A leading group of researchers from Extropic and the Massachusetts Institute of Technology (MIT) has introduced the concept of a thermodynamic computer—a fundamentally new computing architecture capable of radically reshaping the energy balance in the field of artificial intelligence. According to my analysis, this approach could become one of the most significant breakthroughs in AI hardware over the past decades.
The key idea is that instead of traditionally combating physical noise and thermal fluctuations—which are considered interference in modern processors (including GPUs)—thermodynamic computing proposes using these random processes as an integral part of the computational mechanism. According to the authors' estimates, this approach could improve the energy efficiency of performing certain AI tasks by up to 10,000 times compared to classical computing.
Why This Matters for the AI Industry
Modern large language models and deep learning systems require enormous computational resources. The largest tech giants are already investing billions of dollars in building data centers, while the demand for electricity continues to grow exponentially. The thermodynamic approach offers an elegant solution: since many AI tasks—such as finding the most likely answer or optimal solution—are inherently probabilistic, using random physical processes may not only be acceptable but also more efficient than deterministic computing.
From Theory to Practice: A Long Road Ahead
It is important to emphasize that, at this stage, this involves fundamental research and simulation results, not a ready-made commercial product. The authors have demonstrated the advantages of the new approach for specific classes of tasks, but it may take years before real thermodynamic chips emerge. Nevertheless, this work reflects the industry's growing interest in alternative computing architectures—alongside quantum and neuromorphic computers.
Against the backdrop of the rapid growth in the scale of AI models, finding ways to drastically reduce energy consumption is becoming not just a scientific challenge but an economic necessity. The cost of computing is already one of the main limiting factors for the development of artificial intelligence.
My expert opinion: Thermodynamic computing is exactly the type of innovation that could redefine the rules of the game in the industry. If the technology is successfully commercialized, we may witness not only a reduction in energy costs but also a lower barrier to entry for developing powerful AI systems, which will inevitably accelerate the pace of innovation across the entire sector.