The Dirty Secret Behind Your Favourite AI Tools

The environmental cost of artificial intelligence is rising fast — yet the industry remains largely silent. Here’s why that needs to change.

TL;DR — What You Need To Know

• AI systems like ChatGPT and Google Gemini require immense electricity and water for training and daily use
• There’s no universal standard or regulation requiring AI companies to report their energy use
• Estimates suggest AI-related electricity use could exceed 326 terawatt-hours per year by 2028
• Lack of transparency hides the true cost of AI and hinders efforts to build sustainable infrastructure
• Organisations like the Green Software Foundation are working to make AI’s carbon footprint more measurable

AI Is Booming — So Are AI’s Environmental Impact

AI might be the hottest acronym of the decade, but one of its most inconvenient truths remains largely hidden from view: the vast, unspoken energy toll of its everyday use. The focus keyphrase here is clear: AI’s environmental impact.

With more than 400 million weekly users, OpenAI’s ChatGPT ranks among the five most visited websites globally. And it’s just the tip of the digital iceberg. Generative AI is now baked into apps, search engines, work tools, and even dating platforms. It’s ubiquitous — and ravenous.

Yet for all the attention lavished on deepfakes, hallucinations and the jobs AI might replace, its environmental footprint receives barely a whisper.

Why AI’s Energy Use is Such a Mystery

Training a large language model is a famously resource-intensive endeavour. But what’s less known is that every single prompt you feed into a chatbot also eats up energy — often equivalent to seconds or minutes of household appliance use.

The problem is we still don’t really know how much energy AI systems consume. There are no legal requirements for companies to disclose model-specific carbon emissions and no global framework for doing so. It’s the wild west, digitally speaking.

Why? Three reasons:

• Commercial secrecy: Disclosing energy metrics could expose architectural efficiencies and other competitive insights
• Technical complexity: Models operate across dispersed infrastructure, making attribution a challenge
• Narrative management: Big Tech prefers to market AI as a net-positive force, not a planetary liability

The result is a conspicuous silence — one that researchers, journalists and environmentalists are now struggling to fill.

The stats we do have are eye-watering

MIT Technology Review recently offered a sobering benchmark: a 5-second AI-generated video might burn the same energy as an hour-long microwave session.

Even a text-based chatbot query could cost up to 6,700 joules. Scale that by billions of queries per day and you’re looking at a formidable energy footprint. Add visuals or interactivity and the costs balloon.

The broader data centre landscape is equally stark. In 2024, U.S. data centres were estimated to use around 200 terawatt-hours of electricity — roughly the same as Thailand’s annual consumption. By 2028, AI alone could push this to 326 terawatt-hours.

That’s equivalent to:

• Powering 22% of American homes
• Driving over 300 billion miles
• Completing 1,600 round trips to the sun (in carbon terms)

Water usage, often overlooked, is another major concern. AI infrastructure guzzles water for cooling, posing risks during heatwaves and water shortages. As AI adoption grows, so too does this hidden drain on natural resources.

What’s being done — and who’s trying to fix it

A handful of organisations are beginning to push for accountability.

The Green Software Foundation — backed by Microsoft, Google, Siemens, and others — is creating sustainability standards tailored for AI. Through its Green AI Committee, it champions:

• Lifecycle carbon accounting
• Open-source tools for energy tracking
• Real-time carbon intensity metrics

Meanwhile, governments are cautiously stepping in. The EU AI Act encourages sustainability via risk assessments. In the UK, the AI Opportunities Action Plan and British Standards Institution are working on guidance for measuring AI’s carbon toll.

Still, these are fledgling efforts in an industry sprinting ahead. Without enforceable mandates, they risk becoming toothless.

Why transparency matters more than ever for AI carbon emissions

We can’t manage what we don’t measure. And in AI, the stakes are immense.

Without accurate data, regulators can’t design smart policies. Infrastructure planners can’t future-proof grids. Consumers and businesses can’t make ethical choices.

Most of all, AI firms can’t credibly claim to build a better world while masking the true environmental cost of their platforms. Sustainability isn’t a PR sidecar — it must be built into the business model.

So yes, generative AI may be dazzling. But if it’s to earn its place in a sustainable digital future, the first step is brutally simple: tell us how much it costs to run.

You May Also Like:

• AI Powering Data Centres and Draining Energy
• AI Increases Google’s Carbon Footprint by Nearly 50%
• The Thirst of AI: A Looming Water Crisis in Asia
• You can read more from the IEA by tapping here.