We finally know officially how much energy and water a ChatGPT query uses

# The Real Environmental Cost of AI: Official ChatGPT Usage Numbers vs. Daily Life

For over two years, alarming headlines have dominated discussions about AI’s environmental impact. Stories claimed that ChatGPT queries consume massive amounts of electricity and water, with some estimates suggesting each query used as much as three water bottles. But new official data from OpenAI CEO Sam Altman reveals the truth: these fears were dramatically overblown.

## The Official Numbers Are In

In June 2025, Sam Altman published [the first official figures for ChatGPT’s resource consumption per query](https://i.imgur.com/ZiUk4dT.png):

– **Electricity:** 0.34 watt-hours per query
– **Water:** 0.000085 gallons (about 0.32 milliliters) per query

To put this in perspective, Altman notes that the electricity usage is “about what an oven would use in a little over one second, or a high-efficiency lightbulb would use in a couple of minutes.” The water usage is “roughly one fifteenth of a teaspoon.”

## How Wrong Were the Previous Estimates?

The contrast with widely-circulated estimates is striking:

**Electricity:** Many studies claimed 2.9 watt-hours per query—about 8.5 times higher than the actual figure. The most recent academic research had gotten close, with Epoch AI estimating 0.3 watt-hours, but even reputable sources were still citing the inflated numbers.

**Water:** This is where the overestimation was most dramatic. Popular claims ranged from 500 milliliters per 5-50 queries to “one water bottle per query.” The reality? At 0.32ml per query, you’d need over 1,500 ChatGPT queries to equal one standard water bottle.

## Putting AI in Context: Your Daily Digital Life

Let’s compare ChatGPT usage to everyday activities:

### Electricity Consumption

**ChatGPT query: 0.34 watt-hours**

For comparison:

– **Laptop usage:** 30-70 watts = 30-70 watt-hours per hour
– **Smartphone:** 2-6 watts = 2-6 watt-hours per hour
– **LED lightbulb:** 8-12 watts = 8-12 watt-hours per hour
– **Desktop computer:** 200-500 watts = 200-500 watt-hours per hour

**Reality check:** One ChatGPT query uses about the same electricity as:

– 30 seconds of laptop use
– 3-10 minutes of smartphone use
– 2-3 minutes of LED light operation
– 4-7 seconds of desktop computer use

Even if you asked ChatGPT 100 questions per day (an extremely heavy usage pattern), you’d consume 34 watt-hours—less electricity than running a single LED bulb for three hours.

### Water Consumption

**ChatGPT query: 0.32 milliliters**

For comparison:

– **Hamburger:** ~2,500 liters (2.5 million milliliters)
– **Cup of coffee:** ~140 liters (140,000 milliliters)
– **Slice of bread:** ~40 liters (40,000 milliliters)
– **Glass of beer:** ~75 liters (75,000 milliliters)
– **Single almond:** ~4 liters (4,000 milliliters)

**Reality check:** One ChatGPT query uses the same amount of water as:

– 1/7.8 millionth of a hamburger
– 1/437,500th of a cup of coffee
– 1/12,500th of a single almond

To match the water footprint of eating one hamburger, you’d need to make approximately 7.8 million ChatGPT queries.

## Why Were the Estimates So Wrong?

Several factors contributed to the massive overestimation:

1. **Early model inefficiency:** Initial studies were based on older, less efficient AI models
2. **Conservative assumptions:** Researchers made worst-case assumptions about cooling and infrastructure
3. **Methodological issues:** Some studies included training costs or broader data center operations beyond just query processing
4. **Geographic variations:** Water usage varies dramatically by data center location, and some studies used high-consumption regions as baselines
5. **Incomplete data:** Without official numbers, researchers had to make educated guesses that erred on the side of caution

## The Bigger Picture

This doesn’t mean AI has zero environmental impact. At global scale, with billions of queries daily, the aggregate consumption is substantial. ChatGPT processes over 1 billion queries per day, which translates to:

– **Daily electricity:** ~340 million watt-hours (340 MWh)
– **Daily water:** ~85,000 gallons

But context matters enormously. For individual users, even heavy AI usage represents a tiny fraction of their environmental footprint—far smaller than dietary choices, transportation, home heating, or even other digital activities.

## What This Means for AI Policy and Personal Choices

The revelation that AI’s per-query environmental impact has been dramatically overstated has important implications:

**For individuals:** AI guilt is largely misplaced. Using ChatGPT extensively has less environmental impact than drinking an extra cup of coffee or leaving a light on for a few extra hours.

**For policymakers:** Regulations should focus on actual environmental impacts rather than inflated estimates. The data suggests AI’s resource usage, while significant at scale, is manageable within existing infrastructure.

**For researchers:** This highlights the importance of transparency from AI companies and the danger of making policy based on worst-case estimates rather than actual data.

## The Path Forward

As Altman notes, “the cost of intelligence should eventually converge to near the cost of electricity” as data center production becomes more automated. This suggests that efficiency improvements will continue, potentially making AI even more environmentally sustainable over time.

The lesson here isn’t that environmental concerns about technology are invalid—they’re crucial for responsible development. Rather, it’s that accuracy matters. Overblown fears about AI’s environmental impact may have deterred beneficial uses of the technology while distracting from larger environmental issues.

Now that we have official data, we can have informed discussions about AI’s true environmental trade-offs rather than debates based on inflated estimates. The numbers show that for individual users, the environmental cost of AI assistance is remarkably small—smaller than many routine daily activities we don’t think twice about.

—

*This article is based on official usage data released by OpenAI CEO Sam Altman in June 2025, along with comparative data on everyday activities from various environmental studies and energy consumption databases.*

I think the real waste is the iCloud that stores every snap shot on every phone everywhere regardless of if anybody wants to keep each and every one.

I don’t use iCloud for that reason. I waste 50 or 60 shots for every 1 I want, and i transfer that to my laptop where I have google desktop drive to save only what I choose to keep