Standard digital chips spend extra energy forcing noisy sub-threshold transistors to behave as clean, deterministic 0s and 1s. Thermodynamic (or probabilistic) computing runs transistors near their natural noise floor instead and uses that randomness as the computation itself, sampling from a programmable probability distribution such as an Ising model. Extropic's Z1 chip is the leading example: 269,568 probabilistic bits (pbits) with fixed sparse connectivity sample at 50 MHz under 1 watt, a design suited to sampling-heavy and probabilistic workloads rather than the dense deterministic matrix multiplication GPUs are built for.