MatX (COM US)

High-throughput accelerator chips and hardware architectures optimized for large language model training, inference, and long-context workloads