Rumor in, receipts out

“AI data centers are eating the world’s electricity.”

The International Energy Agency says data centers used about 415 TWh in 2024, around 1.5% of the world’s electricity.¹ Its projection is about 945 TWh by 2030, slightly more than Japan’s electricity use today.¹

In the U.S. the share is larger. LBNL estimates data centers used 4.4% of U.S. electricity in 2023 and could use 6.7% to 12.0% by 2028.^2,3 Note that these figures cover all data centers, not only those running AI.

Checked Oct 1, 2026. Sources: IEA, LBNL, U.S. Department of Energy. Projections are scenarios, not measurements.

Read the full answer →

“Each chatbot question uses a bottle of water.”

Google says a median Gemini text prompt uses about 0.26 mL of water (“about five drops”), 0.24 Wh of energy and emits 0.03 g of CO₂e.⁴ A 500 mL bottle (our assumption of a typical size) would hold about 1,900 of those prompts. That division is our own arithmetic.

This is a company estimate for one product, from May 2025 data. Google says the data and claims have not been verified by an independent third party. Other bots, and heavier tasks like video or long documents, will differ.⁴

Checked Oct 1, 2026. Source: Google Cloud Blog.

Read the full answer →

“Data-center water use is just cooling.”

LBNL separates two kinds of water. Direct water, used at the data center itself and mostly for cooling, was an estimated 66 billion liters in 2023, up from 21.2 billion in 2014. Indirect water, used by power plants to make the electricity, was nearly 800 billion liters in 2023.²

The indirect number is about 12 times the direct one (our arithmetic, 800 ÷ 66). That does not mean local cooling water doesn’t matter: a town’s water supply cares about the direct number where the building is.

Checked Oct 1, 2026. Source: LBNL, Dec 2024. Estimates, not meter readings.

Read the full answer →

“Data-center energy use has always been exploding.”

LBNL says U.S. data-center energy use stayed fairly stable at about 60 TWh from 2014 to 2016. It reached about 76 TWh (1.9% of U.S. electricity) in 2018 and 176 TWh (4.4%) in 2023.² DOE describes that as roughly a tripling since 2014.³

LBNL ties much of the rise to GPU-accelerated servers for AI, which grew from under 2 TWh in 2017 to more than 40 TWh in 2023.² Google separately says the energy of a median Gemini prompt fell 33-fold in 12 months, a company claim.⁴

Checked Oct 1, 2026. Sources: LBNL, DOE, Google.

“Nobody measures how much energy and water data centers use.”

Totals exist. The figures cited in POLITICO’s Sept 30, 2026 report put EU data-center electricity use in 2025 at 20.7 TWh and water use at more than 8 million cubic meters, but POLITICO notes they come from incomplete data.⁵

The advocacy group Lighthouse Reports has filed a complaint under the Aarhus Convention, saying EU rules keep the public from seeing figures for individual data centers. It is a complaint, not a finding that anything was breached; an admissibility review is expected in November.⁵

Checked Oct 1, 2026. Source: POLITICO. Unresolved.

“Today’s AI is already conscious.”

A 2023 report by 19 researchers assessed several AI systems against indicators drawn from scientific theories of consciousness. Its analysis suggests that no current AI systems are conscious, and also that there are no obvious technical barriers to building systems that meet the indicators.⁶

That is one report from 2023, not a settled answer for today’s newest models. The accurate summary is: it is an open research question, and today’s chatbots are not shown to be conscious.

Checked Oct 1, 2026. Source: arXiv 2308.08708 (abstract).

Read the full answer →

“AI is always right if it sounds sure.”

NIST calls it “confabulation”: AI presenting false content with confidence.⁷ In 2023 a federal judge fined two New York lawyers and their firm $5,000 after they filed court cases that ChatGPT had made up.⁸

What to do: ask for the source, then open it yourself. See How AI works and Prompting: check the answer.

Checked: NIST and court opinion read Sept 24, 2026 (both cited on our other guides).

Read the full answer →

“Super intelligence is here.”

IBM describes artificial super intelligence as a hypothetical software-based AI system with an intellectual scope beyond human intelligence, and calls today’s systems narrow AI.⁹ AWS calls it a theoretical concept.¹⁰

The word is in the news because the White House now uses it: on Sept 29, 2026 President Trump signed an executive order directing the federal government to use “Super Intelligence” instead of “artificial intelligence,” CBS News reports.¹¹ A change in what the government calls AI is a naming choice, not a new technology. We keep using “AI.” Glossary entry.

Checked Oct 1, 2026. Sources: IBM, AWS, CBS News.