When AI Leaves the Data Centre, Electricity Demand Explodes
Key Takeaways: How Physical AI Could Reshape Global Electricity Demand
Physical AI refers to AI systems that act in the physical world through machines such as robots, autonomous vehicles and automated industrial equipment, potentially creating electricity demand far beyond that of data centers alone.
Physical AI could consume far more electricity than AI data centers alone. Ember’s illustrative bullish scenario estimates physical AI could add roughly 6,160 TWh of annual electricity demand by the end of the 2030s.
Data centers and physical AI could together require about 7,350 TWh of electricity per year. That would be equivalent to roughly 25% of estimated global electricity consumption in 2026, although Ember describes its physical AI scenario as indicative rather than a forecast.
Automated factories and autonomous transportation could be the largest sources of physical AI electricity demand. Ember estimates lights-out factories could require 1,700 TWh annually, followed by driverless electric trucking at 1,150 TWh and robotaxis and autonomous cars at 1,000 TWh.
Much of physical AI’s electricity demand could replace fossil-fuel energy rather than simply add new energy consumption. Electric trucks, mining equipment, farm machinery and autonomous vehicles could displace diesel and gasoline use as transportation and industry electrify.
Physical AI could turn the AI infrastructure challenge into a much broader electrification challenge. Supporting widespread autonomous machines would require not only more generation but also expanded transmission, substations, transformers, charging infrastructure, batteries and other grid equipment.
Many commentators are worried about AI data centres placing a heavy demand on the electric grid. According to Ember, they are missing the bigger story.
The explosive growth of GenAI has focused attention on the electricity required to train and run large language models (LLMs). Utilities are scrambling to accommodate multi-hundred-megawatt and even gigawatt-scale data centres. Technology companies are signing enormous power contracts, building gas generation and investing heavily in renewables.
But Ember’s report “The Age of Power” released September 16, 2026 points to a much larger future electricity demand emerging: physical AI.
By the end of the 2030s, Ember’s scenario puts electricity consumption from data centres at roughly 1,193 TWh per year, with a faster-growth case reaching 1,719 TWh. That is extraordinary.
But physical AI could add roughly 6,160 TWh of annual electricity demand on top of that, bringing total AI-related electricity demand in Ember’s illustrative scenario to approximately 7,350 TWh per year. To put that in perspective the world’s nations will consume an estimated 29,600 TWh in 2026. So this illustrative new load is equal to 25% of the world’s current consumption.
The electricity required when AI begins acting in the physical world will eventually dwarf the electricity required for AI reasoning, according to Ember.
The largest potential source is “lights-out” automated factories. Ember models approximately 1,700 TWh a year if highly automated factories become widespread. Industrial robots, electric motors, machine vision, autonomous materials handling and 24-hour production all turn AI into physical activity—and physical activity requires energy.
Next comes driverless electric trucking, potentially consuming another 1,150 TWh annually. Autonomous electric trucks could operate longer hours than human-driven vehicles, dramatically increasing utilization while replacing diesel. (which is a good thing for the climate).
Robotaxis and autonomous cars could require another 1,000 TWh. Currently robotaxis are driving 350 million to 425 million miles per year according to annualized estimates from various sources in 2026, with more than 90% of that in China and the US.
A world with 100 million humanoid robots operating for 20 hours a day could consume roughly 950 TWh annually.
Then comes autonomous:
construction and mining equipment, 350 TWh;
agricultural machinery, 300 TWh;
robotic delivery vans and pavement robots, 300 TWh;
warehouse, port and logistics robots, 200 TWh;
electric buses and shuttles, 160 TWh;
drones and autonomous aviation, another 50 TWh.
This fundamentally changes how we should think about AI and energy.
Thinking is relatively cheap. Moving atoms is expensive.
A chatbot answering a question uses electricity somewhere inside a data centre. But when the answer causes a two-ton truck to move 1,000 kilometres, a robot to weld automobile frames for 20 hours, a mining vehicle to haul ore, or a humanoid robot to perform physical work, AI becomes connected to vastly larger energy flows.
There is an important qualification: Ember labels this a bullish scenario—not a forecast. Many of these assumptions depend on extraordinarily rapid adoption of robotics, autonomous transportation and industrial automation.
And much of the electricity demand would not represent entirely new energy consumption. It would replace fossil fuels. Electric trucks replace diesel. Electric mining equipment replaces diesel machinery. Autonomous electric farm equipment replaces tractors burning fuel. Robotaxis replace gasoline-powered vehicles.
So the story is not simply that AI creates massive new energy demand. AI could accelerate one of the largest electrification waves in history. And that has profound consequences for infrastructure.
If anything approaching Ember’s scenario occurs, the constraint will not just be generating electricity. We will need dramatically more transmission lines, substations, transformers, distribution infrastructure, charging networks, batteries, electric motors and power electronics.
And we will need generation technologies that can be deployed at enormous scale.
Solar and wind become particularly important because they can be manufactured and installed rapidly and modularly. Batteries become essential for balancing increasingly variable generation and demand. Hydro and other generation will also have roles, while grids will need massive expansion and modernization.
We tend to think of the AI revolution and the energy transition as two separate megatrends. They will increasingly be intertwined megatrends.
AI begins in data centres as intelligence. Its much larger impact will come when that intelligence escapes the data centre and begins controlling factories, vehicles, robots, farms, mines, warehouses and machines throughout the physical economy.
The AI boom will ultimately become an electrification boom.
Sources:
Ember: The Age of Power, Executive summary, slide deck and full report
https://ember-energy.org/latest-insights/the-age-of-power/
About Jim Harris:
Jim Harris is a #1 International Bestselling author published in 80 countries. His new book The Age of Fakes!: How AI Abuse, Fake News, and Deepfakes Threaten Business and Society is a #1 International bestseller in multiple categories and countries. You can see it at https://www.amazon.com/Age-Fakes-Deepfakes-Threaten-Business/dp/3982801001/ref=sr_1_1
Follow Jim on Linkedin.com; X (Formerly Twitter), YouTube and Substack
Interested in having Jim as an AI keynote speaker at an upcoming event? Email him at [email protected]