New Heat-Powered AI Runs on Less Energy Than Before
Scientists found a way to make AI computers use energy almost like our brains do — faster, cheaper, and cooler.
A study of a thermodynamic computer—a stochastic physical device that performs calculations at the thermal energy scale—found it can match a simple multilayer perceptron on a standard machine-learning classification task. Using the Wasserstein speed limit to assess efficiency, researchers showed that different inference protocols allow the computer to run within 40% of the thermodynamic limit of efficiency without losing accuracy, or to perform inference increasingly rapidly at fixed accuracy and efficiency. The findings indicate that a thermodynamic computer designed for a particular task retains considerable freedom in how it operates.
- Researchers built a 'thermodynamic computer' that runs AI tasks using heat instead of electricity, like a brain uses calories.
- The AI performs as well as standard AI but can save up to 40% of the energy that would normally be wasted.
- This could lead to cheaper, cooler, and more eco-friendly AI devices — from phones to data centers.
Why It Matters
AI could soon run on waste heat, making it cheaper, cooler, and greener for everyone — from your phone to the cloud.