Extropic, Normal Computing Unveil AI Chips Claiming Up to 10,000x Lower Energy Use
Updated
Updated · Quanta Magazine · Jul 15
Extropic, Normal Computing Unveil AI Chips Claiming Up to 10,000x Lower Energy Use
1 articles · Updated · Quanta Magazine · Jul 15
Summary
Extropic and Normal Computing have introduced new thermodynamic computing systems aimed at running AI by exploiting thermal noise rather than suppressing it, with Extropic calling its chip the first scalable probabilistic computer.
Extropic said its semiconductor grid can run generative AI using about 10,000 times less energy than existing algorithms, while Normal’s new CN101 shifts from an earlier analog board to digital silicon for easier scaling.
Normal’s earlier prototype used eight interconnected resonator clusters to perform tasks such as matrix inversion, and researchers argue larger thermodynamic networks could eventually beat conventional digital neural networks on speed, heat and power.
The field remains early-stage: Extropic’s energy claims and Normal’s CN101 have not yet been peer-reviewed or independently assessed, and prior demonstrations often injected artificial noise rather than relying on ambient thermal fluctuations.
Researchers frame thermodynamic computing as a potentially nearer-term alternative to quantum computing, using standard semiconductor-style hardware to tackle AI and molecular simulation with far lower heat dissipation if scaling works.
Can machines that 'think' like folding proteins solve AI's massive energy problem?
If harnessing 'noise' makes AI computation nearly free, what unforeseen societal consequences will follow?
Thermodynamic Computing: Unlocking 10x–100x AI Performance per Watt for a Sustainable Future
Overview
Modern artificial intelligence (AI) is facing rapidly growing energy demands, pushing the limits of traditional computing and making energy the main constraint for global computing resources. As current AI algorithms become more energy-intensive, the cost of scaling them is skyrocketing, which has driven a shift from small performance improvements to transformational changes in chip design. This urgent need has led to the rise of thermodynamic computing, a new approach that uses the principles of thermodynamics to process information more efficiently. By rethinking how we design and power our digital world, thermodynamic computing offers a promising solution to maximize AI performance while minimizing energy use.