Google has launched the Ironwood artificial intelligence chip, designed for training large language models, providing an alternative to Nvidia.

Google has announced the large-scale launch of its advanced artificial intelligence chip, the seventh-generation Tensor Processing Unit (TPU), named Ironwood. This internally designed chip will be publicly available in the coming weeks, after a testing period. Ironwood is optimized for training large language models and real-time operation of chatbots, boasting a speed more than four times greater than the previous generation. The chips can be connected in large groups, facilitating the running of the most complex AI models. Google is competing with Microsoft, Amazon, and Meta in building global AI infrastructure, and Ironwood offers a customized alternative to Nvidia graphics processors, with cost and energy efficiency advantages. Among the first customers is the startup Anthropic, which will use up to a million chips for the development of its AI model. The chip launch coincides with updates to Google’s cloud infrastructure, which has reported significant revenue from cloud services. CEO Sundar Pichai highlighted the growing demand for AI infrastructure, announcing an increase in investment estimates for 2025.