Author(s): Aqeel Abbas

Originally published on Towards AI.

Google says the energy behind a median Gemini prompt fell 33-fold in twelve months. Texas is pausing new grid connections anyway. Both numbers are real, and the reason they coexist is the actual story.

Two claims about AI’s energy footprint have been circulating simultaneously in 2026, and they sound like they can’t both be true. Google reports that the energy behind a median Gemini prompt fell 33-fold over twelve months. Epoch AI independently pegs a typical ChatGPT query at roughly 0.3 watt-hours — about ten times below older, widely circulated estimates. By that measure, AI got dramatically more efficient, fast.

The article explains that the apparent contradiction is resolved by separating per-query efficiency from total demand: efficiency gains are real thanks to better architectures, infrastructure, batching, and more compute per watt, but overall electricity use still rises because cheaper per-unit compute drives higher usage volume. It argues that AI load growth outpaces efficiency improvements, accelerators-heavy data centers grow faster than conventional infrastructure, and the strain is highly concentrated geographically where grid interconnection queues and power availability become bottlenecks. The piece details how this constraint is pushing a “nuclear bet” with multiple large corporate commitments to nuclear (including restarts and small modular reactor projects), while consumers already feel impacts through rising wholesale prices and regulated cost allocation debates. It concludes with implications for builders and deployers: track both unit and aggregate metrics, factor regional grid capacity and energy mix into architecture, understand how routing to smaller models interacts with hardware efficiency, and treat power availability—not chip supply—as a scaling gate.

Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI