Uncon­ven­tional AI

If the data centers needed to train and run tomorrow’s AI models multiply as fast as expected, they’ll soon exceed the energy grid’s capacity to power them. One way out of this dilemma is to design a new computing substrate that uses a fraction of the energy of today’s hardware. Unconventional’s goal is to build novel software and hardware that are better suited to reproduce the stochastic operations and non-linear dynamics of biological neural networks, ultimately emulating the energy efficiency of the human brain.