@atomicpoet (if you run your AI locally on your own hardware it requires more energy than if you run it in a data center though)

Vivien At home, you can turn off the hardware when it’s not in use whereas a data centre is always on.

@atomicpoet It’s also always in use.

Vivien Nonsense. Usage patterns vary widely. Most people aren’t running continuous AI workloads. Training massive models or serving thousands of inferences per second is the exception, not the rule.

For intermittent use—like occasional inference or lightweight models—running locally can actually be more efficient because you avoid the baseline power draw of data centres, which remain active even at low utilization.

@atomicpoet I’m skeptical there exists data centres with low utilization.

Vivien I thought your point is that running AI locally consumes more energy than data centres.

@atomicpoet That is correct. It does so by having newer generation hardware, more fit to the task than consumer hardware.

Vivien Sure, but again, usage patterns vary. I can always unplug when the hardware is not needed.

@atomicpoet I’m not sure there is an idle data centre anywhere. That would be a huge loss of money.

Vivien How is a machine in your house a “data centre”?

@atomicpoet It is not. It is different because while your machine in your house may be underutilized, a data centre is never (what I mean by idle), so there is no wasted power.