Let's not gloss over the electrical supply. These chips won't work for free.

LLM inference uses on the order of 1 Wh per query. That's under 10 meters of driving on an EV or running air conditioning for under 5 seconds.

	▲	bluefirebrand 3 hours ago \| parent \| next [-]
		One query is not going to be a useful benchmark when people are deploying AI swarms in loops to solve simple problems
	▲	deadbabe 4 hours ago \| parent \| prev [-]
		Or a human riding a stationary bike for 36 seconds.