EpistrySign in
Contested· ContestedContested

One California-grown almond requires water equivalent to the amount needed to process 38,000 average LLM queries.

The claim that one California almond requires water equivalent to ~38,000 average LLM queries is arithmetically plausible based on published figures: a single almond requires roughly 3.2–12 liters of water (commonly cited ~3.8 L/almond), while an average LLM query (e.g., GPT-class) consumes on the order of 0.0001–0.0005 L of water. Depending on which values are used, the ratio ranges from a few thousand to tens of thousands, so 38,000 falls within a plausible band but is highly sensitive to assumptions.

Largest vulnerability: Significant uncertainty — conflicting evidence is the main limit: Support ratio 42% (direction: mixed) — evidence is genuinely split.

Sources (25)

The competing case

The evidence that takes a side here is one-sided — by weight, most of the pool takes no side. Here is what the directional evidence shows.

Competing claims we weighed

  • A single California-grown almond requires far more than 3.8 liters of water — life cycle assessments indicate the full water footprint per almond kernel is substantially higher when irrigation, processing, and regional variability are accounted for, making the 38,000-query equivalence a significant underestimate.
  • LLM query water consumption is substantially higher than 0.0001–0.0005 L per query when on-site cooling water and indirect water embodied in energy generation are fully counted, meaning the 38,000-query figure dramatically overstates how water-intensive almonds are relative to AI.
  • Water consumption metrics for LLM queries are not yet standardized and the figures underlying the 38,000-query comparison are too uncertain and model-specific to support any reliable numeric equivalence with agricultural water use.
  • Life cycle energy and water assessments of almond production show that irrigation is the dominant input, but the per-nut water figure varies so substantially by orchard location, year, and cultivation practice that a single representative value cannot be assigned, undermining any fixed ratio to AI query water use.

Epistry weighs the case for and against — see the full breakdown, including the opposing evidence and why it held or failed, in the app.

Analyzed by Epistry · last updated · almond cultivation, water usage, large language models, LLM energy consumption, water footprint, California agriculture
One California-grown almond requires water equivalent to the amount n… — Epistry