Consumption

Each chapter records, in its footer, the electricity and water it required to come into existence. The work concerns a technology that consumes mineral and water resources. Environmental costs are therefore not incidental data. They are part of the text. This page opens the project’s accounts and cites the sources we use to calculate the consumption associated with our processing.

What goes into the calculation

Generation. The tokens used by the model to write each chapter are measured by adding together all writing attempts. This calculation provides what we call estimated generation water , meaning the cooling of the datacenter that ran the model.

Video. The energy, in watt-hours, consumed to generate the video images for each chapter.

Printing. The energy consumed by the dot-matrix printer while printing the chapter, measured at the power outlet.

Machine. The energy consumed by the computer to generate a chapter.

Paper. Estimated water consumption for producing one page of continuous-feed paper.

Water per token: the figure and its basis

Water consumption for generation is estimated at 2 mL per thousand tokens. No published figure exists for the model used in this work, so the value is based on a 2025 inference benchmark, which reports around 0.35 mL per thousand tokens for an OpenAI model (GPT-4o) and around 8.7 mL per thousand tokens for a reasoning model. The model used in 2047 runs with low reasoning effort, so 2 mL per thousand tokens is a conservative estimate between the two. This includes on-site cooling and water used in electricity generation. This figure is an estimate, not a measurement, and remains open to revision.

Sources