d-Matrix ☆

d-Matrix argues the inference market splits away from training hardware entirely: its Corsair cards compute inside SRAM to cut latency and power for interactive serving. Backed by Microsoft's M12 and Temasek.