Guides
Is AI Bad for the Environment? The Honest Numbers (2026)
AI's environmental impact explained: energy use, water, data centers, and what the industry is doing — with honest numbers, no hype.
Short answer: AI does use a lot of energy — but "bad for the environment" needs context. Data centers running AI already consume around 1–2% of the world's electricity, and that share is growing fast. At the same time, AI is also being used to fight climate change, from optimizing power grids to designing better batteries. This guide lays out the honest numbers, what's driving the growth, and what the industry is doing about it.
The environmental debate around AI tends to split into two camps that both oversimplify: "AI will boil the planet" and "it's just computers, relax." Neither is right. Here's what's actually going on.
Where AI's environmental footprint comes from
AI runs in data centers — giant buildings full of specialized computer chips. Two things consume resources:
- Electricity to run the chips. Training a large AI model takes enormous computing power, and every query you send (every ChatGPT question, every AI image) uses energy too. A single AI query can use several times the electricity of a traditional web search — estimates range from barely above a search to 10x or more depending on the model.
- Water to cool the chips. Those chips run hot. Many data centers use water-based cooling, and a large facility can consume millions of liters of water per day — a real concern in drought-prone areas.
There are two phases with very different footprints. Training a model is a one-time (well, periodic) massive energy spend — like building a factory. Inference — answering your queries — is the ongoing cost, and as billions of people use AI daily, inference is becoming the bigger share of total energy use.
How big is the footprint, really?
Some perspective helps:
- Data centers of all kinds use roughly 1–2% of global electricity today — comparable to a medium-sized country's consumption. AI is the fastest-growing slice of that.
- For comparison, aviation accounts for about 2–3% of global CO2 emissions. AI's footprint is currently smaller than flying — but growing much faster.
- Your personal AI use is tiny in absolute terms. The concern isn't your chat history; it's the aggregate of billions of users plus the race to build ever-larger models.
The uncomfortable truth: nobody knows exactly how big AI's footprint will get, because it depends on choices being made right now — how fast data centers get built, what powers them, and how efficient the chips become.
The data center building boom
Tech companies are building data centers at a historic pace to keep up with AI demand — multi-billion-dollar facilities with names like Colossus and Stargate. This is where the environmental debate gets concrete: each new mega-facility needs enormous amounts of electricity and water, and communities near them are asking hard questions about who benefits and who pays.
See our AI data centers FAQ for the full breakdown of what these facilities are, why AI needs them, and the most common questions people ask.
The grid problem nobody planned for
Here's the part of the story that gets less attention than carbon footprints: electricity grids weren't built for this. A single large AI data center can draw as much power as a small city — and dozens are being planned simultaneously. Utilities in several regions are warning that data center demand could outpace new power generation, raising two uncomfortable questions:
- Who pays for grid upgrades? If utilities build new power plants and transmission lines mainly to serve data centers, regulators have to decide whether those costs land on tech companies or get spread across all ratepayers — including households.
- Will it raise your electricity bill? In some regions, the answer is already trending toward yes. This is becoming a live political issue, not just an environmental one.
This is also why the nuclear deals matter beyond carbon: tech companies are trying to bring their own power supply rather than draining the public grid. Whether regulators and communities accept that bargain will shape where the next wave of data centers gets built.
What the industry is doing about it
To their credit, the big AI companies know this is a problem — for the planet and for their own growth, since they can't expand without power. What's happening:
- Nuclear deals. Microsoft, Amazon, and Google have all signed agreements to buy nuclear power for data centers — including restarting mothballed plants. Nuclear provides the 24/7 carbon-free power that solar and wind alone can't guarantee for facilities that never sleep.
- More efficient chips. Each generation of AI chip does more work per watt. Efficiency is improving fast — though so far, demand is growing even faster (the classic rebound effect).
- Better cooling. Newer facilities use closed-loop and liquid cooling that drastically cut water consumption compared to older designs.
- Clean energy purchases. The hyperscalers are among the world's largest buyers of renewable energy, even if it doesn't yet cover their total growth.
Honest assessment: these efforts are real but racing against explosive demand. Efficiency gains are being outpaced by growth — for now.
AI's other side: helping the environment
It's worth noting the ledger has two sides. AI is actively being used to:
- Optimize electricity grids and reduce waste in energy systems
- Design better batteries, solar materials, and carbon-capture approaches
- Monitor deforestation, track wildlife, and model climate scenarios
- Cut emissions in logistics, manufacturing, and building management
Whether AI ends up net-positive or net-negative for the climate is genuinely undecided — it depends on how fast the footprint grows versus how much it helps decarbonize everything else.
What you can actually do
Individual guilt over AI queries isn't productive, but informed choices are:
- Use the right tool for the job. Don't use a giant AI model for something a web search can answer — a search typically uses far less energy.
- Support transparency. Favor companies that publish their energy and water usage. Public pressure works.
- Pay attention locally. If a data center is proposed near you, engage with the planning process — water rights and grid impacts are legitimate community concerns.
- Keep perspective. Your flight, your car, and your home heating dwarf your AI usage. Focus your climate energy where it counts most.
Frequently asked questions
How much energy does one ChatGPT query use?
Estimates range from barely above a Google search to 10x or more depending on the model — still a tiny amount in absolute terms. The issue is scale: billions of queries add up.
Is AI worse for the environment than crypto?
Currently, no — cryptocurrency mining (especially Bitcoin) uses far more energy than AI. But AI's energy demand is growing faster, and could eventually rival or exceed it if growth continues unchecked.
Do data centers drain local water supplies?
They can, depending on design and location. Older evaporative-cooling designs in dry regions are the worst case. Newer closed-loop systems use far less. This is one of the most legitimate local concerns about new facilities.
Will AI's energy use keep growing forever?
Probably not forever — efficiency improvements, better chips, and eventually market saturation will bend the curve. But most projections show strong growth through the rest of this decade at least.
Should I feel guilty about using AI?
No. Individual AI use is a rounding error next to heating, transport, and diet. The footprint question is systemic — it's about how data centers get powered, not about your chat history. Use AI thoughtfully, but save your climate guilt for the big stuff.
The bottom line
AI's environmental footprint is real, growing, and worth taking seriously — but it's not the apocalypse some headlines suggest, either. The next few years of choices about what powers data centers will matter enormously. Stay informed, use AI thoughtfully, and direct your climate concern where the numbers say it counts.