Extropic Unveils First Working Thermodynamic Computing Chip for AI

Extropic, a startup pioneering a novel computing paradigm, announced the public debut of its first working thermodynamic computing hardware, marking a significant step in developing energy-efficient AI. The company shared the milestone with a tweet stating, "> Hello Thermo World," on October 29, 2025, signaling the readiness of its Thermodynamic Sampling Units (TSUs). This development aims to address the escalating energy demands of artificial intelligence workloads.

Extropic's thermodynamic computing chips are designed to harness natural thermodynamic fluctuations within electronic components, a stark contrast to traditional computing that seeks to minimize such noise. These TSUs are inherently probabilistic, making them a suitable fit for probabilistic AI workloads and offering a radically more energy-efficient alternative to Graphics Processing Units (GPUs). The company emphasizes a "probabilistic software, meet probabilistic hardware" approach.

The first operational Extropic chip has already been shared with a select group of partners, including frontier AI laboratories, startups focused on weather modeling, and government representatives. Extropic also introduced THRML, an open-source Python library, to enable developers to create thermodynamic algorithms and simulate their execution on TSUs. The company plans to make its first operational chips available for market later in 2025.

Founded by Guillaume Verdon and Trevor McCourt, Extropic's mission is to redefine computation with a "physics-first view," merging generative AI with the physical world. Verdon, known for his work in quantum computing, and McCourt lead a team aiming to create an ultimate substrate for generative AI by leveraging out-of-equilibrium thermodynamics. This approach seeks to achieve unparalleled AI performance per watt.

Thermodynamic computing positions itself as a potential disruptor in the AI hardware landscape, offering substantial speedups and energy savings compared to conventional silicon chips. While quantum computing minimizes random fluctuations, Extropic embraces them, aiming to provide a scalable and mass-manufacturable probabilistic computing platform. The company's innovation challenges established players like Nvidia by proposing a fundamentally different way to process AI tasks.