Unconventional AI is a company rethinking the foundations of computing to make artificial intelligence dramatically more energy efficient. Founded by specialists in AI systems, analog circuits, computing theory, and neuroscience, the company explores how hardware and software can be designed together to move beyond conventional approaches to scaling.

Its work spans research, experimental models, and new computing architectures. Unconventional AI investigates physical and dynamical systems, including coupled oscillators, as potential substrates for machine learning. Its Un-0 image-generation model demonstrates this approach and is available with open weights, training materials, and ablation code. The company also develops concepts and tools such as an instruction set architecture for dynamical-system hardware, alongside research into sparsity, memory technologies, analog and digital computing, and neural co-evolution.

Unconventional AI shares technical articles and research updates, and supports emerging work through the Unconventional Grant, which funds projects exploring new computational paradigms across hardware, systems, neural architectures, and theory. Its goal is to help enable AI systems that achieve greater capability with substantially less energy.