There are many applications that stress the limits of compute resources, from smartphones and self-driving cars to high-performance computing (HPC). In recent years artificial intelligence (AI) has threatened to eclipse or subsume all other application domains as the dominant consumer of computational power. Although AI algorithms can run on just about any processor, they run most efficiently on specialized engines that move some of the key algorithms from software to hardware. Designing an AI chip is a major undertaking requiring specialized knowledge, tools, and methodology at every stage of the process. This white paper discusses the first key stage: pre-silicon planning. It begins with an overview of the AI market to explain why it dominates the chip industry and to establish the motivation for the investment needed to design customized domain-specific chips.