On March 19, 2025, NVIDIA introduced Dynamo, heralded as the "operating system for AI factories." At its heart lies the "Distributed Inference Service Library," engineered to tackle token supply and demand challenges within the open-source realm, thereby bolstering the efficiency and adaptability of AI factories. During the unveiling event, NVIDIA CEO Jensen Huang highlighted Dynamo's capabilities with illustrative charts, underscoring its groundbreaking performance when paired with the Grace Blackwell NVLink72 superchip. Dynamo boasts a 25-fold increase in performance over the previous-generation Hopper chip and a staggering 40-fold improvement in inference model performance. Huang emphasized that this leap will empower AI factory vendors to achieve greater economies of scale, further accelerating the growth of the AI industry.
