According to @JeffDean, Google has released a technical paper detailing a methodology to measure the environmental impact of Gemini inference, providing a verifiable baseline for energy-per-inference modeling for traders (source: Jeff Dean on X, Aug 21, 2025). @JeffDean reports the median Gemini Apps text prompt consumes 0.24 watt-hours, establishing a concrete per-prompt energy intensity metric (source: Jeff Dean on X). Using this figure, 1 million prompts would consume about 240 kWh and 10 million prompts about 2.4 MWh, enabling direct scaling of workload energy and cost models (source: Jeff Dean on X). For crypto markets, this per-inference benchmark can be applied when assessing data center power needs and efficiency for compute-linked infrastructure adjacent to mining and AI-related tokens (source: Jeff Dean on X).