Episode 105

We need to talk about Physical AI, the most lethal oil- and fossil-fuel-killing engine ever built.

While headlines spread panic over data centers breaking the grid, digital reasoning is just the warm-up act. By the late 2030s, global data centers will draw around 1,200 TWh a year. But when intelligence moves into the physical world, powering 24/7 “lights-out” factories, driverless electric trucks, robotaxis, and 100 million humanoid robots, Physical AI will demand an astounding 6,200 TWh annually, pushing total AI electricity demand past 7,350 TWh.

Thinking is cheap; moving atoms isn’t. Because an autonomous machine requires precise sub-second control, adjusting torque 200 times a second compared to just three times for a combustion engine, Physical AI cannot run on burning stuff. Every clean electron feeding an automated machine directly liquidates diesel, swapping 20% to 30% efficient combustion engines for 90% efficient electric drives. The utility “dash to gas” to fuel this boom is a mathematical dead end. Heavy gas turbines are sold out with six-year delivery backlogs, while factory-manufactured solar and battery storage deploy from stock at the speed of compute. Running an automated physical economy on imported fossil fuels is merely automating a country’s own hostage crisis.

Solar is eating gas, batteries are eating the night, EVs are eating oil, and Physical AI is coming for whatever is left.