete with China, we need to be building more data centers.

What Can We Actually Do About It?

As Individual Users:

The MIT researchers I spoke with had a refreshingly honest take: "Putting everything on the end user and making them feel guilty is not the right approach. So, I don't think the end user should feel completely responsible to do these and to manage this problem. However, there is a case to be made for judicious use of resources".

Here's what "judicious use" looks like:

• Ask yourself: Do I really need AI for this? Use it for complex tasks where it adds genuine value, not for things you could easily Google or figure out yourself
• Choose your AI tools wisely - we can ask our AI providers to offer greater transparency. For example, on Google Flights, I can see a variety of options that indicate a specific flight's carbon footprint. We should be getting similar kinds of measurements from generative AI tools
• Remember the energy hierarchy: One simple web search takes one unit of energy. Doing that same search, but using an AI model instead, takes 10 units of energy. And finally, using AI to generate an image? That will take you up to 1,000 units. The thing especially frustrating to me is that now normal Google search includes the AI overview, and it’s insanely difficult to turn off this feature, so you’re already emitting more without even trying when doing a simple search.

The Hierarchy Looks Like This:

1. Regular Google search: 0.2g CO2 (baseline)
2. Google AI search: 2-6g CO2 (10-30x more)
3. ChatGPT text generation: ~5-10g CO2 per query
4. AI image generation: 50-100g CO2 (250-500x more than regular search)

Systemic Solutions:

The real solutions need to come from the companies and policymakers. MIT professor Elsa A. Olivetti argues that this will require a comprehensive consideration of all the environmental and societal costs of generative AI, as well as a detailed assessment of the value in its perceived benefits.

Some promising developments:

• Better transparency: We need carbon footprint labels for AI queries, just like we have for flights
• Energy efficiency improvements: data centers can invest in more energy-efficient processors and server architectures, lean on virtualization to improve resource flexibility, adopt more effective cooling technologies
• Grid cooperation: data centers can collaborate more closely with electric companies to create a shared energy economy

Companies Making Progress:

Companies like Amazon, Meta, Alphabet all claim accurately to be some of the world's largest buyers of clean energy now. The problem? Those investments aren't always very local. So they might be investing in solar energy in Texas while they're consuming that energy somewhere else in a different U.S. state.

Bottom Line

The AI boom has taken off at an awkward time for the fight against climate change, because global temperatures are already rising much faster than scientists expected. We're essentially building the infrastructure for an energy-intensive technology revolution while our grids are still heavily dependent on fossil fuels.

Does this mean we should abandon AI entirely? Of course not. Couldn’t even if we wanted to! But it does mean we need to be way more thoughtful about how we use it, push for transparency from the companies providing it, and demand that policymakers consider the full environmental cost of our AI-powered future.

The next time ChatGPT offers to write your grocery list, maybe just grab a pen instead. Your future electric bill (and the planet) will thank you.