Musk Says 10 Watts of Thinking Built Civilization, and AI Is Far Less Efficient
The brain uses about 20 watts in total, Musk says, pointing to room for far more efficient AI.
All-In with Chamath, Jason, Sacks & Friedberg · Elon Musk: OpenAI Betrayal, His Future at Tesla, and the Next Big Thing — GrokipediaYour brain runs on about 20 watts, roughly the power of a dim light bulb. Only about 10 watts of that goes to higher thinking, said Elon Musk, the executive behind Tesla and xAI, in a conversation about AI and power use. The rest is housekeeping, keeping the heart going, breathing, and other basic functions.
Musk described the brain as having a 20 year boot sequence. He called it very power efficient.
With about 10 watts for higher thinking, humans built civilization. Musk listed aircraft, lasers, and the internet among the inventions, along with the discovery of general relativity and quantum mechanics.
By comparison, Musk said current AI compute is far less efficient than the brain. In the conversation, the interviewer pointed to recent papers said to cut the power needed per unit of AI output by three or four orders of magnitude, and asked Musk where efficiency was headed. Musk answered by pointing to the brain as the example of efficient computing.
He said even a 100 megawatt or gigawatt AI supercomputer cannot yet do everything a human can do, though he expects it eventually will. He added that people communicate with each other at only about 10 bits per second, not terabytes.
Given a brain that builds civilization on about 10 watts and about 10 bits per second of communication, Musk said there is a massive chance to make AI use far less power.
Questions people ask
How can the brain do so much on 20 watts?
Brain cells communicate with brief electrical pulses and chemical signals, and only small networks fire at once. The brain also computes in memory rather than shuttling data back and forth, and it uses efficient, approximate processing instead of exact math. Together, those tricks use far less energy than a digital chip doing similar work.
How much power does AI actually use?
A single large AI data center can draw 100 megawatts or more, about as much as a small city, and the largest planned campuses aim for one gigawatt or beyond. Most of that goes to thousands of processors plus cooling and power conversion. A single AI query uses only a tiny fraction, but billions of queries add up quickly.
Will AI chips ever work like brains?
Chip designers are already borrowing ideas like processing in memory, sparse activation, and analog computing. Those approaches can cut energy use substantially, and researchers are reporting large drops in energy per unit of AI output. Matching the brain's full efficiency, however, remains a distant goal.