Extropic Publishes Scalable Probabilistic Computer, Claiming 10,000x Lower AI Energy Use
Updated
Updated · Quanta Magazine · Jul 16
Extropic Publishes Scalable Probabilistic Computer, Claiming 10,000x Lower AI Energy Use
1 articles · Updated · Quanta Magazine · Jul 16
Summary
Early July 2026, Extropic published a paper on what it calls the world’s first scalable probabilistic computer — a chip built from thousands of interconnected semiconductor components.
The company says the system runs generative AI algorithms with about 10,000 times less energy than existing approaches by exploiting thermal fluctuations instead of suppressing noise.
That claim places Extropic in the emerging thermodynamic-computing field, where startups are trying to turn heat and randomness into useful computation to cut power use and heat dissipation.
The approach is still early-stage: other groups have mainly shown prototypes or simulations, and the broader promise of thermodynamic computing remains largely unproven outside initial demonstrations.
With 10,000x energy savings, is this new tech the AI industry's GPU killer?
Will harnessing chaos, not order, unlock the future of artificial intelligence?
Extropic’s Thermodynamic Computing: Achieving 10,000x Energy Efficiency for Generative AI with Probabilistic Hardware
Overview
As artificial intelligence rapidly expands, its growing energy demands—driven by the power-hungry GPUs needed to train large models—have become a critical challenge. Extropic, founded by a former Google X quantum computing researcher, is tackling this problem with a new kind of hardware called thermodynamic computers. By rethinking how AI computations are performed, Extropic aims to make them far more energy-efficient than current solutions. Their approach leverages probabilistic computing, using specialized chips that harness thermal noise, offering a promising path to sustainable AI as electricity consumption from AI is projected to soar in the coming years.