Google Unveils Ironwood Chip, Boosting AI Inference Power

Google Cloud has introduced its latest Tensor Processing Unit (TPU) named Ironwood, claiming this advanced AI accelerator is over 24 times more powerful than the world’s fastest supercomputer when deployed at scale. This announcement was made during the Google Cloud Next ’25 event and signifies a major shift in Google’s ongoing AI chip development, which has spanned a decade. Unlike previous TPUs that were built to handle both training and inference tasks, Ironwood is uniquely designed for inference—the phase where trained AI models are utilized to make decisions or generate outputs. Amin Vahdat, Google’s Vice President and General Manager of ML, Systems, and Cloud AI, highlighted the chip's purpose during a virtual press conference preceding the event. He emphasized that Ironwood is tailored to meet the demands of the next generation of generative AI, which requires significant computational and communication resources. Vahdat described this era as the “age of inference,” where AI agents will actively gather and create data to provide collaborative insights and answers beyond merely presenting raw information. This strategic focus on inference is expected to enhance the ability of AI systems to function more efficiently and effectively, as businesses and developers increasingly rely on AI for more complex applications. With Ironwood, Google aims to set new benchmarks in the performance and capabilities of AI technology, reinforcing its commitment to innovation in cloud computing and artificial intelligence. This development is likely to have wide-reaching implications for sectors that depend heavily on advanced AI solutions, potentially transforming how they operate and make decisions.