Energy-efficient AI accelerator chips (XPUs) that minimize data movement to reduce inference and training energy costs.
Zettascale develops XPU chips, a family of AI accelerators designed to keep data close to compute, minimizing the energy cost of data movement that dominates datacenter power budgets. The first XPU prototype, Grasshopper, is built on FPGA and targets dense math, sparse models, and practical numeric precisions. Monolith clusters multiple XPUs to behave as a single larger chip, supporting agents, experience generation, and training on the same silicon. Zettascale XPU is a product of Zettascale.