New 'Thermodynamic Computer' Mimics AI Neural Networks Using Far Less Energy

Scientists have built a device that can generate images from random data, using orders of magnitude less power than traditional AI systems.

Published on Feb. 21, 2026

Got story updates? Submit your updates here. ›

Researchers have developed a 'thermodynamic computer' that can produce images from random disturbances in data, or noise, mimicking the generative capabilities of artificial intelligence (AI) neural networks. Unlike conventional computing systems that require large amounts of energy to operate, this new device leverages the natural thermal noise in the system to complete computing tasks with significantly less power consumption.

Why it matters

This breakthrough in thermodynamic computing could lead to more energy-efficient AI systems that are able to perform tasks like image generation. By harnessing the inherent randomness in physical systems, rather than trying to eliminate it, the researchers have found a way to dramatically reduce the energy demands of certain types of computations. This could have important implications for the future development of AI and other advanced computing applications.

The details

The thermodynamic computer works by programming the 'coupling strengths' between different circuits in the system. This allows the natural fluctuations in the circuit values, driven by thermal noise, to be 'programmed' to solve specific computational problems, like generating images. Unlike traditional computing that relies on definite binary values (1s and 0s), this approach uses probabilities of values, which can be more efficient for certain types of optimization problems.

  • The new study outlining the thermodynamic computer was published on January 20, 2026 in the journal Physical Review Letters.

The players

Stephen Whitelam

A staff scientist at the Molecular Foundry at the Lawrence Berkeley National Laboratory and the author of the new study on the thermodynamic computer.

Normal Computing Corporation

A company in New York that has built something similar to a thermodynamic computer, using a network of low-energy circuits.

Ramy Shelbaya

The CEO of Quantum Dice, a company that produces quantum random number generators, who commented on the significance of the thermodynamic computing research.

Got photos? Submit your photos here. ›

What they’re saying

“Wouldn't that be much more natural in a thermodynamic setting where you get the noise for free?”

— Stephen Whitelam, Staff scientist (Conference proceeding)

“This article also shows how physics-inspired approaches can provide a clear fundamental interpretation to a field where 'black-box' models have dominated, providing essential insights into the learning process.”

— Ramy Shelbaya, CEO (Live Science)

What’s next

Researchers plan to continue exploring how thermodynamic computing principles can be scaled up to handle more complex AI and machine learning tasks, beyond the initial image generation demonstrations.

The takeaway

This new thermodynamic computing approach represents a significant breakthrough in developing more energy-efficient AI systems. By harnessing the natural randomness inherent in physical systems, rather than trying to eliminate it, the researchers have found a way to dramatically reduce the power demands of certain computations, paving the way for more sustainable and scalable AI applications in the future.