Summary: Google made Ironwood, its seventh-generation TPU, generally available at Cloud Next 2026 while previewing its eighth-generation architecture: TPU 8t (Sunfish), a Broadcom-designed training chip, and TPU 8i (Zebrafish), a MediaTek-designed inference chip, both targeting TSMC 2nm and late 2027. Ironwood delivers 4.6 petaFLOPS per chip and 42.5 exaFLOPS in a 9,216-chip superpod. The v8 split marks the first time Google has purpose-built separate training and inference chips, with Anthropic’s deal expanding to 3.5 gigawatts of compute in 2027, making it the anchor customer for both generations.
Google made its seventh-generation Tensor Processing Unit, Ironwood, generally available to cloud customers on Tuesday at Google Cloud Next in Las Vegas, positioning the chip as “ the first Google TPU for the age of inference ” and the centrepiece of what may be the most aggressive infrastructure investment in the company’s history.…