The infrastructure behind modern AI was never designed for today's scale, energy, or complexity. General-purpose architectures do not exploit the physics of conventional hardware, and the cost of intelligence is unsustainable.

Normal closes that gap in silicon and in software. We build silicon that computes with stochastic physics: a thermodynamic hardware program for the next era of generative AI. And our AI designs silicon through a continually learning EDA stack.

The long-term aim is a design process that compounds: AI that learns from every workload to shape the next generation of silicon, and silicon whose physics feeds back to improve the models. The same loop retargets the hardware, reducing the cost of intelligence with every cycle.

We deploy our hardware and software in partnership with the world's most advanced institutions. We have offices in New York City, San Francisco, London, Copenhagen, and Pangyo.