Chapter
The Energy Question
AI vs. streaming footprints; per-use vs. aggregate framing; debunked figures; water, grid strain, local prices.
This is where the most confident numbers are also the least reliable. Per-use, a text prompt is trivial — smaller than a Google search plus a page load. In aggregate, the buildout is a genuine grid event. Both are true, and conflating them is how the debate goes wrong in either direction.
Reading path
In order. 4 pieces, mixed media.
- 01 · briefAI vs. Netflix and YouTube: Energy FootprintsInternal research brief · 2026-08-23
- 02 · widget
Energy Comparator (dataset)
Open the dataset entryEnergy per use, kWh · log scale
AI video clip, 5 sec*0.9 kWhAI vs. Streaming brief — the most contested figure in the set; generated video is orders of magnitude above text.
Netflix / YouTube, 1 hr HD0.12 kWhActivity chart; IEA / Kamiya lineage.
Zoom call, 1 hr0.049 kWhActivity chart.
Very-long-input AI query (100k tokens)0.04 kWhEpoch AI, “How much energy does ChatGPT use?”, appendix. source
Short email0.0133 kWhActivity chart.
AI image, 1 image0.003 kWhActivity chart.
Long-input AI query (10k tokens)0.0025 kWhEpoch AI, “How much energy does ChatGPT use?”, appendix. source
Voice assistant query0.0005 kWhActivity chart.
1 hour of Netflix ≈ 100–300 typical AI text prompts. * = contested estimate. Hover any bar for its citation.
- 03 · widget
Midwest Site Map (dataset)
Open the dataset entrySchematic — not to scaleMount Pleasant, WI
· pop. ~27kOperationalMicrosoft Fairwater Wisconsin
Proposed · Hyperscale campus on the failed Foxconn parcel; $19.7M in taxes on track for 2026.
The fight · Largely defused: no NDA and 60+ public meetings. Backlash arrived only once the national story did.
Key people: Alyssa Capozziello
- 04 · essayNo Data Centers in My Backyardjasmi.news · 2026-08-03

