MatX

MatX, Inc. is a technology company founded in 2022 by Reiner Pope and Mike Gunter, with its headquarters located in Mountain View, California. The company specializes in designing and developing high-throughput semiconductor hardware specifically optimized for large language models (LLMs). Founded by former Google engineers who worked on the Tensor Processing Unit (TPU) chips, MatX aims to address the computational bottlenecks of modern artificial intelligence by creating processors that maximize performance for large-scale applications. The company's core focus is on delivering increased computing power and efficiency for AI labs and frontier research, with a mission to serve as a foundational compute platform for artificial general intelligence.

Business/Product Breakdown

MatX, Inc. develops and manufactures the MatX One, a next-generation AI processor engineered from first principles to handle the specific requirements of large-scale AI workloads, particularly those with over 7 billion parameters. The MatX One architecture utilizes a splittable systolic array designed for high utilization on varying matrix shapes, combining the low-latency characteristics of static random-access memory (SRAM) with the high-capacity support of high bandwidth memory (HBM). This hardware targets specific workloads including model training, reinforcement learning, inference prefill, and inference decode, with a focus on maximizing floating-point operations per second (FLOPS) and minimizing latency for Mixture of Experts (MoE) and large dense models. Unlike general-purpose graphics processing units, MatX's technology features a specialized interconnect architecture to coordinate communication across large clusters of chips and a programming model that provides direct hardware control.