NVIDIA partners with d-Matrix for 10x faster AI inference

NVIDIA is teaming up with startup d-Matrix to create a hybrid computing setup that combines their hardware. By mixing NVIDIA's Hopper/Blackwell GPUs with d-Matrix's Corsair ASIC, the new system in Parasail cloud can generate tokens 10 times faster. Corsair uses a special D-IMC design that puts compute and memory on the same chip to cut data transfer delays. In this setup, NVIDIA GPUs handle the first part of AI reasoning, while Corsair takes care of the later, time-sensitive decoding. d-Matrix CEO Sid Sheth says GPUs and dedicated AI chips aren't rivals but partners, and this mix aims to make every stage of AI work better.