The computer has to run somewhere
We enjoy building with computers and keeping up with AI. Spending time on the water also makes it hard to forget that a screen is connected to a physical place. Servers need power, cooling, equipment, and networks.
The IEA and Lawrence Berkeley National Laboratory reports below give useful background on data-center energy demand. Read their dates, assumptions, and ranges. An industry estimate doesn’t tell you the footprint of one tool or one site.
Ask what the job actually needs
A small script, a Raspberry Pi, a local PC, and a cloud AI service are different tools. Before choosing one, ask what the job is and how often it needs to run.
Can a simple lookup do it? Can a result be reused instead of calculated again? How long do we need to keep the data? Those are practical questions we can answer while building, without pretending we’ve measured the entire industry.
Keep the experiment honest
If we share a result, we want to explain the setup, what we measured, and what remains a guess. A useful tool still needs maintenance and a way to recover when it fails.
WitnessWater, RiverLink, and FLY AI are in development. For now, GFM Tech is a place to talk about these ideas, learn, and tinker between river days. We’ll share working examples as they’re ready.
Field notes
Join the conversation.
Have a question, a field observation, or a correction? Send us a note.