Google unveiled its latest tensor processing unit (TPU), Ironwood, at Hot Chips 2025, designed specifically for large-scale AI inference with a performance capacity of 42.5 Exaflops. Featuring up to 9,216 chips per node and significant improvements in power efficiency and reliability, Ironwood aims to excel in handling advanced AI models such as LLMs and mixture-of-expert systems. The architecture incorporates advanced memory subsystems and enhanced RAS features, making it a pivotal development for Google's AI computing capabilities.

Anthropic has partnered with Google to access up to one million Tensor Processing Units (TPUs) in a deal worth tens of billions of dollars, significantly expanding its AI compute capacity. The company, which has seen rapid revenue growth, leverages a multi-cloud architecture that includes partnerships with both Google and Amazon to optimize performance and cost, while maintaining control over its model and data.