Extropic develops thermodynamic computing hardware and software designed to process probabilistic machine-learning workloads with substantially lower energy consumption than conventional approaches. Founded in 2022 and backed by $14.1m in seed funding, the company has built thermodynamic sampling units (TSUs) - hardware that directly samples from complex probability distributions by leveraging thermal noise, rather than relying on traditional matrix multiplication.
The company's product lineup includes the XTR-0 development platform, alongside chips designated X0 and Z1. Extropic has also released THRML, an open-source Python library enabling developers to simulate thermodynamic algorithms. This toolkit approach reflects a strategy to make the technology accessible to the broader machine-learning community whilst establishing standards around the new computing model.
The founding team brings experience from major technology companies including Google, IBM, Apple, and Microsoft, with expertise spanning machine learning and computer hardware design. The company's approach treats thermal noise and inherent randomness in electronic systems as design assets rather than liabilities - a fundamental departure from conventional semiconductor philosophy that seeks to minimise noise.