Who needs HBM when you can juggle SRAM speed and LPDDR bulk across racks AI chip startup d-Matrix is pushing into rack scale ...