Skip to the lesson

How AI worksLesson 6 of 6

Part 1 of 3, 0 done

14 min

The electricity behind an answer

Most chatbot answers run on a computer in a data center that someone pays to power. Weigh a prompt, a picture and a training run in watt-hours, and see why the honest answer depends on how many and what kind.

Ticket 01 · The guess

One prompt, in watt-hours

A watt-hour is the energy a one-watt device uses in an hour. In 2025 Google measured the electricity behind the median text prompt to its Gemini app, counting the AI chips, the computers around them, the machines kept ready and the data center's cooling. How many watt-hours did it find?

5 Wh
0 Wh10 Wh
Words to know

Spot something wrong?

Tell us what looks off. No name or email needed.

What kind of problem?

0 / 1000

We'll also send this page, the lesson, ticket 1, so we can find it. Nothing else. Privacy

·
Sources and every number (1)

Sources

  • Measuring the environmental impact of delivering AI at Google Scale (Elsworth et al., arXiv:2508.15734, August 2025): median Gemini Apps text prompt, energy, water, and the narrower existing approach · Google researchers, on arXiv, 2025

Every number

  • 0.24 Wharxiv.org ·2025 · Google's measured energy for the median Gemini app text prompt, May 2025, counting chips, host computers, idle machines and data-center overhead.
  • 0.10 Wharxiv.org ·2025 · The same prompt counted the narrower way, leaving out host processors and memory, idle machines and data-center overhead, and using only the most efficient data centers.
  • 9 secondsarxiv.org ·2025 · Google's comparison: the median prompt uses less energy than watching TV for this long, taking a TV as about a hundred watts.
  • 33 timesarxiv.org ·2025 · How many times less energy the median text prompt used in May 2025 than in May 2024, per Google.
  • 0.26 mLarxiv.org ·2025 · Water consumed for data-center cooling per median text prompt, per Google's comprehensive count.
  • 45–50 mLarxiv.org ·2025 · Earlier public estimates of water per prompt, as Google's paper cites them.
  1. Lessons›
  2. The school›
  3. How AI works
  • ·Before this: Why AI makes things up
Skip to the lesson

How AI worksLesson 6 of 6

Part 1 of 3, 0 done

14 min

The electricity behind an answer

Most chatbot answers run on a computer in a data center that someone pays to power. Weigh a prompt, a picture and a training run in watt-hours, and see why the honest answer depends on how many and what kind.

Ticket 01 · The guess

One prompt, in watt-hours

A watt-hour is the energy a one-watt device uses in an hour. In 2025 Google measured the electricity behind the median text prompt to its Gemini app, counting the AI chips, the computers around them, the machines kept ready and the data center's cooling. How many watt-hours did it find?

5 Wh
0 Wh10 Wh
Words to know

Spot something wrong?

Tell us what looks off. No name or email needed.

What kind of problem?

0 / 1000

We'll also send this page, the lesson, ticket 1, so we can find it. Nothing else. Privacy

·
Sources and every number (1)

Sources

  • Measuring the environmental impact of delivering AI at Google Scale (Elsworth et al., arXiv:2508.15734, August 2025): median Gemini Apps text prompt, energy, water, and the narrower existing approach · Google researchers, on arXiv, 2025

Every number

  • 0.24 Wharxiv.org ·2025 · Google's measured energy for the median Gemini app text prompt, May 2025, counting chips, host computers, idle machines and data-center overhead.
  • 0.10 Wharxiv.org ·2025 · The same prompt counted the narrower way, leaving out host processors and memory, idle machines and data-center overhead, and using only the most efficient data centers.
  • 9 secondsarxiv.org ·2025 · Google's comparison: the median prompt uses less energy than watching TV for this long, taking a TV as about a hundred watts.
  • 33 timesarxiv.org ·2025 · How many times less energy the median text prompt used in May 2025 than in May 2024, per Google.
  • 0.26 mLarxiv.org ·2025 · Water consumed for data-center cooling per median text prompt, per Google's comprehensive count.
  • 45–50 mLarxiv.org ·2025 · Earlier public estimates of water per prompt, as Google's paper cites them.
  1. Lessons›
  2. The school›
  3. How AI works
  • ·Before this: Why AI makes things up