An Investor Says the Brain Is a Million Times More Efficient Than Data Centers
An investor points to brain-like chips as one path to much lower AI power use.

Guru Chahal, an investor with Lightspeed Venture Partners, puts the human brain at 5 to 20 watts, about as much power as a dim lightbulb.
That figure is his estimate, not a measured result. Chahal says of the brain, "It's a million times more efficient than the compute we have in data centers today."
He points to a startup he works with, Unconventional, that is building AI chips to work more like the brain than like a traditional computer.
Chahal described the aim in July as chips that are 100 to 1,000 times more power efficient, a target for the effort. He frames the payoff as a what-if: what if the same AI answer could be made with 10%, 1%, or 0.1% of the power used today.
Questions people ask
Why do AI data centers use so much power?
An AI request makes many chips work together to generate an answer, rather than simply fetching a stored file. A traditional server rack might draw about 10 kilowatts, while an AI rack can draw about 100 kilowatts. Cooling all that heat takes still more energy.
What would more efficient chips mean for regular people?
They could lower the cost of running AI services and slow the growth in electricity and cooling needed for new data centers. In devices, greater efficiency can also mean longer battery life and less heat. None of those benefits arrive until new designs are proven and mass produced.


