Google Unveils Ironwood TPU Built for the Age of AI Inference
Google Cloud introduces Ironwood, a seventh-generation TPU designed specifically for AI inference that delivers 24 times the computing power of the world's fastest supercomputer at scale. The chip marks a major strategic shift toward powering responsive AI agents.
Google Cloud reveals its seventh-generation Tensor Processing Unit named Ironwood, a custom AI accelerator built specifically for inference workloads. When deployed at scale across 9,216 chips in a single pod, the new hardware delivers an impressive 42.5 exaflops of computing power, which is over 24 times greater than the computational output of the world's fastest supercomputer.
This release signals a major strategic pivot for Google, as Ironwood is the first TPU designed exclusively for inference rather than model training. The company frames this shift as the transition into the "age of inference," where AI agents actively generate data and collaborate to provide answers rather than simply processing static information.
The chip brings massive technical upgrades, featuring 192GB of High Bandwidth Memory per chip and delivering twice the performance per watt compared to the previous Trillium generation. This exceptional power efficiency addresses the growing energy constraints within modern data centers while enabling significantly more AI capacity for customer workloads.